Introduction
Across liberal democracies, the integration of ethnic and religious minorities has become a central challenge with implications for democratic resilience and social cohesion. Among these groups, Muslims in Europe represent both a significant demographic presence and a politically consequential community (Dancygier, Reference Dancygier2018). Muslims constitute roughly six percent of the population in major Western European countries,Footnote 1 and as their size and visibility grow, so does their electoral influence. Although many no longer neatly fit the category of “immigrant origin,” they remain a minority out-group that often encounters discrimination and exclusion. Anti-Muslim sentiment has risen in both the United States and Europe, driven by perceptions of Muslims as culturally distant or incompatible with liberal values (Verkuyten, Reference Verkuyten2021). These characterizations are also reflected in media coverage and political discourse and can lead to unequal treatment in employment, housing, and education (Abdelgadir and Fouka, Reference Abdelgadir and Fouka2020). A recent EU Agency for Fundamental Rights report shows that one in two Muslims in the EU experiences racism, harassment, or discrimination, with young Muslims born in Europe and women wearing religious attire especially vulnerable.Footnote 2
Such experiences matter politically. Perceived discrimination correlates with weaker national identification (De Vroome et al., Reference De Vroome, Verkuyten and Martinovic2014), lower trust in government (Maxwell, Reference Maxwell2010), greater dissatisfaction with democracy (Mansoury Babhoutak et al., Reference Mansoury Babhoutak, Kavadias and Echeverria Vicente2020), and in some cases even sympathy for radical alternatives (Mitts, Reference Mitts2019). Yet discrimination does not always lead to withdrawal. For many Muslims, exclusion increases political engagement, particularly toward parties that defend minority rights. Despite growing academic attention to Muslim integration, their voting behavior remains understudied. Although Muslims in Europe consistently support left-wing parties at higher rates than non-Muslims, existing explanations remain inconclusive, especially regarding the role of integration (Oshri and Itzkovitch-Malka, Reference Oshri and Itzkovitch-Malka2025). Figure 1(a) shows that across eighteen Western European countries, Muslims support left-wing parties by more than 30 percentage points relative to non-Muslims. In our survey of Muslim Turks in Germany, those with no identifiable immigrant background, what is often termed “third generation,” are even more left-leaning than first-and second-generation respondents (SI Appendix Section 3). From the perspective of classic theories of immigrant incorporation, this is surprising. One influential line of work expects that social and economic integration will weaken ethnic voting and align minorities more closely with the party system of the majority population (Alba and Nee, Reference Alba and Nee2003; Goerres et al., Reference Goerres, Mayer and Spies2022; Portes and Rumbaut, Reference Portes and Rumbaut2001). A different line of research on discrimination and group threat, however, suggests that systematic exclusion can heighten minority identification and sustain group-based political behavior, even (or especially) among the later generations who are otherwise well integrated (Fleischmann and Phalet, Reference Fleischmann and Phalet2016; Huddy, Reference Huddy, Huddy, Sears and Levy2013; Verkuyten and Yildiz, Reference Verkuyten and Yildiz2007). These patterns challenge conventional expectations that integration reduces group voting and highlight the need for new explanations (Goerres et al., Reference Goerres, Mayer and Spies2022). More broadly, they raise a concrete question: under what conditions does integration dilute Muslims’ distinct voting patterns, and when does discrimination sustain cohesive, left-leaning group voting instead?
Muslim–non-Muslim gap in the vote for leftwing parties in 18 European democracies 2002–2020. Each line represents a country in the sample. Parties included in the left party family are: socialist and social democratic (Source: Parlgove). (a) Trendline is a polynomial regression of the Muslim–non-Muslim gap on year, weighted by country. (b) In red line are country/years where the far-right attained at least a single parliamentary seat, and in black line are county/years where the far-right did not make it to parliament. Data are drawn from the European Social Survey 2002–2020.
Figure 1. Long description
The line graph illustrates the Muslim-non-Muslim gap in the vote for leftwing parties across 18 European democracies from 2002 to 2020. Each line represents a different country in the sample. The x-axis spans from the year 2002 to 2020, while the y-axis measures the percentage gap, ranging from 0 to 80. The graph is divided into two panels: (a) and (b). Panel (a) includes a trendline, which is a polynomial regression of the Muslim-non-Muslim gap on the year, weighted by country. Panel (b) highlights country-years where the far-right attained at least a single parliamentary seat in red lines and country-years where the far-right did not make it to parliament in black lines. The data are drawn from the European Social Survey 2002-2020. All values are approximated.
This article addresses this question by bringing together three strands of literature that have rarely been integrated: research on discrimination and threat, work on Muslim political participation in Western Europe, and theories of immigrant integration and ethnic voting. We argue that experiences and cues of social–political exclusion can sustain cohesive left-wing voting among Muslims by activating ethnic and religious identities and shared grievances.
To test our argument, we focus on the political behavior of Muslim Turks in Germany. We fielded an original survey among nearly 1,000 Muslim Turks that expands on existing measures of integration and discrimination and embeds a novel experiment that randomly exposes participants to videos depicting social exclusion (hate crimes) and political exclusion (far-right success). The findings reveal a consistent pattern. Over generations, Muslim Turks show clear signs of integration, higher language proficiency, more diverse social networks, greater economic stability, and increased political engagement. Yet the same integrated individuals also report stronger perceptions of discrimination and greater ethnic identification. Exposure to exclusion in the experiment evokes anger and fear and increases support for left-leaning parties, particularly among those with strong ethnic ties. These results speak directly to the competing expectations outlined above: contrary to the view that integration erodes group voting, our evidence shows that, in contexts of perceived and politicized exclusion, integration and discrimination can coexist and jointly sustain cohesive left-of-center voting.
Substantively, the paper provides one of the few studies that directly links experimentally primed social and political exclusion to Muslims’ vote choice, rather than to general attitudes, party identification, or turnout intentions. We discuss these findings in light of broader implications, particularly in relation to acute political conditions, hate crimes, and the rise of far-right parties.
Social–political exclusion and the muslim vote
Exclusion toward Muslim minorities takes multiple forms, including hate crimes (Frey, Reference Frey2020), labor market discrimination (Adida et al., Reference Adida, Laitin and Valfort2016), Islamophobic policies (Abdelgadir and Fouka, Reference Abdelgadir and Fouka2020), and the rise of far-right parties advocating restrictive agendas. These processes categorize and stigmatize individuals based on origin and religion. Such experiences might heighten the salience of group boundaries and strengthen ethnic identification and attachment (Huddy, Reference Huddy, Huddy, Sears and Levy2013; Kuo et al., Reference Kuo, Malhotra and Mo2017; Pérez, Reference Pérez2015). Individuals exposed to discrimination or violence are more likely to retreat into their ethnic group and interpret politics through an ethnic lens (Malik, Reference Malik2025; Pantoja and Segura, Reference Pantoja and Segura2003; Pérez, Reference Pérez2015; Zingher and Thomas, Reference Zingher and Thomas2012), with similar patterns observed among religious minorities (Bader, Reference Bader2007; Just et al., Reference Just, Sandovici and Listhaug2014; Wald et al., Reference Wald, Silverman and Fridy2005). Shared grievances can further motivate political engagement (Kranendonk et al., Reference Kranendonk, Vermeulen and Van Heelsum2018), as shown in studies of Muslim voters in Western Europe (Kollar and Vermeulen, Reference Kollar and Vermeulen2025; Spierings, Reference Spierings2016; Spierings and Vermeulen, Reference Spierings and Vermeulen2024).
However, evidence linking exclusion to Muslims’ vote choice remains limited and mixed. Some studies find increased support for left-wing parties (Sanders et al., Reference Sanders, Heath, Fisher and Sobolewska2014; Zingher and Thomas, Reference Zingher and Thomas2012), while others report no effects or even political disengagement (Fisher et al., Reference Fisher, Heath, Sanders and Sobolewska2015; Goerres et al., Reference Goerres, Mayer and Spies2022; Just, Reference Just2024; Simon et al., Reference Simon, Reichert and Grabow2013). Much of this research relies on observational data or focuses on party identification rather than vote choice (Just, Reference Just2024; Oshri and Itzkovitch-Malka, Reference Oshri and Itzkovitch-Malka2025), and experimental evidence on political exclusion remains scarce (Kuo et al., Reference Kuo, Malhotra and Mo2017).
Far-right parties as mobilizers of exclusion. Far-right parties are key drivers of anti-Muslim hostility in Europe (Mitts, Reference Mitts2019). Their platforms advance exclusionary, “nativist” populism that targets religious and ethnic minorities. These parties frame immigrants and minorities as threats, creating a moral divide between “good” citizens and “bad” outsiders (Oshri et al., Reference Oshri, Amsalem and Shenhav2024), and use stigmatizing rhetoric and imagery in campaigns. Their growing influence can also push mainstream parties toward harsher positions on immigration and Islam.
Research consistently links far-right support with anti-Muslim sentiment. Yet less is known about how far-right success shapes the political behavior of Muslims and immigrants. Mitts (Reference Mitts2019) finds that areas with strong far-right performance can foster Muslim extremism, while Sprague-Jones (Reference Sprague-Jones2011) shows that minorities in such contexts are more likely to support multiculturalism. Similarly, Martin (Reference Martin2023) finds that minorities in UK districts with higher British National Party support were more likely to vote Labour.
Returning to Figure 1, panel b disaggregates the Muslim–non-Muslim voting gap into two sets of country-years: those in which the far-right was represented in parliament (shown in red) and those in which it was not viable and therefore absent from parliament (shown in black). The figure reveals a larger Muslim–non-Muslim voting gap for the left in contexts where the far-right holds sway. In this sense, Figure 1(b) offers an aggregate indication for the relationship between political exclusion and the Muslim vote, highlighting how exclusion appears to push Muslims toward left-wing parties.
Building on this suggestive evidence, as well as the above literature, our article links far-right mobilization and the salience of anti-Muslim hate crimes to Muslims’ political behavior, and advances the literature by identifying these effects using a causal design: an experiment that randomizes exposure to far-right electoral success and hate-crime cues.
Expectations
While some research argues that over time and across generations, Muslims’ political behavior should converge with that of the non-Muslim electorate as a result of social and economic integration (Reeskens and Van Oorschot, Reference Reeskens and Van Oorschot2015; Schmidt-Catran and Careja, Reference Schmidt-Catran and Careja2017), the persistent Muslim–non-Muslim vote gap in support of left-wing parties challenges this expectation. If integration leads to assimilation in political preferences, we would expect Muslims’ voting patterns to resemble those of the majority population. Yet, across Europe, Muslims consistently show higher levels of support for left-wing parties. This suggests that factors beyond socioeconomic integration continue to shape Muslims’ political behavior. We argue that experiences of social and political exclusion sustain group-based voting among Muslims.Footnote 3 Exclusion can politicize ethnic and religious identification, making it more salient and thus more influential on electoral choices. In this context, continued support for left-wing parties reflects not a lack of integration but rather a strategic response to perceived marginalization, aligning with parties viewed as more protective of minority rights and anti-discrimination policies.
Data and measures
This study was conducted in Germany, Europe’s most populous country and home to the largest Muslim minority originating from a single country. To our knowledge, it is the first survey explicitly targeting Muslim voters in Germany aside from the Immigrant German Election Study, which also includes immigrants from the former Soviet Union (Oshri et al., Reference Oshri, Itzkovitch-Malka, Mor-Lan and Shenhav2026). Germany hosts approximately 5.5 million Muslims, most of whom are of Turkish origin, reflecting the migration of Turkish “Gastarbeiter” in the 1960s and 1970s. The naturalization of Turks and their descendants, especially after the 2000 citizenship reform, has made them one of the fastest-growing immigrant-origin electorates, increasing from about 750,000 eligible voters in 2017 to 992,000 in 2023 (Goerres et al., Reference Goerres, Mayer, Diaz Garcia and Elis2024).
In July 2023, we fielded an online survey of roughly 1,000 Muslim Turkish citizens using the Respondi panel (SI Appendix Sections 1 and 2). Because exclusion and hate crimes tend to target the largest and most visible minority group (Cikara et al., Reference Cikara, Fouka and Tabellini2022), Muslim Turks are a theoretically relevant population for examining reactions to exclusion. The survey received IRB approval, and participation required informed consent. The sample is nationally representative of Muslim Turks by age (18 and above) and gender, and includes only citizens eligible to vote in federal elections, omitting those without German citizenship.
A two-step sampling procedure was used. Recruitment first targeted individuals speaking Turkish or Arabic and those registered with citizenship from Muslim-majority countries. The sampling began in Brandenburg, Mecklenburg-Vorpommern, Sachsen, Sachsen-Anhalt, and Thüringen, targeting regions where far-right representatives won the 2021 federal elections. The goal was to achieve as many interviews as possible in these areas, as the premise was that Muslims residing in these areas will be exposed to exclusion, more so than Muslims in other regions. Second, invitations were then expanded nationwide. Because the survey company did not have full information on respondents’ citizenship or language, it invited the general adult population and screened respondents for eligibility. Those who met the criteria completed the survey in German.
Measuring integration. Following Harder et al. (Reference Harder, Figueroa, Gillum, Hangartner, Laitin and Hainmueller2018), we conceptualize integration as a multidimensional latent construct capturing psychological, economic, political, social, and linguistic aspects of life in Germany, rather than cultural assimilation. Respondents rated their sense of belonging, financial stability, political engagement, interactions with native Germans, and German-language proficiency (see Table A2 in SI Appendix for a full list of dimensions and questions). Responses were rescaled from 0 to 1 and combined into five integration indices. Importantly, our survey includes Muslims with different immigration backgrounds: Muslim immigrants, who have personally migrated to the host country (first generation); Muslims with immigrant background, whose parents (at least one) migrated to the host country (second generation); and Muslims with no identifiable immigrant background (third generation).
Measuring ethnic identification. To measure Muslim Turks’ ethnic identification, we adapted a 4-item partisan identity scale used by Bankert et al. (Reference Bankert, Huddy and Rosema2017) and Huddy et al. (Reference Huddy, Bankert and Davies2018), which taps “a subjective sense of group belonging, the affective importance of group membership, and the affective consequences of lowered group status, all of which are crucial ingredients of a social identity” (Huddy et al., Reference Huddy, Bankert and Davies2018). Respondents answered four items: “When I talk about Muslims, I usually say ‘us’ as opposed to ‘them’,” “I’m interested in what people think about Muslims,” “When people criticize Muslims, I take that as a personal insult,” and “I have a lot in common with other Muslims.” The items were strongly correlated (α = 0.76). An exploratory factor analysis revealed a single factor with eigenvalue greater than one (eigenvalue = 1.83 and 96% of the variance explained). We therefore created a scale ranging from 0 to 1, with higher values indicating stronger group identification.
Measuring subjective discrimination. Following Goerres et al. (Reference Goerres, Mayer and Spies2022),
we rely on group-based discrimination as a filter for individual experiences. Respondents were first asked whether they belonged to a group discriminated against in Germany (yes/no). Those answering yes identified the basis (ethnic origin, language, gender, disability, or religion) and then reported how often they experienced discrimination in five areas over the past five years (often, sometimes, rarely). We additionally asked how often respondents worry about violent crime and whether they had been a victim of a hate crime in the last 12 months.
Study design. We fielded a pre-registered online survey experiment with random assignment to two treatments or a control group.Footnote 4 Treatment 1 (N = 302) exposed respondents to a video on anti-Muslim hate crimes, while Treatment 2 (N = 304) showed a video on the electoral success of the far-right AfD and its anti-Muslim positions (SI Appendix Section 4). The control group (N = 377) viewed a neutral video. After exposure, respondents reported emotional reactions, political participation, and vote choice. Both treatments capture the same underlying mechanism – exclusion – through different channels: social (hate crimes) and political (far-right success). This design allows us to assess whether exclusion, regardless of its source, generates similar political responses.
Integration, discrimination, and ethnic identification
We begin our exploration by examining the question of integration. Figure 2 shows that Muslim Turks without identifiable immigrant background (born in Germany, whose parents were also born in Germany) are more integrated into German society across all integration indices compared to foreign born Turkish immigrants (first generation immigrants).
Integration scores by Muslims’ immigration background.
Note: Each integration dimension is a standardized index based on questions listed in Table A2 (SI Appendix Section 2). Higher values denote higher levels of integration. Figure A4 in the Appendix expands the figure to include also second-generation Muslims. Data are drawn from the survey conducted by the authors fielded in July 2023.
Figure 2. Long description
The image contains five density plots comparing integration scores by Muslims’ immigration background across different dimensions. The plots are titled Psychological integration, Language integration, Economic integration, Political integration, and Social integration. Each plot features two overlapping density curves representing first-generation and third-generation Muslims. The first-generation Muslims are shown in pink, while the third-generation Muslims are shown in blue. The x-axis of each plot ranges from 0 to 1, representing the integration scores, and the y-axis ranges from 0 to 3, indicating the density of the scores. The plots illustrate how integration scores vary between the two generations across different dimensions. In the Psychological integration plot, the first-generation Muslims show a higher density around the 0.6 mark, while the third-generation Muslims have a more spread-out distribution. In the Language integration plot, both generations show a peak around the 0.6 mark, but the first-generation Muslims have a higher density. The Economic integration plot shows a higher density for third-generation Muslims around the 0.4 mark, while first-generation Muslims have a more spread-out distribution. The Political integration plot indicates that first-generation Muslims have a higher density around the 0.2 mark, whereas third-generation Muslims show a more even distribution. The Social integration plot shows a higher density for first-generation Muslims around the 0.4 mark, with third-generation Muslims having a more spread-out distribution. All values are approximated.
At the same time, Muslim Turks without identifiable immigrant background report experiencing more discrimination than the first and second generations (Figure 3), and they also exhibit higher levels of ethnic identification compared to the first generation (Figure 4). This pattern is consistent with previous research on European Muslims, which shows that second-and later-generation Muslims often perceive more discrimination than first-generation immigrants and report lower trust in state institutions, despite higher levels of structural integration (Fleischmann and Phalet, Reference Fleischmann and Phalet2016; Giuliani et al., Reference Giuliani, Tagliabue and Regalia2018). It also resonates with work on second-generation Muslims in Europe documenting strong and salient ethnic and religious identities among European-born Muslims, sometimes more pronounced than among the first generation (Fleischmann and Phalet, Reference Fleischmann and Phalet2012; Verkuyten and Yildiz, Reference Verkuyten and Yildiz2007). Several mechanisms may underlie this pattern. First, higher levels of integration may increase exposure to discrimination: more integrated individuals are more likely to work in diverse professional environments and interact frequently with non-Muslims, where discriminatory experiences may occur. Second, higher integration may raise expectations of equal treatment and full belonging. Third-generation citizens who perceive themselves as fully part of society may experience discriminatory encounters as more anomalous and thus more aggravating. In contrast, first-generation immigrants may have fewer such interactions and may also be more inclined to anticipate or tolerate unequal treatment.
Feeling of discrimination by Muslims’ immigration background.
Note: Higher values denote higher (reported) discrimination.
Figure 3. Long description
The bar graph compares perceived discrimination among Muslims based on their immigration background across various categories. It features six sets of vertical bars, each representing different aspects of discrimination: feeling discriminated as a group, discrimination by police, discrimination in the internet, exposure to hate crime, discrimination by authorities, discrimination at work, and fear from hate crime. Each set includes three bars in different colors: pink for first generation, gray for second generation, and blue for third generation. The x-axis labels the categories of discrimination, while the y-axis measures the perceived discrimination levels. The data shows that third-generation Muslims report higher levels of discrimination in most categories compared to first- and second-generation Muslims. All values are approximated.
Ethnic identification by Muslims’ immigration background.
Note: Higher scores indicate stronger ethnic identification. The items are “When I talk about Muslims, I usually say ‘us’ as opposed to ‘them’,” “I’m interested in what people think about Muslims,” “When people criticize Muslims, I take that as a personal insult” and ”I have a lot in common with other Muslims.”
Figure 4. Long description
The bar graph compares ethnic identification by Muslims’ immigration background across four categories: Us/Them, Interested in Muslims, Criticize Muslims, and A Lot in Common with Muslims. The x-axis represents three groups: First Generation, Second Generation, and Third Generation. The y-axis measures values ranging from 2.5 to 4. Each group is represented by a different color: pink for First Generation, gray for Second Generation, and blue for Third Generation. The bars indicate that the Third Generation consistently shows higher values across all categories compared to the First and Second Generations. The First Generation has the lowest values in all categories. Error bars are present, indicating variability in the data. All values are approximated.
The descriptive results also speak directly to a broader debate in the literature about the relationship between perceived discrimination and ethnic identification. On the one hand, perceived discrimination can strengthen minority ethnic and religious identification and weaken national identification (Fleischmann and Phalet, Reference Fleischmann and Phalet2016; Giuliani et al., Reference Giuliani, Tagliabue and Regalia2018; Verkuyten and Yildiz, Reference Verkuyten and Yildiz2007). On the other hand, strong minority identification may heighten sensitivity to discrimination, producing a reciprocal feedback loop between identity salience and perceived exclusion (Verkuyten and Yildiz, Reference Verkuyten and Yildiz2007; Verkuyten, Reference Verkuyten2018). Our exploratory study is not designed to disentangle these dynamics; doing so would require longitudinal data, as in recent panel studies (Fleischmann and Phalet, Reference Fleischmann and Phalet2016). Instead, we examine how these forces jointly shape vote choice, to which we now turn in the experimental analysis.
Figure 5 reports the main effects of our two treatments, hate crimes and far-right success, on each of our dependent variables. The “intercept” row reports the baseline level of the control condition. As hypothesized, results show that exposure to the hate crime and far-right threat caused consistently higher levels of support for leftwing parties among respondents. While only 19.9 percent of the respondents in the control condition reported to vote in the next federal elections for the mainstream left party, SPD, 33.4 and 24.5 percent of respondents among the hate crime and far-right conditions, respectively, reported to support the SPD in the next elections. As expected, respondents assigned to the threatening conditions reported feeling more negative emotions (anger, fear, and hostility) than those assigned to the control condition.Footnote 5 The mean reported anger (on a scale between 0 and 1) increases from 0.28 in the control condition to 0.71 and 0.66 in the hate crime and far-right conditions, respectively. A similar picture emerges for the other negative feelings of fear and hostility. Negative or defensive emotions, such as anger, are known to propel political action and are therefore strong predictors of political participation (Groenendyk and Banks, Reference Groenendyk and Banks2014).
Average treatment effect of hate crime and far-right threat conditions on different outcomes.
Figure 5. Long description
Two line graphs compare the average treatment effect of hate crime and far-right threat conditions on various outcomes. The x-axis lists factors such as Anger, Hostility, Fear, Vote Left, Vote SPD, Participation, German Identity, Muslim Identity, and European Identity. The y-axis represents the Coefficient Estimate ranging from 0 to 1. Each graph features red lines with circles and error bars indicating the range of estimates for each factor. The hate crime graph shows higher estimates for Fear, Vote Left, and Vote SPD, while the radical right graph shows higher estimates for Fear, Vote Left, and Vote SPD as well. All values are approximated.
Next, we conducted heterogeneity analyses by immigrant generation, interacting the experimental condition with generation (DV: voting for left parties). Contrary to the expectation of stronger treatment effects among third-generation immigrants, we do not find statistically meaningful interaction effects. That is, the experimental effect of exclusion/discrimination on voting left does not significantly differ across generations (SI Appendix Section 4). This additional analysis points to a complementary and theoretically relevant pattern that is better characterized as mediation rather than moderation. First, recall that our descriptive results show that third-generation immigrants report higher levels of perceived discrimination compared to first-and second-generation immigrants. Second, the experimental results demonstrate that experiences of discrimination causally increase support for left parties across respondents. Third, in the observational analyses reported in the Supplementary Information, we find that third-generation immigrants are more likely to vote for left parties than earlier generations. Taken together, these findings suggest that exclusion does not necessarily operate differently across generations in terms of treatment responsiveness, but rather that third-generation immigrants are more exposed to or more likely to perceive exclusion, which in turn is associated with greater support for left parties. In this sense, the results align with the literature on integration, while adding an important nuance: exclusion appears to be particularly salient among those who are otherwise more integrated, even if its causal effect on political preferences is similar across generations.
In addition to the primary analyses, we explore the association between negative emotions and the strength of ethnic identification. Although we do not derive a specific hypothesis regarding this relationship, we consider it theoretically relevant and examine it in an exploratory manner. We view this analysis as preliminary, pointing to a potentially fruitful direction for future research. We therefore interact the experimental conditions with individuals’ saliency of exclusive Muslim identity to predict negative feelings aroused by the treatment. This analysis shows that, as articulated by Huddy et al. (Reference Huddy, Bankert and Davies2018, p. 191), “defensive group emotions are felt most intensely by the strongest group identifiers.” Indeed, Figure 6, which presents the predicted values for the three emotional reactions at different levels of ethnic identification, shows that Muslims with strong emotional and psychological attachment to their ethnic group, most of whom are third generation (see Figure 4), display defensive emotions when they encounter information that compromises their status. Such information will be taken as implicating their in-group and, therefore, as a call to rally in its defense. For example, after watching the hate crime video, respondents with the lowest ethnic identification score reported a medium level of anger: 0.53 on the scale of 0–1, while those with a high ethnic identification score reported a rather high level of anger: 0.78. In contrast, for the control condition, the ethnic identification score of the respondents did not predict anger at all. Similar results were obtained for fear and hostility.
The effect of experimental threatening hate crime (left-hand column) and far-right conditions (right-hand column) on anger, hostility, and fear across different levels of ethnic identity.
Figure 6. Long description
The line graph presents data on the effect of experimental conditions on anger, hostility, and fear across different levels of ethnic identity. The x-axis represents exclusive Muslim identity, ranging from 0 to 1. The y-axis represents predicted anger, hostility, and fear, ranging from 0 to 1. The graph is divided into two columns: the left-hand column shows the effect of the threatening hate crime condition, and the right-hand column shows the effect of the threatening far-right condition. Each row represents a different emotion: anger, hostility, and fear. The control condition is represented by a solid gray line, the threatening hate crime condition by a dashed black line, and the threatening far-right condition by a dashed red line. The bars with error lines indicate the predicted values and their confidence intervals. All values are approximated.
Adding up the results thus far reveals a striking pattern. Muslims with no identifiable immigration background, that is third-generation respondents, are the most integrated into German society across multiple dimensions. Yet they are also the most supportive of left-wing parties, thereby exemplifying cohesive group voting (SI Appendix Section 3). At the same time, they report the strongest sense of Muslim identification and the highest levels of perceived discrimination. The experimental findings further clarify this dynamic. Exposure to exclusionary cues increases support for left parties across respondents. In addition, individuals who identify more strongly as Muslims exhibit the largest increases in anger, hostility, and fear when confronted with messages signaling social or political rejection. Because third-generation respondents tend to score higher on Muslim identification, they are disproportionately represented among these high identifiers.
Figure 7 examines vote switching by comparing respondents’ reported vote in the 2021 federal elections (pre-treatment) with their intended vote in the next election (post-treatment). Panels a and b show the hate-crime and far-right treatments, respectively, while Figure 7(c) presents the control group. Two patterns emerge. First, exposure to both treatments reduces the share of respondents who are undecided or do not intend to vote, indicating lower political apathy. Second, it increases support for left-wing parties, especially the SPD. This shift appears to reflect the mobilization of previously disengaged voters rather than switching across parties. As cell sizes are small when disaggregated, these results should be interpreted with caution.
Vote switching between the 2021 federal election and intended vote in the next election, by condition: (a) hate-crime cue, (b) far-right success cue, (c) control. Parties are color-coded according to their commonly recognized party colors in Germany. Parties considered left are as follows: SPD, Green, and Linke.
Figure 7. Long description
The image contains three graphs labeled (a), (b), and (c), each showing vote switching between the 2021 federal election and intended vote in the next election under different conditions. The graphs are color-coded to represent different political parties in Germany. The x-axis represents the parties before treatment, and the y-axis represents the parties after treatment. In graph (a), the condition is a hate-crime cue, in graph (b), the condition is a far-right success cue, and in graph (c), the condition is a control. Each graph shows the flow of votes from one party to another, with lines connecting the before and after treatment votes. The parties are color-coded as follows: AfD in light blue, CDU in black, DIE LINKE in purple, Do not know in orange, FDP in yellow, GREEN in green, and SPD in red. The graphs illustrate how different conditions affect vote switching among these parties. All values are approximated.
Indications of mechanisms. The previous section showed that exposure to exclusionary messages increases Muslims’ support for left-leaning parties. To shed light on the mechanisms behind this shift, we conduct an exploratory (non-preregistered) analysis of responses to an open-ended question asked at the end of the survey: “What does Germany mean to you?” All respondents, including those in the control group, answered this question, which allows us to examine how exposure to exclusion shapes the emotional and cognitive content through which respondents describe Germany, and how these meaning-making processes relate to political preferences.
Our text analysis proceeds in two stages. First, we classify the overall sentiment expressed toward Germany in each response using GPT-4o via API. Second, we analyze which words and short phrases are most strongly associated with positive versus negative sentiment, and whether these associations differ between treated and control respondents. Following Gidron et al. (Reference Gidron, Sheffer and Mor2022), we estimate an ElasticNet regression model with separate word-level coefficients for the treatment and control groups. The dependent variable in each model is the sentiment score of the response, and the independent variables are word and bigram features derived from the text. Figure 8 visualizes the words and phrases most strongly associated with sentiment in the treatment group. Specifically, it displays terms that receive nonzero coefficients in the ElasticNet model estimated on treated respondents, with positive coefficients indicating association with positive sentiment and negative coefficients indicating association with negative sentiment. To focus on vocabulary that is distinctive of the treatment condition rather than general language patterns, we display only terms that are diagnostic in the treatment model but not in the control model. This does not imply that these words never appear in control responses, but rather that they do not reliably predict sentiment among control respondents once regularization is applied.
Unique positive (blue bars) and negative (orange bars) words for treatment groups.
Figure 8. Long description
The bar graph compares unique positive and negative words for treatment groups. The x-axis lists words, while the y-axis indicates the frequency of these words. There are two sets of horizontal bars: blue bars representing positive words and orange bars representing negative words. Positive words include love, best, better, world, luxury, peace, hope, good, opportunities, tolerance, ideas, mother, although, justice, despite, economic, better, opportunity, home, most beautiful, freedoms, life, friends, cohesion, well-being, pleased, joy, freedom land, great, wonderful, way of life, security, and beautiful. Negative words include migrants, pushing away, marginalized, treating, stepmotherly, emotionally, controlling, bad, mood, ashamed, frustrated, government, hatred, contempt, stress, poverty, gives, however, care, racism, own, unjust, more lost, outdated, garbage, catastrophe, chaos, people, sad, bankruptcy, high, monotonous, dependence, and origin. All values are approximated.
The results validate the experimental manipulation and clarify the mechanisms behind the observed political shift. Among treated respondents, negative sentiment is strongly associated with terms such as AfD, racism, hatred, and marginalized, indicating that exposure to exclusionary cues makes discrimination, political hostility, and social rejection cognitively salient. At the same time, treated respondents also employ a distinct set of words associated with a positive sentiment, including peace, freedom, tolerance, opportunity, justice, and well-being. These terms are not generic expressions of positivity, but aspirational references that respondents invoke precisely in response to perceived threat and exclusion. This pattern is evident in respondents’ own language. For example, one treated respondent wrote:
“For me, Germany is a country where I can live freely. I can choose how traditional my lifestyle is. Here, I have social security should I become unemployed, and medical care is guaranteed (even if it’s steadily declining). Germany was once a country of freedom of speech and democracy. Unfortunately, it’s deteriorating more and more.”
Another respondent similarly juxtaposed security and decline:
A country that offers certain advantages and is generally safe compared to other countries. However, a country that has been moving in the wrong direction in many respects in recent years (increased shift to the right, deteriorating education and healthcare systems).
Others described Germany as “formerly a safe country” with “good opportunities to build a good life, but now the trend is more negative.”
Across responses, Germany is framed through a problem–remedy logic: negative experiences or perceived decline – often linked to the rise of the far right – are paired with demands for protection, inclusion, and social guarantees. This explains the prominence of aspirational terms such as equality, justice, and welfare, which reflect expectations of what Germany should provide under threat.
More broadly, exclusion appears to trigger a meaning-making process that politicizes group identity. Rather than disengaging, respondents express grievances while emphasizing values associated with left-leaning politics, especially cultural inclusion and economic protection. This suggests a selective shift toward parties perceived as offering institutional protection against discrimination and insecurity.
Conclusion
The study highlights the significant political impact of social and political exclusion on Muslim communities in Germany, Europe’s largest Muslim population. By analyzing data from an original comprehensive survey and experimental videos, the research demonstrates that experiences of exclusion propel Muslims toward cohesive group voting and favoring left-of-center political parties. It specifically underscores how hate crimes and the rise of far-right parties shape electoral behavior among Europe’s fastest-growing ethnic group, suggesting that exclusion not only increases political engagement but also strengthens group solidarity among integrated Muslims.
Empirically, relatively little research has directly examined how exclusion, particularly far-right mobilization, shapes minorities’ vote choice. Existing studies rarely manipulate exclusion experimentally and typically rely on observational designs, offering limited leverage on causal mechanisms. By focusing on a single, well-defined minority community and employing an experimental design that addresses both social and political exclusion, this study advances the literature by providing causal evidence on how exclusion translates into electoral behavior.
On a macro-level, this study can teach us about the electoral prospects of the left in Europe and about its relationship with the far-right, possibly revealing a somewhat “toxic” symbiosis between the two. According to the rationale developed in our project, in the case of Muslim citizens, it is the far-right that keeps the left politically viable: When far-right parties garner more support, Muslims seek refuge in the arms of the political left.
At the same time, important open questions remain. Our analysis focuses on Turkish Muslims in Germany, a relatively integrated minority community. Whether similar dynamics emerge among newer immigrant groups, other religious minorities, or in political contexts with different party systems and histories of exclusion remains an open empirical question. Future research could, for example, examine how the effects of social and political exclusion vary across minority groups, stages of integration, and institutional settings, and whether exclusion mobilizes participation in similar ways beyond electoral behavior.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/XPS.2026.10034.
Data availability statement
Support for this research was provided by the Fritz Thyssen Foundation (grant Ref. 20.17.0.047PO). The data, code, and any additional materials required to replicate all analyses in this article are available at the Journal of Experimental Political Science Dataverse within the Harvard Dataverse Network at https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/17KERB.
Acknowledgements
In addition to the anonymous reviewers and the editor, we thank the members of the Berlin Workshop “Political Parties in Contemporary Democracy” (Berlin, 2024), and the members of the “Actors Without Arena? Bringing the Political Behavior of Youths and Immigrants Into the Mainstream” workshop (Berlin, 2022) for helpful comments on this project. We are also grateful to Julia Wittorf for her superb research assistance. We gratefully acknowledge funding support from the Fritz Thyssen Foundation (grant Ref. 20.17.0.047PO).
Competing interests
The authors are not aware of any conflicts of interest.
Ethics statement
This study received ethical approval from the Research Ethics Committee at The Hebrew University of Jerusalem (No. 2023-02061). The research adheres to APSA’s Principles and Guidance for Human Subjects Research.
Open access
