Cautious or marginalized? Perceptions of the riskiness of engaging in corruption and gender differences in corruptness

Previous studies suggest that women are often perceived as less corrupt and more risk averse, possibly due to long-standing asymmetries in power and marginalization that reproduce certain gender stereotypes. However, much remains unclear regarding the origins of these perceptions. Why are some individuals and societies more prone to perceive women as less corrupt than men? We present the first cross-country examination of these questions utilizing data from the latest wave of the World Values Survey, covering a total of 49 countries. Our results suggest that both perceived riskiness of corruption and attitude toward gender equality, in addition to the overall level of gender inequality in society, matter in explaining a stronger belief in gender differences in corruptness. However, the positive association between a higher perceived riskiness and the perception that women are less corrupt is mostly limited to societies with high levels of corruption and gender inequality.

A prominent line of research in recent years has been the association between gender and corruption (see, e.g., Bauhr & Charron, 2020;Malmberg & Christensen, 2021;Merkle, 2020;Stensöta & Wängnerud, 2018).While a large share of these studies has been interested in exploring the link between gender (in)equality and overall levels of corruption, and the question if women actually have a lesser tendency to engage in corruption or tolerate corruption, some, such as Barnes and others (2018) and Goetz (2007), have concentrated on explaining public perceptions of women as "political cleaners."The distinction between perceptions of being less corrupt and actually being less corrupt is important since perceptions-regardless of whether they are true or false-have also been shown to have real-life consequences.Perceptions of corruption being widespread have, for instance, been linked to a greater likelihood of bribegiving (see e.g., Nguyen & Le, 2022).However, most of the previous studies on the subject have mainly relied on evidence from single country survey experiments.We build on this work further by examining these questions with cross-national data of this question using data from the latest wave (Wave 7) of the World Values Survey.This allows us to contribute to the debate on whether the country level context affects these perceptions (see, e.g., Esarey & Schwindt-Bayer, 2018).
Many studies suggest that there exists a widespread belief that women in government are likely to reduce corruption and that women in general are less corrupt than men (see e.g., Barnes et al., 2018).One example of this belief being translated into public policy comes from Peru in the early 2000s, where the National Police in an attempt to reduce malfeasance among police officers enlisted a greater number of female officers (Goetz, 2007).Likewise, there is evidence from experimental research on representative bureaucracy suggesting that a larger share of women in the police force increases its perceived trustworthiness in the eyes of ordinary citizens (Riccucci et al., 2014).But what are the roots of this belief?Is it in any way grounded in (perceived) empirical reality (i.e., influenced by direct or indirect observations of the social structures of corruption that make it a riskier venture for women), or is it mainly a general reflection of women being less powerful and more marginalized in some societies?We explore the questions surrounding corruption-related gender norms, which have received relatively limited attention in previous empirical work due to a lack of cross-national survey data, by focusing on the association between perceptions of the riskiness of taking part in corrupt exchanges and the belief that women tend to be less corrupt than men.The research question guiding this study is therefore: Why are some individuals and societies more prone to perceive women as less corrupt than men?Is there an association between perceptions of corruption and the belief that women are less corrupt, or is this belief a manifestation of other factors such as gender norms and conservative views on the role of women in society?These questions are important as they have implications for both women's participation in public offices (see e.g., Barnes & Beaulieu, 2014;Benstead & Lust, 2018) and policy suggestions for combating corruption and restoring trust in public officials (see e.g., Barnes et al., 2018;Goetz, 2007).To bring a greater clarity to these issues, we use multilevel statistical analysis of data from the latest wave (Wave 7) of the World Values Survey conducted in the period 2017-2020.This approach enables us to take into account contextual variations in the association between individual corruption perceptions and corruption-related gender norms, while at the same time allowing us to also test the effect of system-level corruption indicators.
Our results suggest that the belief that women have a lesser tendency to be corrupt is indeed primarily a symptom of views reflecting traditional gender roles and higher levels of gender inequality.Moreover, they also suggest that perceptions of the riskiness in taking part in corrupt practices matters too, albeit to a lesser degree.People who perceive that the risk of being held accountable for behaving corruptly is great are more likely to believe that women are less corrupt than men.However, this positive effect of the perceived riskiness is limited to those contexts where corruption is relatively widespread and gender inequality is high.These findings are in contrast with a previous study by Esarey and Schwindt-Bayer (2018) who find that greater women's representation is associated with lower levels of perceived government corruption in more established democracies.We discuss the potential implications of these results in our concluding section.
The article proceeds as follows.We first discuss previous research on the association between gender and corruption, focusing especially on studies of the gender stereotypes that might account for why some people are more likely to perceive women as less corrupt than men.From this we form a number of hypotheses that are tested in the empirical section of this article.Before presenting our results, however, we also describe our data, variables, and research method.In the final section, we summarize and discuss our findings.Here, we also briefly discuss the potential policy implications of the findings.

Gender and the perceived tendency to engage in corruption
The contemporary academic debate on the association between gender and corruption was kick started by two pioneering studies by Dollar and others (2001) and Swamy and others (2001) showing a relationship between a higher share of women in national parliament and lower levels of corruption. 1 In the following two decades, these studies helped to inspire an extensive body of research that has attempted to untangle the seemingly robust association between gender inequality and corruption and explain the underlying mechanisms (for a recent overview, see Merkle, 2020). 2  While much has been done in the previous few decades in an attempt to explain the observed association between gender and corruption, at least one open question remains: Why are some individuals or societies more (or less) likely to perceive women in general as less corrupt?The answer to this question could potentially provide us with another important piece in the puzzle for explaining for instance the relationship between a higher share of women politicians and lower levels of corruption.
Earlier studies indicate that women politicians under some conditions are seen as more honest and trustworthy than men (Barnes & Beaulieu, 2014, 2019), which could suggest that they are more harshly punished by the electorate for taking part in corruption, thereby making it a riskier affair for women to engage in such behavior (Eggers et al., 2018;Esarey & 1 Corruption is often defined as "abuse of entrusted (or, more specifically, public) power for private gain" (see, e.g., Transparency International, 2021). 2 Here, it is important to acknowledge that the relationship between gender (in)equality and corruption is unlikely to be unidirectional.Plenty of studies have suggested that lower levels of corruption contribute to higher levels of gender equality (Esarey & Schwindt-Bayer, 2018;Stockemer, 2011;Watson & Moreland, 2014).Schwindt-Bayer, 2018).This view has been dubbed "the differential treatment theory of gender and corruption" (see Schwindt-Bayer et al., 2018).If this theory is correct, it could mean that citizens are more likely to punish female candidates associated with corruption, which in turn suggests that women have greater incentives to appear incorruptible.Citizens concerned with corruption could then be more willing to vote for female candidates because of "the perceived incorruptibility of women" (Benstead & Lust, 2018, p. 85).Gender stereotypes have indeed been shown in previous studies to have an impact on political attitudes and behaviors of voters (Dolan, 2010(Dolan, , 2014;;Huddy & Terkildsen, 1993;Sanbonmatsu, 2002).
At the same time, one could expect, in line with the collective action perspective on corruption (Persson et al., 2013), that perceptions of women as less corrupt, combined with a greater number of women in public office, could influence the shared expectations of ordinary citizens concerning the need and inevitability of taking part in corrupt action.Looking at it from another perspective, to say that women are less corrupt is another way of saying that men are more corrupt.A belief that men, who tend to hold a monopoly on (public) power in more patriarchal societies, are more corrupt could therefore foster more corruption from a collective action perspective.This belief or shared expectation could even act as a justification for corrupt behavior, although people generally tend to condemn corruption (Rothstein & Varraich, 2017).Perceptions of women as less corrupt could in turn originate from observations of male-dominated corrupt networks from which women are excluded.
In theory, the introduction of women into public positions could therefore have an impact on the citizens' own behavior when it comes to offering bribes or asking for personal favors, for instance.Corruption should hence be perceived as a less attractive strategy to get things done, which in turn would imply that corruption could be reduced by increasing the share of women in government positions.Consequently, perceptions of women as less corrupt could become a self-fulfilling prophesy.If women indeed have a greater tendency to "radiate integrity" this could in fact be an important explanation for why social norms that promote corruption are weakened (see also e.g., Köbis et al., 2015Köbis et al., , 2019)).The lack of adequate individual-level data has hampered earlier attempts to examine this important norm-related question in a crossnational comparative perspective.The question then remains, what are the reasons behind this belief in the greater integrity of women?
Before moving on, however, some clarification is in order regarding the scope of this study.Unlike many of the previously mentioned studies that examine citizen attitudes toward female politicians and women in elite positions (see e.g., Barnes & Beaulieu, 2014, 2019), this study looks at perceptions of women in general.Care should therefore be taken when extrapolating the findings in this study to perceptions of elites.However, recent studies suggest that elites and political strategists are well aware of a potential gender gap in corruptness, or at least the perceptions thereof among the general population, and therefore sometimes strive to take advantage of it in dealing with political scandals by replacing scandalized (male) politicians with female alternatives (Funk et al., 2021;Valdini, 2019).This practice may in turn reinforce gender stereotypes among the general population.Furthermore, it is possible that citizens take cues from "the real world," where men are more often in leadership positions (Boehm & Sierra, 2015), especially in less equal societies, and are therefore more exposed to corruption.Indeed, there is even some evidence suggesting that when women do reach leadership positions, such as a mayor's office, they may reduce corruption risks, at least temporarily (Bauhr & Charron, 2021;Brollo & Troiano, 2016), which may be noticed by the public, further reinforcing gender stereotypes.
Previous research stresses that there are strong theoretical reasons for believing that men and women are actually affected by corruption in differential ways as a result of gender roles that (re)distribute power resources unequally in societies (Boehm & Sierra, 2015;Merkle, 2020).Women are often claimed to be more vulnerable to corruption while at the same time lacking the means to benefit from taking part in it due to gender-based 17471346, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License asymmetries in power.Boehm and Sierra (2015, p. 2) argue that while men might be more likely to encounter corruption due to their traditional roles as heads of households and active market participants, "women may still proportionally be more vulnerable" (emphasis in original).A study from Uganda, for instance, suggests that while men are more active in the private sector, women are more often targeted by corrupt officials (Ellis et al., 2005).Other studies in turn indicate that men are more exposed to bribe demands, possibly because they are more frequent users of public services; however, it might also be because women more often cannot afford to pay bribes due to the fact that women are likely to have a lower income than men (Boehm & Sierra, 2015, p. 2). 3  Recent studies utilizing survey experiments to explain which psychological factors or "gender stereotypes" might account for people's perceptions of women as less corrupt argue that there are at least three plausible explanations (Barnes et al., 2018;Barnes & Beaulieu, 2019).First, women might be portrayed as more honest and ethical.Second, they can be viewed as more cautious and risk averse.Third, they can be perceived as political outsiders due to their marginalized status across many formal and informal institutions.The findings in these studies suggest that the strongest and most consistent evidence can be found in support of the "risk aversion" stereotype, while the "political outsider" stereotype comes in second place.Meanwhile, as stressed by Barnes and Beaulieu (2019, p. 135), there is relatively little empirical support for the conventional wisdom posited by Dollar et al. (2001), among others, that women are perceived as inherently more honest, ethical, and trustworthy, and hence less corrupt, especially if they belong to the elite (see also Schneider & Bos, 2014).Still, there is some evidence suggesting that men might be more likely to respond to the honesty stereotype (Barnes & Beaulieu, 2019).Furthermore, as pointed out by the same authors, it is possible that the honesty stereotype is a more likely explanation in more conservative or less developed states.However, since we lack appropriate instruments for testing this explanation, we will focus on the two other ones.
In sum, the two most likely theoretical explanations, according to previous research, for the belief that women are less corrupt than men are (a) that women are perceived as less willing to expose themselves to the risks that engaging in corruption can entail (the risk aversion theory), and (b) that women are marginalized and often lack the same power resources that men tend to have, and therefore have fewer opportunities to benefit from abuses of power (the marginalization theory).Following this latter line of reasoning, one could conclude in line with Merkle (2020, p. 76) that "women might simply be less associated with corruption, as they are typically among the powerless in society."

Hypotheses
In this article, we test both of the explanations discussed previously using cross-country data from a wide variety of contexts.To begin with, we examine if an individual's assessment of the risks involved in participating in corruption is in any way related to their view of differences in corruptness across gender.As was already stated, some studies posit that involvement in corruption is considerably riskier for women, and that they therefore should be less likely to participate (see e.g., Esarey & Schwindt-Bayer, 2018).Others again point at previous studies that have indicated that women have a tendency to be more risk averse than men (Jianakoplos & Bernasek, 1998;Watson & McNaughton, 2007).Research on gender stereotypes moreover seems to confirm that women generally tend to be at least perceived to be more cautious and risk averse than men (Eckel & Grossman, 2002;Huddy & Terkildsen, 1993).Perceptions of bribery and other forms 3 According to the UN 1995 Human Development Report, 70% of the poor, who are also said to be more dependent on public services, are women (Boehm & Sierra, 2015, p. 2).
of corruption as a "risky business" could therefore activate the gender stereotype that depicts women as more risk averse.Based on this, one could expect those who perceive corruption as riskier to be more likely to believe that women are less corrupt.Hence, we choose to formulate the following hypothesis: Hypothesis 1a.People who view corruption as riskier are more likely to believe that women are less corrupt than men.
However, the societal context is also likely to influence an individual's corruption perceptions.In societies where taking part in corrupt exchanges is less of a deviance and more of a (informal) norm, the riskiness of behaving corruptly should be significantly lower (see e.g., Persson et al., 2013).Hence, we would expect to find that people on average are less likely to perceive women as less corrupt than men in societies where public sector corruption, especially of the kind that ordinary people are more likely to run into (i.e., petty corruption), is perceived to be more common.
Several studies argue that institutional factors can mitigate the relationship between gender and corruption (see e.g., Alatas et al., 2009;Alhassan-Alolo, 2007;Esarey & Chirillo, 2013;Esarey & Schwindt-Bayer, 2018).Alhassan-Alolo (2007) for instance found that Ghanaian civil servants hold similar attitudes toward corruption no matter their gender.The author argues that women might be as likely to take part in corrupt exchanges as men as long as the environment presents adequate opportunities, networks, and social norms that facilitate corrupt behavior.Meanwhile, Alatas and others (2009) show using data from economic experiments conducted in Australia, India, Indonesia, and Singapore that significant gender differences could only be found in the Australian context.Hence, they argue that the relationship could be culture specific and possibly tied to how patriarchal the society in question is.According to them, "[i]n relatively more patriarchal societies where women do not play as active a role in the public domain, women's views on social issues may be influenced to a greater extent by men's views" (Alatas et al., 2009, p. 678).Consequently, one could expect the perceived gender differences in corruptness to be considerably smaller or even non-existent in these kinds of contexts.
Others, such as Esarey and Chirillo (2013) and Esarey and Schwindt-Bayer (2018), argue that variations in the association between gender and corruption across different institutional contexts are due to the observed gender differences in risk aversion.It is argued that "this is because women are more averse to the risks of violating political norms and because gender discrimination makes violating institutional norms a riskier proposition for women than men" (Esarey & Chirillo, 2013, p. 382).Women's attitudes toward corruption should therefore be tied to how socially stigmatized such behavior is and the likelihood that it will be punished.If there are strong social norms that encourage participation in bribery, nepotism, or other forms of corruption, and it is seen as "an ordinary part of governance supported by political institutions" (Esarey & Chirillo, 2013, p. 383), then there is less likely to be a gender gap with regard to corruption.
However, as noted by Lee and Chávez (2020, p. 476), "[b]ehavioral theories suggest that as women's opportunities and roles more closely simulate those of men, so too will their perceptions and attitudes."Therefore, we could also expect the actual gender gap in corruptness to be larger in more corrupt societies with a higher level of gender inequality, which in turn could also have an impact on the perceived gender gap.Nevertheless, we will still proceed from the theoretical argument that a higher risk should be associated with a greater tendency to perceive women as less corrupt than men.This risk should in turn be higher in societies with strong institutions and anti-corruption norms where corruption is relatively rare.Hence, we form the following hypothesis: Hypothesis 1b.In societies judged to have higher levels of corruption there is a lesser tendency to believe that women are less corrupt than men, ceteris paribus.
We also test if the association between individual perceptions of the riskiness of behaving corruptly and the belief that women are less corrupt than men is stronger or weaker in societies that are generally perceived to have higher levels of corruption.As several scholars have pointed out, the actual forms (and consequences) of corruption can vary quite radically across different types of societies (Graycar & Monaghan, 2015;Johnston, 2005;Kaufmann & Vicente, 2011).Petty or "everyday" corruption and other related phenomena such as clientelism are on one hand said to be more widespread in relatively poor and newly democratized countries.Wealthy consolidated democracies are on the other hand claimed to be characterized by more ambiguous "structural" or "legal" forms of corruption that are less visible to ordinary citizens, and to have less outright bribery or embezzlement.
While the general belief might be lower in societies with less corruption, one could still expect this relationship to be stronger in societies with more widespread corruption where assessments of the riskiness of corruption could be more firmly anchored to actual (direct or indirect) experiences of corruption.The same should also apply for gender inequality since it tends to be strongly correlated with higher levels of corruption (Branisa et al., 2013;Dollar et al., 2001;Swamy et al., 2001).Citizens in these contexts can be argued to be more likely to show a greater awareness of the asymmetries in power that tend to exclude and disfavor women in corrupt exchanges and perceive corruption as a salient issue.The risk aversion stereotype, like other gender-related stereotypes and norms, is also likely to be more prevalent in societies where gender inequality is greater, which in turn suggests that this stereotype is more likely to be triggered in these contexts.This leads us to form the following hypothesis: Hypothesis 1c.The positive association between risk assessments and the belief that women are less corrupt is stronger in societies where corruption is perceived to be more common.
Finally, we contrast the risk aversion theory with the political outsider or marginalization theory.We test this latter theory using both individual-and societal-level indicators of gender (in)equality.
While these are admittedly far from unproblematic indicators of the perceived marginalization of women, we would still argue that they can act as proxies for a belief that women are (and should be) excluded from power.A belief in gender equality does not exclude perceptions of women as political outsiders; however, we find it likely that people who believe in traditional gender roles would also have a stronger tendency to believe in the social exclusion of women.Following the argument of Merkle (2020), among others, that the association between women and corruption might be a reflection of power asymmetries or inequalities between men and women, we hypothesize that there is a higher tendency to perceive women as less corrupt among individuals who hold gender equality in lesser regard.The view that women are less corrupt than men can be considered a form of what some scholars call "benevolent sexism," that is, a view of women as the fairer (and weaker) sex who requires men to provide chivalrous protection and govern society (Benstead & Lust, 2018;Glick & Fiske, 2001;Merkle & Wong, 2020).Therefore, we might expect that those individuals who hold more gender stereotypical and traditional views might be more likely to perceive women as less corrupt.Likewise, we test if this belief is stronger in societies with a higher level of gender inequality, measured using a more objective indicator.We therefore expect that individuals who are more committed to gender equality are less likely to believe in stark gender differences in corruptness.We also expect to find this in more equal societies.Hence, we form the final two hypotheses: Hypothesis 2a.People who value gender equality to a higher degree are less likely to believe that women are less corrupt than men.

Hypothesis 2b.
In societies where the level of gender inequality is higher there is a greater tendency to believe that women are less corrupt than men, ceteris paribus.

DATA , VA R I A BLE S , A N D M ET HOD S
The research model guiding our empirical analyses can be seen in Figure 1.First, we establish the associations between individual perceptions of the riskiness involved in engaging in corruption plus the respondent's attitude toward gender equality and the tendency to believe that women are less corrupt than men (Hypotheses 1a and 2a).Second, we examine the relationships between our key country-level variables (the Corruption Perceptions Index and Gender Inequality Index) and our dependent variable "WLC" (Hypotheses 1b and 2b).Last, we examine if and how the relationship between individual perceptions of the riskiness of corruption and WLC hinges on the Country-level Corruption Perception (CCP), as indicated by the dotted arrow from this box (Hypothesis 1c).

Data
The individual-level data originate from the latest wave (Wave 7) of the World Values Survey conducted in the period 2017-2020 (Haerpfer et al., 2020).This dataset is ideal for the purposes of this study since it contains an extensive array of questions regarding both corruption perceptions and gender equality.Next we present information on key variables while more detailed information regarding question wordings and the coding of variables can be found in Table B1 in the Appendix.

Dependent variable
The dependent variable of the study is WLC ("Women less corrupt"; Table 1).It is operationalized using an item asking respondents if they believe that women are less corrupt than men (Can you tell me how strongly you agree or disagree with the following statement: On the whole, women are less corrupt than men).The five response alternatives are (0) "Hard to say," (1) "Strongly agree," (2) "Agree," (3) "Disagree," and (4) "Strongly disagree."The variable is reversed and recoded to range from 1 ("Strongly disagree") to 5 ("Strongly agree") and "Hard to say" is treated as a neutral response alternative (3) positioned in the center of the scale (see Figure B1 in Appendix B for the distribution of the dependent variable).
However, there are at least two potential issues with this item.First, due to the relative vagueness of this item, it must be considered that disagreement with the statement that "women are less corrupt than men" can hold at least two different meanings: [a] that the respondent thinks that men and women are equally corruptible, or [b] that the respondent thinks that men are less corrupt than women.In the first case, this view might reflect a more gender egalitarian disposition, while in the second case it might reflect what scholars call "hostile sexism," a sort of antipathy against women who are perceived as a threat to men's power (Benstead & Lust, 2018;Glick & Fiske, 2001).This means that a low level of "benevolent sexism" in theory could be due to a high level of "hostile sexism," not more gender egalitarian views.Unfortunately, we have no way of checking which of these interpretations is more correct for each respondent due to a lack of data on this.However, as noted by Glick and Fiske (2001), there seems to be a positive correlation between these two separate, but complementary, types of sexist dispositions that is especially strong on the country level. 4Hence, we would also expect to find that higher amounts of hostile sexism often go hand in hand with benevolent sexism.
Second, it could be argued that the question is somewhat leading, which could in turn have an impact on individual responses.However, since alternative indicators for this are not currently available, this question item is used in the analyses. 4Glick and Fiske (2001) find only moderate positive correlations on the individual level.Their explanation for this correlation between two seemingly conflicting feelings or beliefs is that they are directed toward different female subtypes.For example, a person can simultaneously have strongly positive feelings toward housewives and strongly negative feelings toward career women or feminists.In general, though, the authors argue that these same individuals are likely to experience ambivalence, or conflicting feelings ("ambivalent sexism"), toward individual women.

Independent variables
Our first key individual-level independent variable is RoC ("Riskiness of Corruption") This variable is operationalized using a survey item asking for the respondent's view on the risk of being held accountable for taking part in corrupt exchanges (How high is the risk in this country to be held accountable for giving or receiving a bribe, gift or favor in return for public service?).
The response alternatives range from 1 ("No risk at all") to 10 ("Very high risk").This scale is recoded to range from 0 to 1 (1 = highest perceived risk).
Our second key individual-level independent variable is Attitude to gender equality (Gender Equality).It is operationalized using a pre-existing index that measures the respondent's attitude toward gender equality: the Emancipative Values-2: Equality sub-index (Welzel, 2013).This index is constructed using three items indicating how strongly respondents disagree with the following statements: [1] "Education is more important for a boy than a girl," [2] "When jobs are scarce, men should have priority over women to get a job," and [3] "Men make better political leaders than women."5It ranges between 0 and 1 where a higher value indicates that the respondent values gender equality to a higher degree.
Our first key country-level independent variable is country-level corruption perception (CCP).This variable is operationalized with Transparency International's Corruption Perceptions Index (CPI).This is the most widely used indicator of corruption worldwide and has been compiled on a yearly basis since 1995.It uses (since 2012) a scale of 0 to 100, where 0 is highly corrupt and 100 is very clean.Here, the scale has been reversed and recoded so that it ranges from 0 (very clean) to 1 (highly corrupt).This variable also functions as a moderating variable to check if the impact of RoC varies depending on the level of CCP.
Our second key country-level independent variable is GII (Gender Inequality Index).It is a composite measure that reflects inequality in achievements between women and men in the following three dimensions: reproductive health, empowerment, and the labor market.The maternal mortality ratio and the adolescent fertility rate are indicators for the health dimension, while the share of parliamentary seats held by each gender and the secondary and higher education attainment levels of women are considered to be indicators of the empowerment dimension.Finally, the labor dimension is measured by women's participation in the workforce (United Nations Development Programme, 2021).This index uses a scale from 0 to 1 where higher values indicate higher levels of gender inequality.

Control variables
To verify the associations found here, we include a number of control variables known to influence attitudes toward gender roles in general (Bolzendahl & Myers, 2004;Fodor & Balogh, 2010).At the individual level, we include the standard sociodemographic factors age (measured in years), gender (male or female), educational level (following the International Standard Classification of Education, ISCED), scale of incomes (step 1 to 10), marital status (married or not), and place of living (urban or rural).We also control for religiosity using the following item in the World Values Survey: "How often do you attend religious services?"("Never" to "More than once a week"). 6ince this study is interested in beliefs regarding gender differences in corruptness, it will also control for the respondent's perceptions of corruption.It operationalizes individual-level corruption perception (ICP) using two items that measure two different aspects of corruption. 7 One asks about perceptions of corruption in general (ICP1: How would you place your views on corruption in [your country] on a 10-point scale where "1" means "there is no corruption in [my country]" and "10" means "there is abundant corruption in [my country]"); and a second focuses more on experiences of petty or street-level corruption (ICP2: We want to know about your experience with local officials and service providers, like police officers, lawyers, doctors, teachers, and civil servants in your community.How often do you think ordinary people like yourself or people from your neighborhood have to pay a bribe, give a gift or do a favor to these people in order to get the services you need?Does it happen never, rarely, frequently, or always?).
At the country level, we include a measure of democracy: The Polity index combined score for autocracy (−10) to democracy (10) (Polity V 2018). 8Table B2 in Appendix B shows the correlations between all the variables included in the main analyses.

Method
Multilevel regression analysis is used to take into account that the respondents are nested in countries (Rabe-Hesketh & Skrondal, 2012).We utilize multilevel linear regression analysis in our main models; however, we also use multilevel logistic regression analysis to test the robustness of our conclusions.We use unweighted data, which means that the results are not necessarily representative.(The regression results are presented in Table 3, where models 1 (M1) to 4 (M4) are fixed effects bivariate regressions between the dependent variable in question and our four independent variables, while M5 includes all control variables.M6 in Table 3 includes a cross-level interaction effect between perceived riskiness of behaving corruptly [RoC] and CCP as well as a random intercept for RoC in order to examine how the association between RoC and WLC varies depending on the generally perceived extent of corruption in the country).We also discuss the implications for the hypotheses with the help of plots of marginal means in order to clarify the implications following the recommendations of Brambor et al. (2006).

EM PI R ICA L A NA LYSE S
We begin by presenting country-level differences in WLC, perceived riskiness of corruption, attitude toward gender equality, country-level corruption perception, and the GII in Table 2.As we can see from the table, there is considerable cross-country variation in the extent that people agree with the statement that women tend to be less corrupt than men (WLC).The strongest average support for this claim on a scale from 1 to 5 can be found in Egypt (3.57) and Pakistan (3.50), the weakest in Ukraine (2.45) and Germany (2.51).
Meanwhile, the average perceived risk of being held accountable for taking part in corruption is highest in Ethiopia and Jordan (.83 in both countries) and lowest in Peru (.42) and Ukraine (.44).There seems to be some evidence in support of the hypothesized positive 7 Although both items measure perceptions of corruption, they are quite weakly correlated (Pearson's r = .198).We can hence include them both in the same models.
T A B L E 2 Country-level differences (N = 42).Comparing the country scores for WLC and our country-level corruption indicator (CCP) we see a positive association, which would seem to suggest that there is a greater tendency to perceive women as less corrupt than men in countries where corruption is judged to be more common.This finding is contrary to our hypothesis.However, there are also cases such as Ukraine and Guatemala where corruption is perceived to be relatively widespread, but WLC is low.

Riskiness of corruption
As expected, there seems to be quite a strong positive association between the level of gender inequality and people's view of women as less corrupt than men: Pakistan stands out as the country with the highest level of gender inequality and the second highest average WLC score.South Korea and Germany, the countries with the lowest and second lowest GII scores meanwhile stand out with relatively low WLC scores (2.62 in the case of South Korea).A similar trend can also be seen regarding our individual-level gender equality indicator.
Table 3 displays the multilevel linear regression results for our dependent variable WLC or "Women less corrupt."From the intraclass correlation coefficient (ICC) in the so-called null model (M0), we can see that approximately 7.1% of the variance can be explained by contextual factors, while the rest (92.9%) is explained by individual factors.Models 1 to 4 show the individual effects of our four independent variables without any other variables.Model 5 includes all four independent variables plus our control variables.Finally, Model 6 adds a cross-level interaction effect to check if the association between the perceived riskiness of taking part in corrupt exchanges (RoC), and our dependent variable varies across contexts depending on the actual level of (perceived) corruption.
Our first key individual-level variable of interest, RoC, shows in M1 a highly significant positive coefficient (B = .096,p = .000),suggesting that people who perceive a higher risk of being held accountable for taking part in corrupt exchanges are more likely to perceive women as less corrupt than men.One standard deviation (SD) change in RoC has the effect of approximately .029 on WLC.This finding also holds with very little change when we add the rest of the variables in M5, which means that we find support for our Hypothesis 1a.Turning to our second individual-level variable of interest, attitude to gender equality, we see a strong and highly significant negative coefficient (B = −.463,p = .000)in M2.One SD change in attitude to   gender equality has the effect of approximately .123 on WLC.This suggests people who value gender equality and equal opportunities for men and women to a higher degree tend to be considerably less likely to state that women are less corrupt than men.This effect remains and even becomes stronger when we add the rest of the variables in M5.In other words, as expected we find strong support for Hypothesis 2a.Next, we examine the effect of our first country-level variable of interest, the estimated level of corruption in a specific country or CCP.M3 shows a highly significant positive coefficient (B = .595,p = .000),suggesting that the belief that women are less corrupt than men tends to be more common in countries where experts estimate that corruption is more widespread.This is contrary to our Hypothesis 1b.However, once we add the rest of the variables, or more specifically, the GII, the sign of this coefficient changes and becomes negative but non-significant (B = −.351,p = .092). 9 Hence, it would seem like the observed positive association was actually due to the higher levels of gender inequality in more corrupt societies.Nevertheless, due to the lack of significance following the conventional threshold, we are forced to conclude that we find no support for Hypothesis 1b.The broad confidence interval on the left-hand side of the graph suggests that we would need more data from so-called "less corrupt countries" in order to corroborate the observed relationship.
Turning to our second country-level independent variable, the GII, we see in M4 a very strong and highly significant positive coefficient (B = .714,p = .000).This indicates that people in countries with higher levels of gender inequality are considerably more likely to claim that women are less corrupt than men.Controlling for the other factors in M5 does not change the fact that this variable shows the by far largest coefficient.In other words, we find very strong support for Hypothesis 2b.
Turning our attention to the control variables in M5, we see that both indicators of ICP show significant coefficients but different signs.The indicator of general corruption perceptions (ICP1) shows a relatively weak positive effect while the indicator of petty or "street-level" corruption shows a stronger negative effect.Perceptions of corruption among local service providers hence seem to weaken the tendency to believe that women are less corrupt than men, in contrast to general perceptions where the association is considerably weaker and even positive.This is in line with our argument that perceptions that are more firmly anchored in direct or indirect (e.g., via friends or family) experiences of corruption are more likely to influence people's view of gender differences in corruptness.Men and people living in urban environments are also less likely to agree with the statement that women are less corrupt than men, while older people and people who are married tend to agree with it.Of these six variables, age seems to have the greatest impact.The rest of the control variables fail to reach statistical significance.
Finally, we examine if there are any differences in the effect of our key individual-level explanatory variable RoC across different corruption contexts.Hence, the final model, M6, adds a cross-level interaction effect called "RoC#CCP."As we can see from M6, the coefficient for this interaction effect is positive but non-significant (B = .250,p = .162).Still, as argued by Brambor and others (2006, p. 12), it is still too early to draw any conclusions based on the lack of significance without looking closer at this interaction effect.Figures 2 and 3 therefore visualize the implications of this interaction and how the marginal effect of the perceived riskiness of corrupt behavior changes with the level of corruption in a country.
The figure above suggests that the effect of the perceived riskiness of being held accountable for behaving corruptly is significant and positive (B = .170,p = .014in the most corrupt 9 The variance inflation factor values for CCP and GII are 2.63 and 2.71, respectively.context) only in contexts with relatively high levels of corruption (.7+ on the current to 1 scale); i.e., where corruption is perceived to be relatively widespread.One SD change in RoC has the effect of approximately .052 on WLC in a high-corruption context.In societies where the corruption level is judged to be below this point, risk perceptions do not seem to matter at all.Similar patterns are found when we substitute our corruption indicator for our gender inequality indicator (see Figures A1 and A2 in Appendix A).These findings support our Hypothesis 1c.Not only is the positive association stronger in societies where corruption is perceived to be more common and gender inequality is greater, it is practically non-existent elsewhere.

F I G U R E 2
The heterogeneous effect of RoC on WLC across CCP, w controls (95% CI).
F U R E 3 Contrasting effects of RoC on WLC across level of corruption, w controls (95% CI).We also perform a few tests in order to check the robustness of our findings (see Appendix A).Firstly, we show our interaction effect without any controls (see Figure A3).Secondly, we use an alternative (dummy) version of our dependent variable and multilevel logistic regression analysis.The interaction effect in this model shows a significant positive coefficient (B = .480,p = .043),suggesting that the probability of agreeing with the statement that women are less corrupt than men increases from .39 to .46 (+.07) when you move from the lowest perceived risk to the highest in a high corruption context.Figure A4 in Appendix A illustrates this effect.You can compare this with the by far strongest individual-level effect, the respondent's view of gender equality, where the estimated probability decreases from .55 to .35(−.2) when moving from the most negative view of gender equality to the most positive.Finally, we control for a couple of potential outliers (see Figure A5).Our conclusions are not substantially affected by any of these robustness tests.

DI SC US SION A N D CONC LUSIONS
The purpose of this article was to harness new cross-national survey data to test some of the main explanations as to why women are perceived as less corrupt than men.More specifically, we tested two of the dominant theories in the literature as to why this might be the case: risk aversion theory and marginalization theory.
Unsurprisingly, we find that individual attitudes toward gender equality and the overall level of gender inequality in a society are the by far strongest predictors of the tendency to agree on the notion that women are less corrupt than men.These findings provide strong support for the marginalization theory: those who believe in traditional gender roles or stereotypes where men tend to dominate in both the public and private spheres are more likely to agree with this assumption.It can hence be viewed as a form of benevolent sexism where women are perceived as the "weaker sex" who are less likely to possess the power resources (e.g., social networks and opportunities) or "masculine traits" (e.g., "wickedness" and "iron-heartedness") needed to successfully behave in a corrupt way and get away with it (Benstead & Lust, 2018;Glick & Fiske, 2001;Merkle & Wong, 2020; see also Alhassan-Alolo, 2007, p. 235).Similarly, this belief is considerably more common in societies where gender inequality is estimated to be higher based on indicators of women's reproductive health, empowerment, and workforce participation.These findings can be interpreted as support for Barnes and Beaulieu's (2019, p. 159) argument that in societies where women's marginalization is greater it "may prove a more persuasive general explanation for the gender gap in corruption perceptions." In addition, we also find some support for the risk aversion theory: those who perceive the risk of being held accountable for taking part in corrupt exchanges as high are more likely to agree with the statement that women are less corrupt than men.However, this association can mainly be found in societies where corruption is generally estimated to be more commonly occurring according to the CPI and where gender inequality is greater according to the GII.One potential explanation for these findings is that this is considerably more likely to happen in contexts where "street-level corruption" (i.e., corruption involving police officers, doctors, teachers, and other street-level civil servants whom women are more likely to encounter due to their greater dependence on public services; Boehm & Sierra, 2015, p. 2), is more common.These contexts also tend to be characterized by extensive gender inequality, which could also explain why the risk aversion stereotype has a greater chance of being triggered here.Meanwhile, those who perceive the lowest risk in behaving corruptly in high-corruption contexts seem to be the least likely to agree that women are less corrupt than men.
The fact that we find support for both the marginalization and the risk aversion theories is perhaps not that surprising since it could be argued that both are closely linked and rooted in traditional gender roles.This might also explain why the perceived riskiness of corruption mainly seems to matter in contexts where gender inequality is high.At the same time, however, we can also conclude that the belief that women tend to be less corrupt is on average considerably stronger in societies where gender inequality is high.Citizens in highly unequal and corrupt societies who perceive corruption as a very risky venture are most likely to hold this belief, which could potentially alter the citizens' own behavior in relation to female civil servants or politicians since they do not come to suspect or expect corruption as strongly as with male ones.Increasing the presence of women in the public sector while increasing accountability could hence contribute to reducing perceptions of corruption in these contexts.As argued by Barnes and Beaulieu (2019, p. 159), women politicians may also take advantage of a situation where corruption is of high political concern by crafting an effective political image that plays on these stereotypes of women as "outsiders" or "careful and cautious."However, as noted by Alhassan-Alolo (2007, p. 237), if women's so-called "superior morality" fails to live up to the expectations it risks "undermining overall efforts at gender mainstreaming on the basis of equality."Gender equality is valuable in and of itself and should be pursued regardless of its potential effects on corruption.
These findings also suggest that gender equality-promoting policies might reduce the potential utility of portraying women as an "anti-corruption force" or "political cleaners."As gender equality increases, people no longer come to expect as strongly as before that women are intrinsically less corrupt than men.In other words, increasing the representation of women in public organizations with the aim of reducing corruption could paradoxically result in this anti-corruption strategy losing its effect, as people become more and more used to seeing women in positions of power.
From a gender standpoint, it is also interesting to note that women themselves seem to be more likely than men to agree that women are generally less corrupt.This finding could potentially indicate that women perceive themselves in general as more risk averse, marginalized, and/or vulnerable to corrupt behavior and hence less likely to (be able to) take part in corrupt exchanges.Alternatively, it could be part of a more general gender bias; however, more research is needed to explore this question further.
More studies are also needed to determine if and when a public official's gender and a citizen's belief regarding gender differences in corruptness actually influences how the citizen chooses to approach the public official.10Likewise, while our cross-sectional approach is useful for examining broad country-level variations in the associations between the different variables, the nature of our data makes it difficult to draw any definitive conclusions regarding causal directions.Furthermore, it is important to acknowledge that our tests of the risk mechanism and the marginalization mechanism are not direct tests of either mechanism.In both cases, we utilize proxies in order to examine the mechanism in question, due to a lack of better existing alternatives.Future studies should strive to develop more appropriate survey instruments for measuring both the stereotype that corruption is riskier for women and the belief that women are marginalized in society.Preferably, these studies could also attempt to measure if people believe that women are inherently more moral or honest than men, so that the "fairer sex" mechanism (Esarey & Chirillo, 2013) could also be examined in an adequate way, which we could not do here due to a lack of a suitable instrument.Moreover, the relationship between gender (in)equality and corruption is unlikely to be unidirectional.Plenty of studies have suggested that lower levels of corruption contribute to higher levels of gender equality (Esarey & Schwindt-Bayer, 2018;Stockemer, 2011;Watson & Moreland, 2014).Hence, we need more studies with causal identification strategies to determine the direction of this association.
Another limitation is that our dependent variable treats corruption as a single homogenous phenomenon.Differences in the perceived corruptness of men versus women could depend on the characteristics of the corruption phenomenon itself.One path forward is therefore to differentiate between the different forms and types of corruption such as need and greed corruption (see Bauhr & Charron, 2020).The perceived difference could for instance be greater in the case of need corruption since women are argued to be more exposed to this type of corruption due to their traditional caretaking roles.In short, more research is needed on this subject.

A PPE N DI X A ROBUSTNESS CHECKS F I G U R E A 1
The heterogeneous effect of RoC on WLC across GII, w controls (95% CI).

F I G U R E A 2
Contrasting effects of RoC on WLC across of gender inequality, w controls (95% CI).

F I G U E 4
The heterogeneous effect of RoC on WLC across CCP, alternative dependent variable, w controls (95% CI)..000 .000 .000 .000 .

17471346, 0 ,
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 3 Multilevel regression results for "women less corrupt" (WLC).Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

17471346, 0 ,
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License F I G U R E A 3 The heterogeneous effect of RoC on WLC across CCP, w/o controls (95% CI).

17471346, 0 ,
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E B 2 Correlations between variables.
T A B L E 1 17471346, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/polp.12578 by Mid Sweden University, Wiley Online Library on [31/01/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License between perceived RoC and WLC, at least on the country level: Both indicators show low scores in Ukraine and Germany, for instance, and high scores in Egypt and Jordan.However, there are also countries that clearly do not fit the pattern, such as Greece and Guatemala with low WLC and high RoC scores, or Bangladesh and Iran with high WLC and low RoC scores.There also seems to be a rather weak positive correlation between RoC and our country-level corruption indicator.Contrary to what was expected, taking part in corrupt exchanges tends to be perceived as riskier in countries perceived by experts as more corrupt; however, there are exceptions such as Mexico and Iran where the average perceived risk is relatively low.