Introduction

There are many examples of female Muslim politicians being targeted by politicians of the Populist Radical Right (see Farris, 2017; Oudenampsen, 2016), sometimes leading to female Muslim politicians receiving extraordinary amounts of discursive backlash (Saris & Ven, 2021; van Oosten, 2022). At the same time, Muslim women tend to outnumber Muslim men in politics (Hughes, 2016), especially in contexts where party selectors craft candidate lists: Muslim women tick two diversity boxes while also challenging stereotypes of Muslim women as oppressed, simply by being politicians (Dancygier, 2017). Despite these challenges and the unique positioning of Muslim women in politics, the question remains how voters evaluate them. Does being a female Muslim politician pose electoral challenges, or is there an electoral benefit? In this paper, I test whether intersectionality plays a role in how voters evaluate female Muslim politicians.

An intersectional analysis is distinct from a unitary or multiple one (Hancock, 2007). Where a unitary analysis foregrounds one background characteristic (race or gender) and a multiple analysis adds up the effects of multiple ones (race and gender), an intersectional analysis highlights the interaction between them (race interacts with gender) (idem). In order to study the intersectional position of minoritized women in politics quantitatively, many scholars call the use of interaction effects and candidate experiments viable methodological solutions (Block et al., 2023; Klar & Schmitt, 2021, p. 493, 495). This paper tests the limits of both the method of data collection (candidate experiments) and the method of analysis (interaction effects) by studying what is arguably a most-likely case: female Muslim politicians.

Though there has been much research on intersectionality and politicians in the US (Brown, 2014a, 2014b; Collins, 1998; Holman & Schneider, 2018; Lemi & Brown, 2019; Reingold et al., 2020), intersectionality and politicians in the European context is poorly understood. In Europe, Muslim women play a crucial role in many nationalist debates in western countries such as France, Germany and the Netherlands (Dancygier, 2017; Korteweg & Yurdakul, 2021). The general framing tends to imply that Muslim women are significantly different from both non-Muslim women and Muslim men because being Muslim influences what it means to be a woman and being a woman influences what it means to be Muslim. As Islam and gender are thus “mutually reinforcing”, an intersectional lens is indispensable (Crenshaw, 1991, p. 1283). This is particularly apparent when female Muslim politicians attempt to enter politics (Dancygier, 2014; Hughes, 2016; Murray, 2016). However, whether female Muslim politicians face a “double disadvantage” or a “strategic advantage” (Gershon & Lavariega Monforti, 2021) depends heavily on the specific political and societal context in which they operate. In order to study this, I presented 3056 respondents in France, Germany, and the Netherlands a total of 18,336 short bios of hypothetical politicians while randomizing their religion, ethnorace and gender. I asked respondents to assess these politicians by asking evaluation and choice-questions. Candidate conjoint experiments rarely include Islam as an experimental condition and when they do, intersectional analyses are rarely conducted (one notable exception being Benstead et al., 2015).

In line with Hancock (2007), I analyze the results in a unitary, multiple and intersectional way. In the intersectional analysis I use interactions while controlling for direct (unitary) effects. Although I do not find voters assess women and ethnoracially minoritized politicians negatively, I find robust and consistent evidence that voters have a strong negative and unitary bias against Muslim politicians. However, this analysis did not garner any evidence for intersectional effects of religion and gender. Given the sizable sample and effect sizes, I do not consider a lack of statistical power the cause of these null results. Though I remain confident that interaction effects are the most fitting method of analysis, I argue that conjoint experiments are not the most fitting method of data collection due to the cognitive overload causing respondents to single out one attribute to base their choices on.

Theoretical Framework

Do (intersections of) politician religion, gender, and ethnorace influence how voters assess them? In an intersectional approach, the effects of race and gender are not merely added together but are ‘more than the sum of mutually exclusive parts’ (2007, p. 65). In this theoretical framework, I discuss how racism, sexism, and Islamophobia are widespread and how context matters in the extent to which ethnoracial minorities, women and Muslims are underrepresented in politics. Political elites can thwart the ambition of minoritized candidates (Dancygier et al., 2021) as they might have reasons to expect that members of these groups might not do well at the ballot box. Yet, experiments in which candidate attributes are randomly presented to respondents reveal that voters are generally slightly more positive about ethnoracially minoritized, women, and Muslim politicians than about their majoritized counterparts (Aguilar et al., 2015; Bai, 2021; Brouard et al., 2018; Campbell & Cowley, 2014; Carnes & Lupu, 2016; Hainmueller et al., 2014; Schwarz & Coppock, 2022; van Oosten et al., 2024; Visalvanich, 2017; Weaver, 2012). That is why I pre-registered (van Oosten, 2020) hypotheses stating that I expect respondents to not be biased towards politicians with these background characteristics. But what if these background characteristics are combined? I discuss the literature on intersectionality and possible quantitative operationalizations of the concept, whilst structuring the theoretical framework with Hancock’s (2007) three distinctions of a unitary, multiple and intersectional analysis.

Unitary Analysis of Ethnorace, Gender, and Religion

I turn to the largely separate literatures on the political underrepresentation of 1) ethnoracially minoritized citizens, 2) women, and 3) Muslims. First, ethnoracially minoritized citizens. Racism is widespread, also across Europe (Benson & Lewis, 2019; FRA, 2017; Lentin, 2008). It reaches across many domains of life: social media (Patton et al., 2017), education (Harwood et al., 2018), night life (May & Goldsmith, 2017), sports (van Sterkenburg & Blokzeijl, 2017). Moreover, research has shown that racist attitudes shape voting behavior (Weller & Junn, 2018). Ethnoracially minoritized politicians are underrepresented in most parliaments (Fernandes et al., 2016, pp. 2, 4; Hughes, 2013). Research from the US teaches us that party leaders are more hesitant to select and support Black and Latinx candidates (Doherty et al., 2019) and the same applies to candidates of immigrant descent in Europe (Dancygier et al., 2021). Is this fear of electoral repercussions real? Do voters discriminate against ethnoracially minoritized politicians? Much experimental research has been done on this in in the US (Lerman & Sadin, 2016) and some in Brazil (Aguilar et al., 2015), Uganda (Carlson, 2015), Afghanistan (Bermeo & Bhatia, 2017), and Europe (Brouard et al., 2018). Although old-fashioned racism has been shown to influence attitudes and subsequent voting behavior in some groups of individuals (Tesler, 2012, 2013, 2015), experimental research averaging out all respondents together rarely shows a statistically significant negative direct effect of bias on voter assessments of ethnoracially minoritized politicians (e.g., Aguilar et al., 2015; Brouard et al., 2018; Carnes & Lupu, 2016; Hainmueller et al., 2014; van Oosten et al., 2024; Visalvanich, 2017; Weaver, 2012) though some researchers sporadically find negative effects (Krupnikov et al., 2016; Peterson, 2017; Sances, 2018).

Second, sexism is widespread (Eagly & Wood, 2012; Hall et al., 2019), spanning many domains of social life: the workplace, academia and STEM (Bocher et al., 2020; Phipps et al., 2018), health care (Verdonk et al., 2009), and media coverage of female politicians (Aaldering & van der Pas, 2020); and women in politics have to deal with more harassment and violence than men do (IPU, 2018; Krook, 2018, 2019; Saris & Ven, 2021). Women are descriptively underrepresented in national parliaments across the world (EIGE, 2019; Hughes, 2013, p. 501), so much so that young girls think that politics is more for men (Bos et al., 2021). Party elites are hesitant to support female candidates due to the fear that others will not support the candidate, leading to so-called strategic discrimination: party selectors may not be inhibited by sexist ideas themselves, but they are hesitant to select women to top positions due to a fear that voters will punish them for it(Bateson, 2020). Is this fear of electoral repercussions real? Do voters discriminate against female politicians? Experimental research rarely shows a statistically significant negative direct effect of bias on voter assessments of female politicians, in fact, the female candidate does better than the male one on average (Schwarz & Coppock, 2022).

Third, Islamophobia is widespread across western countries (Cesari, 2013; Khalimzoda et al., 2025; Lajevardi & Oskooii, 2018): from the United States (Lajevardi, 2020) to Europe (Abdelkader, 2017; Finlay & Hopkins, 2020; Simon & Tiberj, 2018) and Australia (Mansouri & Vergani, 2018). It reaches across many domains of life: social media (Awan, 2014), finding a job (Di Stasio et al., 2019; Weichselbaumer, 2020), even mere interactions with people on the street (Choi et al., 2021), and much more (Helbling & Traunmüller, 2018; Meer & Modood, 2012). Muslims are underrepresented in national legislatures (Hughes, 2016), not so much in the Netherlands (Aktürk & Katliarou, 2021, p. 399-400) but particularly so in France and Germany (idem, p. 393).

Political parties struggle with dilemmas of including Muslims: one the one hand they want to convey to voters that they are dedicated to diversity, on the other hand they fear a possible backlash from voters who scrutinize Muslim political leaders on liberal values and individual freedoms (Dancygier, 2017). Is this fear of electoral repercussions real? Do voters discriminate against Muslim politicians? Little experimental research has been done on this, yet US and UK research points towards no statistically significant direct effects of politicians being Muslim on assessments by voters (Bai, 2021; Campbell & Cowley, 2014). This has never been researched in mainland Europe though, where outcomes might be different than the US and UK. Research in Denmark shows that voters do penalize candidates with an Arabic (versus Danish) last name (Dahl & Nyrup, 2021, p. 209); this might extend to Muslim politicians outside of Denmark.

In summary, existing research from a unitary perspective suggests an absence of singular effects of ethnorace, gender, and religion by themselves. I thus expect to find null effects for my unitary analysis.

 Therefore, I pre-registered (van Oosten, 2020) the following hypotheses at OSF:

-       H1.a. Voters prefer ethnoracially minoritized politicians equally to ethnoracially majoritized politicians.

-       H1.b. Voters prefer Muslim politicians equally to Christian and non-religious politicians.

-       H1.c. Voters prefer female politicians equally to or more than male politicians.

Multiple Analysis of Islam and Gender

Do voters tend to discriminate against politicians with multiple underrepresented or minoritized background characteristics? In political science, gender and ethnorace are commonly studied together (Brown, 2014a, 2014b; Celis et al., 2014; Collins, 1998; Holman & Schneider, 2018; Hughes, 2013; Lemi & Brown, 2019; Reingold et al., 2020), which extends to experimental studies (Campbell & Cowley, 2014; Krupnikov et al., 2016). Because Islam is so rarely included in experiments or studies on descriptive representation and politics (except Dancygier, 2017; Hughes, 2016), very little is known about the dynamics of descriptive representation of Muslim women in the west (except in Tunisia, Benstead et al., 2015; in Jordan, Kao & Benstead, 2021). Beyond (experimental) political science on descriptive representation, however, Muslim women have received much (scholarly) attention (e.g., Ahmed, 2020; van Es, 2019; Zimmerman, 2015).

There are many accounts in which Muslim women are discriminated against: for instance, Muslim women are less likely to be invited to job interviews than white women and men (Weichselbaumer, 2020). The discrimination against Muslim women even applies in seemingly mundane situations. When a Muslim woman drops a bag of oranges on the street, research (Choi et al., 2021) shows that they receive significantly less help from passersby. This is particularly the case amongst Muslim women who had conveyed to these passersby that they have more conservative views on gender equality issues, a penalty that does not apply to non-Muslim women (idem).

In the context of European politics in which populist parties are gaining political ground, Muslim women are the objects of fierce femonationalist debates. The literature on femonationalism contends that critics of Islam, mainly populist radical right politicians weaponize gender equality as a cause in order to discredit Muslims (Farris, 2017; Rahbari, 2021), thus causing Muslim women to occupy a particularly contentious place in societal debates (Korteweg & Yurdakul, 2021). Liberal values such as gender equality pose dilemmas of inclusion for party selectors who want to diversify their ranks but fear including Muslims will make their party vulnerable to criticisms concerning, particularly, gender equality (Dancygier, 2013; Dancygier, 2017). The weaponization of gender equality to discredit Muslims is reminiscent of similar weaponizations of gay (van Oosten, 2022) and Jewish people (van Oosten, 2024a; 2024b; 2024c) in an effort to weaken alliances by underlining internal divisions (van Oosten, 2025a), yet because Muslim women in politics are much more common than gay Muslim politicians, the weaponization of gender equality is most relevant to dilemmas of inclusion (Dancygier, 2017). However, in proportional systems with relatively long party lists, party leaders will be more keen on selecting Muslim women than Muslim men through their embodiment of Islam and female empowerment simultaneously (Dancygier, 2017), often conditional on the denouncement of conservative values that are seen as inherently Muslim (Aydemir & Vliegenthart, 2016; 2022; Bird, 2005, p. 439; Kundnani, 2012; Murray, 2016). These counteracting trends make it difficult to predict whether being a Muslim woman in politics, compared to being a Muslim man, is an advantage or a disadvantage. Either way, to what extent are these studies on Muslim women intersectional? In the following section, I will discuss Hancock’s (2007) distinction between a multiple and an intersectional analysis.

Intersectional Analysis

Intersectionality as a research paradigm (Hancock, 2007) criticizes the analysis of unitary categories as “discrete and pure strands” (Brah & Phoenix, 2004, p. 76). Simply adding up the effects of multiple categories also is not sufficient as the categories are considered to be “mutually reinforcing” (Crenshaw, 1991, p. 1283). Indeed, researchers cannot categorize individuals into singular neat straightforward boxes, but all individuals are located on the intersection of numerous categories that arguably influence each other back and forth. This applies to both subordinate (Purdie-Vaughns & Eibach, 2008) and dominant (Carbado, 2013) background characteristics (Hancock, 2013, p. 506), leading to “strategic advantage” or “double disadvantage” (Gershon & Lavariega Monforti, 2021).

Hancock (2013; 2007) is the clearest critic of additive conceptualizations of intersectionality. Indeed, an intersectional approach is more than just adding up the effects of being Muslim and being female. Understanding intersectionality as more than a sum of its parts is key to its conceptualization and operationalization. Without understanding intersectionality in this way one would not be able to make sense of some otherwise puzzling phenomena. Indeed, research on politicians’ multiple identity categories reveal mutually reinforcing mechanisms that occasion either “strategic advantage” or “double disadvantage” (Gershon & Lavariega Monforti, 2021). Double disadvantage posits that the disadvantages politicians face are more than a sum of their subordinate group memberships. Strategic advantage means that multiple disadvantaged background characteristics could amount to less disadvantage than a sum of its parts, in other words, belonging to more than one disadvantaged group actually cancels out part of the negative effect of the disadvantaged categories. Muslim women are more often elected than Muslim men, particularly in proportional systems where Muslims make up a significant share of the electorate, such as Germany and the Netherlands (Hughes, 2016; Dancygier, 2017).This means that having multiple disadvantaged background characteristics simultaneously could end up being an advantage compared to those who are only singularly disadvantaged (Purdie-Vaughns & Eibach, 2008) such as Muslim men or non-Muslim white women.

In quantitative experimental research, one often mentioned yet “underutilized” (Klar & Schmitt, 2021, p. 495) operationalization of intersectionality is to use interaction effects (Block et al., 2023; Bowleg, 2008, p. 319; Choo & Ferree, 2010, p. 146; Cole, 2009, p. 177; Hancock, 2007, p. 70; McCall, 2005, p. 1788) to measure whether the combination of two factors is more than a sum of its parts. McCall (2005) links intersectionality to an intracategorical approach that is situated at the middle of the continuum between anticategorical and intercategorical approaches (2005: 1773). On the one hand, intersectionality rejects categories by revealing the intracategorical complexities within them and on the other hand intersectionality embraces categories as strategic (McCall, 2005: 1773). Since the introduction of conjoint experiments (Hainmueller et al., 2014, 2015), it is increasingly common for researchers to randomize multiple attributes at the same time, such as ethnorace and gender (Atkeson & Hamel, 2020; Lemi & Brown, 2019). Although some researchers find that combining certain categories leads to more than a sum of its parts (Golebiowska, 2001; Philpot & Walton, 2007; Weaver, 2012), some researchers test for interaction effects but find no statistically significant results (Doherty et al., 2019, p. 1288; Kao & Benstead, 2021, pp. 16–17; Ono & Burden, 2019, p. 604).

The studies that report not finding any statistically significant results when using interaction effects in conjoint experiments, might only be the tip of the iceberg. There are three main reasons why null results might be the most common outcome, even though this remains relatively invisible in the literature. First, studying social phenomena with interaction effects requires high levels of statistical power. With limited sample sizes, interaction effects could more easily give false negatives. Second, in academia, journals favor publishing relatively large effect sizes over null effects, making publication bias a problem (Quintana, 2015, pp. 6–7). Publication bias hinders scholarly advancement because researchers do not know the full scope of research findings when formulating their own research questions. Third, conjoint experiments might cause cognitively taxed respondents to single out just one attribute to base their choices on using the identity that is most salient in their political context. That is why I now turn to the political contexts of France, Germany, and the Netherlands.

The way female Muslim politicians are situated varies significantly across the three countries France, Germany and the Netherlands, which is crucial for understanding voter evaluations. In all three countries, the presence of Islamophobic populist radical right rhetoric, widespread anti-Muslim stereotypes, and distinct nationalist ideologies—such as secularist nativism and civilizationism (Brubaker, 2017; Kešić & Duyvendak, 2019; Marzouki et al., 2016) — lead voters to assess politicians through a unitary lens (Petsko et al., 2022). Differences in electoral systems also matter: proportional representation in the Netherlands encourages more virulent femonationalist and homonationalist rhetoric compared to Germany and France, where PRRPs must appeal more to conservative Christian voters due to their electoral constraints. These systems also shape political inclusion strategies, with symbolic inclusion of secular Muslim-origin politicians common in Germany and the Netherlands, and more individualized distancing strategies in all three countries, a phenomenon I call broadstancing (van Oosten, 2024d; 2025d). Together, these contextual dynamics influence how voters perceive and evaluate female Muslim politicians.

Methods

Although candidate experiments date back to the eighties (Sapiro, 1982), they now more commonly take the form of conjoint experiments (Hainmueller et al., 2014, 2015) and in that form are currently booming in the field of political science (e.g., Kao & Benstead, 2021; Leeper et al., 2019; Martin & Blinder, 2020; Reeves & Smith, 2019). Just as in most other candidate experiments, I present the respondent with profiles of hypothetical politicians, whilst randomly varying the attributes religion, gender, and ethnorace. See Table 1 for the values of each attribute per country. After respondents have viewed the profiles, I ask them to evaluate each single politician and choose between pairs of two politicians. The possibility to study multiple group memberships at the same time with conjoint experiments enables an intersectional analysis, yet this possibility is still “underutilized” (Klar and Schmitt, 2021: 495).

In Europe, ethnoracially minoritized groups are most often referred to as having a “migration background” (Rosenberger & Stöckl, 2018; Verkuyten et al., 2016), possibly as a way to avoid references to ethnicity or race (Simon, 2017: 2328). We adopt the same operationalization because we expect our respondents to become confused if we present them profiles referring to a politician’s ethnorace, as this is not a way people are generally accustomed to discussing ethnoracial difference in Europe. In each country, I chose the most common migration backgrounds with populations that are most likely to experience discrimination (FRA: European Union Agency for Fundamental Rights, 2017). There are two exceptions: 1) politicians with a Turkish background in France, which I chose in order to have one common migration background in all three countries (Ersanilli and Saharso, 2011). 2) politicians with a background in the Former Soviet Union (FSU) in Germany, which I chose in order to have a case in which ethnicity is seen as white, while they have a migration background nonetheless (Goerres et al., 2018; Spies et al., 2020). I matched a common name to their ethnic/racial group and signaled gender through first name and pronoun she/he. An example of a profile is: “Sebnem Yılmaz has a Turkish background and she practices Islam”, followed by a randomized policy position on socio-economic and socio-cultural issues (see all exact survey questions here: van Oosten, 2025c).

I constructed the profiles in the third person to mimic a journalist describing them. Moreover, a politician explicitly stating migration background and religion in the first person could seem unconvincing to respondents. All combinations of ethnorace and religion are possible. Some combinations are less likely to be encountered in real life (such as a politician with an FSU background who practices Islam, a politician with a Turkish background who practices Christianity or a politician with a French background who practices Islam) but as there are Muslims in the FSU, Christians and Turkey, and Muslims without a migration background in France, none of the combinations are impossible so I kept them in anyway.

I presented 18,336 randomly constructed profiles to 3056 citizens of France (N=1199), Germany (N=954), and the Netherlands (N=903), administered by survey agency Kantar Public between March and August of 2020. All data is publicly available (van Oosten, 2025). While planning these online survey experiments I did not know I would be collecting data during the outbreak of a pandemic. Fortunately, replications of online surveys and experiments show that data collected in spring 2020 does not significantly differ from data collected before the outbreak of COVID-19 and subsequent lockdowns. If anything, outcomes are more conservative, making the chance of false negatives more likely than false positives due to some respondents being less attentive (Peyton et al., 2021).

Kantar Public has many policies in place to ensure high attentiveness in their panelists. Members of the Kantar panel only receive invitations to participate in their surveys once a month. In all three countries participants received 5 euros’ worth of points for their participation, with which they can buy small items in a gift shop. This means that although respondents are externally motivated to participate, the rewards are not high enough and the invitations are not frequent enough to create a following of online workers clicking through surveys in order to receive as many rewards as possible. Indeed, respondents gave lengthy answers in the comment box at the end of the survey, showing how involved respondents felt with the questions, revealing their internal motivation to give well-thought-through answers. Moreover, Kantar imposes policy to achieve what they call “panel-hygiene”. Kantar pays explicit attention to how fast people answer surveys and when they answer the surveys too quickly or with too many repetitive answers they get expelled from the panel and do not receive any more invitations. Kantar reaches out to communities underrepresented in their panel, to increase their panel being a reflection of the citizenry of each country.

In the survey, I presented a single politician profile and asked respondents to answer the following three questions (see all exact survey questions here: van Oosten, 2025c) on a scale from 0 to 10: Do you think this politician represents you? How much do you trust this politician? How capable do you think this politician is of performing well on the job? I then presented another politician profile and asked the same three questions. Then I asked respondents to choose between one of the two profiles by asking Which politician are you most likely to vote for? In my analysis I distinguished two dependent variables, evaluation and choice. I constructed the evaluation variable by adding up the scores of the questions on representation, trust, and capability (Chronbach’s Alpha: 0,89). I recoded both the evaluation and choice variables to range from 0 to 1.

I prepared the data using R-package “tidyr” (Wickham, 2020), ran analyses using “miceadds” (Robitzsch et al., 2021) and made visualizations with “ggplot2” (Wickham et al., 2020), (code is available here: van Oosten, 2025c). First, for the unitary analysis, I ran linear regression models with politician religion, gender and ethnorace as separate independent variables. Second, for the multiple analysis, I ran linear regression models with recoded dummy variables indicating whether a respondent saw a profile of a politician who was either a female Muslim, male Muslim, female Christian, male Christian or female non-religious politician, compared to male non-religious politicians. Third, for the intersectional analysis, I ran linear regression models while interacting religion and gender and controlling for the main effect. In all models, I 1) controlled for respondent age, age squared, and sex, 2) used population weights based on respondent level of education, migration background, region, urbanization, and gender and 3) clustered at the level of the respondent. I accepted hypothesis tests with a p-value of smaller than 0.05. Before gathering the data, I pre-registered (van Oosten, 2020) the hypothesis and methods at Open Science Framework (OSF) and all code, appendices and survey questions are available on OSF (van Oosten, 2025c).

Results

Do (intersections of) politician religion, gender, and ethnorace influence how voters assess them? I structure this section in the same way as I structured the theoretical framework: according to Hancock’s (2007) distinction between unitary, multiple and intersectional analyses. Figure 1 presents the results of the unitary analysis, Figure 2 of the multiple analysis and Figure 3 presents the results of the intersectional analysis. Although I find a consistently strong negative bias against Muslim politicians, Muslim women and Muslim men do not seem to be assessed differently from one another. I discuss the possible causes of these findings and implications for future research.

Unitary Analysis

First, I take a unitary approach to the research question. Here, I analyze religion, gender, and ethnorace unitarily, i.e. as separate categories. I ran linear models with each attribute value (Muslim, Christian, female, Turkish etc.) as an independent variable, excluding the reference categories (non-religious, male, without migration background). This is the most common method of analysis and presentation in studies using conjoint experiments (e.g., Dahl & Nyrup, 2021; Hainmueller et al., 2014, 2015).

The most important outcome of Figure 1 is very straightforward: voters have a strong negative bias against Muslim politicians compared to non-religious politicians. This outcome holds across all three countries and both dependent variables. For the rating-dependent variable, effect sizes range from 5.2 (France, Germany), to 5.6 percent (the Netherlands). Effect sizes are, as usual, larger in the choice-dependent variable[1] and range from 9.5 (France), to 11.0 percent (Germany) and 14.0 percent (the Netherlands). This effect is in contrast with previous research in the US and UK (Bai, 2021; Campbell & Cowley, 2014) yet similar to research in Denmark varying Arabic and Danish names (Dahl & Nyrup, 2021). In the Netherlands, the negative bias against Muslim politicians is most pronounced, which could be a consequence of there being more Islamophobic attitudes in the Netherlands than in France and Germany (European Commission, 2016; Heath & Richards, 2019, p. 25; Ribberink et al., 2017, pp. 264, 266), or populist politicians using Muslims as a scapegoat (Brubaker, 2017, p. 1193).

In comparing France, Germany, and the Netherlands some oddities stand out. First, Germany takes up an intermediate position in this analysis whereas research reveals less Islamophobia in Germany, compared to French and Dutch society (Brubaker, 2017, p. 1193). Second, on one of the two dependent variables in Germany and the Netherlands I also find a negative bias towards Christian politicians compared to non-religious politicians, whereas I do not find this in France. This is surprising given the French political history of secularism (Kuru, 2008) where one would be more inclined to expect that a politician who explicitly confesses practicing Christianity would be penalized by voters. In fact, in Figure 1 (France/choice), French voters are statistically significantly more negative against Muslim politicians than Christian politicians, whereas the confidence intervals overlap in Germany and the Netherlands. Indeed, in all other models the bias against Muslim politicians is strongly negative compared to non-religious politicians, but not statistically significantly more so than the bias against Christian politicians. Third, Figure 1 also shows that French voters have a negative bias towards Turkish politicians. This is surprising because the Turkish community is relatively small and very diverse in terms of migration histories in France, especially compared to Germany and the Netherlands (Ersanilli & Saharso, 2011). In the Netherlands, we also see voters have a negative bias against Turkish politicians, though barely missing the test of significance. Germany has the largest population of citizens with a Turkish background of the three countries and this is the only country where assessments of Turkish politicians are the closest to zero.

Figure 1 also shows that German voters are slightly more likely to choose a female politician than their male counterpart, though this does not pass the test of significance. This trend is not reflected in how German voters evaluate female politicians. In France, I find no negative bias against female and male politicians. In the Netherlands, however, I find a negative bias against female politicians both in the way they are evaluated and whether voters are likely to choose them.

Multiple Additive Analysis

I also take a multiple approach to the research question. For the purposes of parsimony, societal relevance and to fill gaps in the literature, I focus on religion and gender (particularly Muslim women) instead of any other combination of the three attributes I randomize in this study (ethnorace, religion, and gender). Here, I analyze religion and gender in an “additive” (Hancock, 2007, p. 70) or multiple sense. I ran linear regression models with recoded dummy variables indicating whether a respondent saw a profile of a politician who was either a female Muslim, male Muslim, female Christian, male Christian or female non-religious politician, compared to male non-religious politicians (the reference category). The “quick addition” (Hancock, 2007, p. 64) I use thus consists of adding up the levels of the gender and religion attributes respondents saw, not adding up effect sizes.

Indeed, Figure 2 shows that in France and the Netherlands, voters are statistically significantly more negative about male and female Muslim politicians than non-religious men. The analysis in which I flip the reference categories so non-religious men are the independent variable and all others are the reference categories, shows that in France and the Netherlands, voters are statistically significantly more likely to choose male non-religious politicians than any other combinations of gender and religion. These differences between France and the Netherlands on the one hand and Germany on the other are in line with the literature on Islamophobia in all three countries (Brubaker, 2017, p. 1193; European Commission, 2016; Heath & Richards, 2019, p. 25; Ribberink et al., 2017, pp. 264, 266).

In France and Germany, male Muslim politicians receive slightly more negative assessments than female Muslim politicians. In the Netherlands, the opposite is true. This points towards “double disadvantage” in the Netherlands and “strategic advantage” in France and Germany (Gershon & Lavariega Monforti, 2021). Although party selectors are more prone to choose Muslim women over Muslim men in proportional systems (Dancygier, 2017; Hughes, 2016), pointing towards “strategic advantage”, this suggests voters in France and Germany do the same, to a slight extent. In the Netherlands it seems that Muslim women face “double disadvantage” (Gershon & Lavariega Monforti, 2021), although only slightly compared to Muslim men.

 The major caveat of Figure 2 is that the confidence intervals overlap. This means that although some combinations of politician attributes might be statistically significantly different from the reference category, none of the categories with at least one subordinate element are statistically significantly different from each other. This means that these categories are not “more than the sum of mutually exclusive parts” (Hancock, 2007, p. 65). I could have used various methods for this multiple additive analysis. For instance, I could have also added up the effect size of being Muslim and that of being female. The approach I chose, however, approximates but fails to reach the intersectional analysis. Therefore, this approach allows me to be more clear about what I understand the distinction between a multiple and an intersectional analysis to be. In my approach to a multiple additive analysis, the independent variables are additive dummy variables of more than one attribute (religion and gender) whereas an intersectional analysis comprises interaction effects between attributes (religion interacts with gender) while controlling for the main effects of each separate direct effect (religion, gender). In other words, a multiple additive analysis adds up the sum of all parts, an intersectional analysis is more than the sum of its parts. In the following section I will articulate this further.

Intersectional Analysis

In the intersectional approach to the research question I analyze religion and gender intersectionally, i.e. as “mutually reinforcing” (Crenshaw, 1991, p. 1283) categories that comprise more than the sum of their parts. I continue to focus on the attributes of religion and gender. I ran linear models with two-way interaction effects controlling for the direct effect of each attribute.

Figure 3 does not reveal any statistically significant interaction effects. The direct (unitary) effect of being a Muslim politician is statistically significant in France (12.4 percent) and the Netherlands (12.1 percent), though not in Germany where the analysis does not reveal statistically significant interaction effects. This is surprising given 1) the centrality of gender equality mobilized to discredit Muslims in (femo)nationalist debates (Farris, 2017; Korteweg & Yurdakul, 2021; Rahbari, 2021; Yurdakul & Korteweg, 2021); 2) the literature stating that Islamophobia impacts Muslim women particularly (Choi et al., 2021; Weichselbaumer, 2020); and 3) the literature on Muslim women who strategically resist the combined impact of being female and Muslim (Ahmed, 2020; van Es, 2019; Zimmerman, 2015). One could argue that the intersection of Islam and gender would be a most likely case of finding statistically significant interaction effects and the null effects in this study are, therefore, telling.

Discussion

Despite multiple calls for using interaction effects to estimate intersectionality quantitatively (Bowleg, 2008, p. 319; Choo & Ferree, 2010, p. 146; Cole, 2009, p. 177; Hancock, 2007, p. 70; McCall, 2005, p. 1788), this analysis does not find any statistically significant effects. Does this mean intersectionality does not play a role in how voters assess politicians? Possibly, but not necessarily. In the following paragraphs, I will discuss four potential explanations for these null findings in ascending order of importance: 1) false negatives, 2) heterogeneous treatment effects, 3) elimination of campaigning effects and most importantly, 4) contextual factors causing unitary effects.

First, with small effect sizes, limited sample sizes and thus limited statistical power, interaction effects are more likely to give false negatives. The sample sizes in this study were quite large. In total, I presented 18,336 randomly constructed profiles to 3056 citizens, repeating the experiment six times for each respondent. I presented 7194 profiles to citizens of France (N=1199), 5724 in Germany (N=954) and 5418 in the Netherlands (N=903). The effect sizes for Muslim politicians are also much larger than what all similar experimental studies find for minoritized politicians (e.g., Bai, 2021; Campbell & Cowley, 2014; Carnes & Lupu, 2016; Hainmueller et al., 2014; Kirkland & Coppock, 2018; Krupnikov et al., 2016; Visalvanich, 2017; Weaver, 2012). The high sample and effect sizes reveal that statistical power is probably not causing false negatives in this particular study. Whatever the cause, I expect that similar experimental studies have produced null results more often, though due to publication bias the scope of this is largely unknown (Quintana, 2015, pp. 6–7).

Second, heterogeneous treatment effects could obscure findings. Possibly, some female Muslim politicians have a “strategic advantage” whereas others have a “double disadvantage” (Gershon & Lavariega Monforti, 2021). Together, these effects average out to null effects. In some electoral contexts, party elites prefer selecting Muslim women over Muslim men, whereas in other electoral contexts Muslim men are at an advantage (Dancygier, 2017; Hughes, 2016). Possibly, some voters advantage Muslim women, whereas others advantage Muslim men. One average, this results in null effects. Finding out what causes some voters to prefer one intersection over the other could explain future findings.

Third, I measure voter reactions at first glance without any effect of possible (hate) campaigns. Research on candidate gender tells us that stereotype reliance only happens when stereotypes have been activated during a campaign (Bauer, 2015) and counterstereotypic campaigning improves voter evaluations (Bauer, 2017). In addition, stereotypes of Muslims can be mobilized to influence voting behavior (Jardina & Stephens-Dougan, 2021). In campaigns and press coverage, minoritized women are more visible yet receive more negative coverage (Ward, 2016, 2017). I do not know of any research on which politician background characteristics tend to be foregrounded in campaigns. Quick experiments like the one at hand tend to eliminate the effects of (hate) campaigns, interactions with other actors and the time to develop a narrative surrounding a politician upon which a voter might be able to reflect on what their evaluation and subsequent choice might be. Beyond the effects of campaigns, other contextual factors can influence voter assessments causing one unitary background characteristic to be foregrounded. That is the fourth explanation for the null findings. I turn to this next.

Fourth, contextual factors prescribe whether people see others, who by definition inhibit complex intersectional identities, through either singular and simplistic lenses or intersectional and complex lenses (Petsko et al., 2022). One reason voters may rely on singular lenses is lens accessibility—the ease with which a particular social lens is retrieved from memory (idem). In France and the Netherlands, negative stereotypes about Muslims may be more accessible than in Germany, explaining clearer unitary effects in those contexts. Lens fit and perceiver goals further shape whether a singular lens is used: if a stereotype fits existing narratives (e.g., Muslims as threats to liberal democracies) and serves personal or national identity goals, voters are more likely to rely on it (idem). Moreover, Muslim politicians are distinctive because of their rarity, especially those who openly identify as practicing Muslims (idem). This distinctiveness may heighten reliance on a Muslim lens when evaluating them. Most importantly, conjoint experiments are cognitively demanding,  leading to information overload, a longstanding concern in conjoint experiments (Lines and Denstadli, 2004). When confronted with cognitively burdensome tasks, respondents are more likely to highlight just one identity-specific lens at a time (Petsko et al., 2022, p. 764), often the most distinctive or accessible one. This helps explain the dominance of singular effects in the experiment, even if intersectional perceptions play a greater role in real-world settings.

The most straightforward solution would be to study real voting behavior for intersectionally minoritized politicians (see Mikkelborg, 2025), though many political contexts do not have such data available. Another solution might be to survey voters and assess real-life politicians who have already been subjected to nationalist discussions and (hate) campaigns – which arguably are intersectional frames. Many national election surveys already survey voter assessments of party leaders or otherwise high-ranking politicians. As female Muslim politicians rarely lead parties in western countries and usually take up lower-ranking positions (Dancygier, 2017, p. 165; van der Zwan et al., 2019), this sort of data has yet to be gathered. A difficulty is that lower-ranking politicians are by definition lesser known to the general population and surveying voter assessments of such lesser-known politicians might, therefore, not make sense. Another solution is that future researchers could study the extent to which they publicly embrace or denounce their personal connection to the Muslim faith using either qualitative case studies (already done by van Kortewe and Yurdakul, 2021; Es, 2019; Vermeulen, 2018; Murray, 2016; Yurdakul and Korteweg, 2021) or Large Language Models (LLMs) to analyze parliamentary speeches and media performances and ascertain whether female Muslim politicians are subjected to significantly different frames than male Muslim politicians.

Conclusion

Do (intersections of) politician religion, gender, and ethnorace influence how voters assess them? Hancock’s (2007) distinction between unitary, multiple and intersectional analyses has been indispensable to structuring this paper. This study reveals voter have a strong negative and unitary bias against Muslim politicians. Whereas politician ethnorace and gender do not make a consistent difference in how voters assess politicians, religion (especially Islam) does. The intersectional analysis reveals no statistically significant interaction effects. Although we find some suggestions of both strategic advantage and double disadvantage for female Muslim politicians, the findings point towards unitary effects of a politician being Muslim on the assessments of voters.

I do not rule out the existence of intersectional effects in real life and though I continue to believe interaction effects are the most sensible method of analysis, I do not believe conjoint experiments are ideal for intersectional analyses. Conjoint experiments are cognitively taxing, increasing the likelihood that participants default to basing their choices on the most salient identity that is presented to them. When the Muslim identity is made salient and fits dominant societal narratives, such as Muslims being framed as a cultural threat, other identities fall out of focus. Thus, even when intersectional categories are present in the design, respondents are likely perceiving politicians through singular and simplistic lenses rather than complex, intersectional ones.

In real life, people often take in multiple aspects of a person’s identity at once—especially when it comes to Muslim women and men. Public perceptions, media coverage, and political rhetoric often reflect these intersectional dynamics. While this study does not uncover statistically significant interaction effects, that does not mean such effects are absent in real-world contexts. Rather, it highlights the need to rethink how we study intersectionality. Interaction effects remain the most appropriate analytic strategy for capturing these dynamics, but we must carefully consider whether conjoint experiments are the right methodological tool. To make meaningful progress, we should continue using interaction effects while also seeking research designs that are better equipped to detect the simultaneous influence of multiple identities.

[1] This is very common when comparing dependent variables measuring rating and choice, see Hainmueller et al. (2014).