Ethnic diversity and extremist attitudes in soccer teams: A multi-level investigation

It has been argued that sports contexts may be suitable venues for reducing intercultural hostility, including its more extreme forms, yet empirical evidence remains scarce. The present study investigated the main and interactive influence of two sets of factors on support for violent extremism: (a) ethnic diversity of teams, that is, the heterogeneity of the team members’ ethnic origins, and (b) team members’ intercultural perceptions, including perceived social capital, contact experiences and diversity ideologies within the team. Individual-level data from 257 players nested within 36 German soccer teams were combined with assessments of the ethnic diversity of each team based on a genealogical database. Multi-level analyses were conducted. Some evidence suggested that higher ethnic diversity within a team and frequent inter-ethnic contact between its players were associated with more extremism. However, cross-level moderation analyses showed that ethnic diversity was associated with less support for violent extremist groups when inter-ethnic contact quality was high. Perceptions of colorblind team ideologies that focus on minimizing/ignoring differences between groups were associated with lower threat perceptions and extremism. While social capital generally played little of a role, one social capital indicator, norms of behavior, was unexpectedly associated with higher threat perceptions. Overall, the present findings suggest that increasing ethnic diversity in sports teams may in itself not reduce extremist attitudes and sometimes may even backfire. Rather, how intercultural relations are managed within these contexts seems decisive. Prioritizing venues for positive contact experiences between soccer players of different backgrounds seems essential. © 2020 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Since the beginning of the refugee "crisis" in 2015, Germany, similar to other countries, has experienced a new increase in right-wing and Islamist extremism (Koehler, 2018). In 2016, 23,100 people endorsed a right-wing-extremist ideology in Germany, of which 12,100 people were considered violent extremists (Bundesamt für Verfassungsschutz [BfV], 2018). Simultaneously, the number of Islamist extremists grew to 24,400, many of whom are willing to resort to violence (BfV, 2017). Both right-wing and Islamist extremists tend to be predominantly male and in their late teens or early adulthood (Bakker, 2006;BKA et al., 2016;Kailitz, 2013). Furthermore, in Germany, men make up 92 % of perpetrators involved in extreme right-wing attacks (Götschenberg, 2018) and 79 % of the Islamist extremists who traveled to Syria (BKA et al., 2016).
While amateur soccer teams have increasingly been targeted by extremist groups to recruit young people (Atran, 2011;Dembowski, 2007;Dorsey, 2014Dorsey, , 2016Peucker, 2009;Pilz, 2008), soccer has also received particular attention for its potential to combat extremism (Glaser & Elverich, 2008;Kaddor et al., 2017;Peucker, 2009). Indeed, soccer-based extremism interventions have been implemented worldwide (Kosseim, 2011; United Nations Office on Drugs and Crime [UNODC], 2018). Yet, empirical insights into the group, intergroup, and intercultural factors at play remain scarce. Recently, social capital has also been suggested to play an important role in sports teams (Forsell, Tower, & Polman, 2018), as well as in the radicalization processes leading to extremism (Ellis & Abdi, 2017;Harpviken, 2019), and is therefore of high relevance for the present research. Moreover, social capital is closely linked to intercultural contact and should thus be studied in research that focuses on it (Schmid, Hewstone, Hughes, Jenkins, & Cairns, 2009). Hence, in the present study, we aimed to investigate the unique role of social capital alongside that of ethnic diversity, intercultural ideologies, and interethnic contact for extremist attitudes among soccer players, because these concepts play major roles in explaining attitudes within relevant group and intercultural settings. We define violent extremism as "the beliefs and actions of people who support or use violence to achieve ideological, religious or political goals" (United Nations Educational, Scientific and Cultural Organization, 2017, p. 19). In addition to focusing on support for such violent extremism, we treat experiences of symbolic and realistic threats as dependent variables (Stephan & Stephan, 2000;Stephen, Ybarra, & Morrison, 2009) because both are central factors influencing violent extremist beliefs (Obaidi et al., 2019;Stephens et al., 2019;Tahir, Kunst, & Sam, 2019), whilst being less likely to be impacted by socially desirable response tendencies.

The effect of ethnic diversity
One central but not yet fully understood factor in research on extremism is ethnic diversity. Soccer teams can be regarded as major social platforms to interact with others of diverse cultural backgrounds (Peucker, 2009). For instance, in Germany -the context of the present research -most soccer teams are composed of various ethnicities (Pilz, 2006). Previous studies have provided inconsistent findings regarding the effect of ethnic diversity on intergroup and intercultural attitudes. Some have demonstrated that high ethnic diversity at the macro level (e.g., neighborhood or country) can be associated with more intergroup threat (Quilliam, 1995;Semyonov, Raijman, Tov, & Schmidt, 2004). Higher ethnic diversity or the perception of it has furthermore been associated with greater racial bias (Craig & Richeson, 2014), more restrictive attitudes towards immigration (Alba, Rumbaut, & Martoz, 2005), and lower levels of generalized trust (Alesina & La Ferrara, 2002) -factors associated with higher levels of perceived threat and willingness to engage in violent extremism (Dalgaard-Nielsen, 2010;Obaidi et al., 2019). Moreover, in the Netherlands (Dinas & Van Spanje, 2011), Belgium (Rink, Phalet, & Swyngedouw, 2008), and Sweden (Valdez, 2014), a higher percentage of immigrants within a municipality was associated with increases in voting for right-wing parties.
By contrast, other studies have found no effect (Hjerm, 2007;Hooghe, Reeskens, Stolle, & Trappers, 2009), mixed effects (Oliver & Wong, 2003;Schlueter & Scheepers, 2010;Van Assche, Roets, Dhont, & Van Hiel, 2014), or beneficial effects (Ellison & Powers, 1994;Laurence, 2009;Schmid, Ramiah, & Hewstone, 2014) of macro-level ethnic diversity on intercultural relations. For instance, Oliver and Wong (2003) found that ethnic diversity was generally associated with less outgroup prejudice in neighborhoods. Yet, it was associated with more intergroup hostility in diverse metropolitan areas, where economic competition tends to be high. Furthermore, Van Assche et al. (2014) showed that ethnic diversity was associated with more positive general attitudes towards outgroup members (i.e., immigrants) among participants scoring low on authoritarianism yet with more negative attitudes among those scoring high on authoritarianism. This suggests that investigating ethnic diversity in itself might not be sufficient to understand its effects on intergroup outcomes but that an assessment of individual-level moderators is needed. One goal of the present research was thus to investigate whether the effect of team-level ethnic diversity on extremism is moderated by the attitudes and perceptions of the teams' players.

The role of social capital
As mentioned earlier, another central factor, which is often investigated together with ethnic diversity and associated with vulnerability to extremism, is social capital (Ellis et al., 2016;Ellis & Abdi, 2017). Social Capital Theory (Bourdieu 1986;Coleman, 1988;Putnam, 2000) encompasses access to resources (Bourdieu, 1986), the ability to use reciprocal relations to achieve individual goals (Coleman, 1988), and collaboration with others at the group level to pursue common goals, develop norms of reciprocity, and trust (Putnam, 2000). Whereas bonding capital describes the interaction of individuals that share many core elements, bridging social capital refers to the synergy of individuals who differ substantially from each other (Putnam, 2000).
In extremism research, a lack of having strong, positive connections with people who differ profoundly from oneself (i.e., low bridging social capital) has been found to increase vulnerability to developing extremist mindsets (Ellis et al., 2016;Ellis & Abdi, 2017). By contrast, developing high social capital through, for instance, trusting reciprocal relationships has been associated with decreased involvement in right-wing extremism (Boehnke, Hagan, & Merkens, 1998).
Much attention has also been paid to the ability of sports to create social capital (Coalter, 2007;Forsell et al., 2018;Nicholson, Brown, & Hoye, 2013), making it highly relevant for the present study. Soccer teams' potential effectiveness to reduce perceptions of intergroup threat and prevent extremist radicalization may lie in their capacity to create a sense of belonging and connectedness, two important aspects of social capital theory (Ellis & Abdi, 2017).
In the present research, we therefore examined whether players who experienced higher social capital in their teams would show fewer extremist tendencies and threat perceptions.

The importance of contact experiences
More recently, Schmid and colleagues (2009) argued that ethnic diversity and social capital should be examined jointly with contact experiences as they are inter-related factors. Therefore, intergroup contact was examined as another independent variable in the present research. Several studies have demonstrated that, rather than mere diversity, the frequency and perceived quality of interethnic contact is often decisive for positive intercultural relations (Barlow et al., 2012;Binder et al., 2009;Islam & Hewstone, 1993). Intergroup Contact Theory (Allport, 1954;Pettigrew, 1998) suggests that intercultural attitudes may improve in more diverse environments (e.g., ethnically diverse soccer teams) because more opportunities for interethnic contact arise (Schlueter & Scheepers, 2010). These claims have continuously been supported by research showing significant negative associations between interethnic contact and prejudice (Pettigrew & Tropp, 2006; but see Paluck, Green, & Green, 2018, for a recent critique). Several studies have further highlighted that contact quality and frequency may have different effects on intercultural relations (Islam and Hewstone, 1993;Johnston & Glasford, 2018;Van Dick et al., 2004). For instance, Binder and colleagues (2009) found that, when contact quality was controlled for, contact frequency no longer predicted intergroup anxiety. Moreover, demonstrating how both factors can have interactive effects, Barlow et al. (2012) showed that more frequent contact was associated with more prejudice when its quality was negative but with less prejudice when its quality was positive. Furthermore, Christ et al. (2014) demonstrated that living in a macro system (i.e., district, neighborhood), in which people interacted more frequently with outgroups, was also associated with decreased prejudice.
Research directly assessing effects of interethnic contact on extremism remains scarce. Fiske (2002) suggested that people who become extremists typically show few strong outgroup ties (e.g., intercultural friendships). Rydgren (2008), on the other hand, found that contact quality had different effects in the context of right-wing extremism depending on the country of investigation. A lack of immigrant friends was associated with a higher likelihood to vote for relatively extreme right-wing parties in Norway and Belgium, but not in Austria, Denmark, France, or the Netherlands 1 . Given these inconsistent findings and limited research, one goal of the present study was to investigate how contact quality and frequency relate to extremist attitudes among soccer players.

The role of intercultural ideologies
To further complement our holistic approach, we included perceived diversity ideologies because strong evidence suggests that they play an essential role in explaining intercultural relations, including extremism (Guimond et al., 2013;Guimond, da la Sablonnière, & Nugier, 2014). Such ideologies are concerned with how diversity is approached and managed. Two diversity ideologies that have received much attention are multiculturalism and color-blindness. A multicultural ideology appreciates cultural diversity and values group differences (Berry, 1997;Gündemir, Martin, &, Homan, 2019;Rattan & Ambady, 2013). By contrast, a colorblind ideology seeks to minimize and ignore group differences, assuming that people are similar and should thus be treated equally regardless of their ethnicity (Guimond et al., 2014;Purdie-Vaughns, Steele, Davies, Ditlmann, & Crosby, 2008). Importantly, multiculturalism has been found to foster more positive intergroup attitudes (e.g., more support for migration policies, lower perceived outgroup threat) by strengthening self-esteem among both ethnic minority and majority members and alleviating identity threat, whereas color-blindness is thought to yield more negative outcomes (e.g., greater racial bias, opposition against affirmative action policies) as it provides a justification for racial inequalities and disregards disadvantaged group members' minority identities (González, Verkuyten, Weesie, & Poppe, 2008;Neville et al., 2013;Richeson & Nussbaum, 2004;Sasaki & Vorauer, 2013;Verkuyten, 2009;Ward & Masgoret, 2008).
The effects of color-blindness and multiculturalism may furthermore be moderated by whether people belong to majority or minority groups (Jansen, Vos, Otten, Podsiadlowski, & van der Zee, 2016). Multiculturalism tends to be perceived more negatively by ethnic majority members as they might feel threatened in their cultural dominance (Verkuyten, 2004), whereas it is perceived more positively by ethnic minority members due to the appreciation of diversity (Verkuyten, 2007;Verkuyten & Martinovic, 2006). Research suggests the opposite effect for color-blindness. Whereas ethnic majority members tend to prefer color-blindness (Jansen et al., 2016), minority members perceive higher outgroup threat when interactions with the outgroup are accompanied by a colorblind ideology (Purdie-Vaughns et al., 2008;Rattan & Ambady, 2013). Yet, sometimes color-blindness can have cascading effects among minority group members. Based on the common ingroup identity model (Gaertner, Mann, Murrell, & Dovidio, 1989), Dovidio et al. (2016) proposed that color-blindness may result in lower inequality perceptions and subsequently more positive intergroup attitudes among minority members (also see Saguy, Tausch, Dovidio, & Pratto, 2009). Based on these mixed findings and predictions, we investigated how perceptions of color-blindness and multiculturalism in soccer teams would relate to extremism support and threat perceptions, and whether these effects would be moderated by whether players belong to ethnic majority or minority groups.
In addition to the perceived team ideologies, it is possible that the ideologies held by the coach are particularly influential because coaches typically hold more power than players. Meeussen and colleagues (2014) demonstrated that when leaders endorsed a multicultural (vs. colorblind) ideology, minority members felt more accepted. If leaders, however, endorsed a colorblind ideology, minority members experienced higher relationship conflict within and distanced themselves more from their workgroup (Meeussen, Otten, & Phalet, 2014). Given these findings, we investigated the independent effects of intercultural ideologies perceived to be prevalent in the team and to be held by the coach(es) separately.

Interactions between team-and individual-level variables
Importantly, while macro-level ethnic diversity and individual-level contact and intercultural ideologies may have main effects on extremism and threat perceptions, cross-level interactions are theoretically possible (De Keersmaecker et al., 2017;Fasel, Green, & Sarrasin, 2013;Magnusson & Magnusson, 2013;Mischel, 2004;Van Assche, 2018) and have been touched upon by previous research (e.g., Van Assche et al., 2014). Sniderman and colleauges (2004) proposed that contexts may increase (i.e., galvanize) an individual-level reaction for those who already endorse a certain attitude, or elicit (i.e., mobilize) a new attitude among people who originally had not endorsed it. Numerous studies have assessed this theory within the context of ethnic diversity, showing primarily galvanizing effects (Karreth, Singh, & Stojek, 2015;Van Assche, 2018). For instance, Van Assche and colleagues (2016) found that, when ethnic diversity was high, intergroup attitudes only became more negative for people who already held negative intercultural ideologies. Roccas and Amit (2011) demonstrated that when participants were manipulated to perceive more ethnic diversity, solely people who already endorsed a conservative ideology displayed more anti-immigrant attitudes, whereas pro-immigrant attitudes increased for more liberal participants. Hence, how ethnic diversity affects intercultural relations may depend on individual-level factors within the specific context. Following this rationale, we examined whether soccer teams' ethnic diversity has distinct effects depending on the intercultural ideologies players perceive and the frequency and quality of contact they experience.

The present research
Given that previous research has shown that social capital, contact experiences, as well as diversity ideologies, complementarily explain the effect of ethnic diversity on intergroup outcomes (Schmid et al., 2009;Van Assche et al., 2016), and all play a potential role in explaining extremism, investigating one without the other would provide an incomplete picture. To gather comprehensive insights, a nuanced focus on the role of these variables furthermore requires the assessment of both individual-level and cross-level interactions. In terms of individual-level interactions, we examined whether the associations of interethnic contact within a team with support for extremism would be moderated by the quality of this contact, and whether the associations of perceived multiculturalism and color-blindness with support for violent extremism would differ depending on whether players belonged to the ethnic majority or minority group (Figure 1 ). In terms of cross-level interactions, we examined whether the association of ethnic diversity (at the team level) with extremism and threat perceptions would be moderated by the intercultural ideology players perceived to prevail in their team (i.e., multiculturalism or color-blindness) and their experiences of interethnic contact (i.e., frequency and quality; Figure 1).
To investigate these relationships, we collected data from ethnic majority-and minority group members playing in 36 soccer teams, located in the federal state of North-Rhine Westphalia, Germany. This area was targeted due to its high ethnic diversity (25.8 % minority members in 2016; Ministerium für Kinder, Familie, Flüchtlinge und Integration des Landes Nordrhein-Westfalen, 2017) and reports of its increased risk of radical extremism (Gohel, 2011; Ministerium des Innern des Landes Nordrhein-Westfalen [MI NRW], 2018). As one of our main goals was to investigate the effects of ethnic diversity at the team level, we pooled participant data with information about the actual (rather than perceived) diversity in a team. Here, we distinguished between four ways of computing this diversity to offer nuanced insights (see method section for details). Although our analyses were theoretically motivated, we throughout controlled for multiple tests to prevent Type 1 error inflation.

Participants
To be eligible, participants had to be active, adult, amateur, and male soccer players within North-Rhine Westphalia and fluent in English or German. Given the complexity of the data and uncertainty about expected population parameters, we did not conduct a power analysis but followed established recommendations. Various researchers have highlighted the importance of having a sufficient sample size at the highest level of the multi-level model (i.e., for level 2 in the present study; Hox, 1998;Maas & Hox, 2005;Snijders & Bosker, 1993). Although some have argued for a minimum of N = 50 or more at level 2 (Maas & Hox, 2005), several studies have demonstrated that multi-level analyses become sensible for N ≥10 (Huang, 2018)  In total, we collected data from 435 participants. However, to prevent confounding of level-1 and level-2 associations (Gelman & Hill, 2006), we excluded data from players who were the only one of their team participating, respondents who did not fill out all study variables, and those who played in multiple teams (n = 178). This resulted in a final data set of 257 players nested in 36 teams. Demographic information about participants is presented in Table 1. For details regarding ethnicity, refer to SOM, Figures S1-S2.

Procedure
The study was approved by a national center for research data and the institutional review board at the department of the first author. Questionnaires were forward-back translated by bilingual teams in collaboration with independent translators to ensure translation accuracy. Data were collected online between October 2018 and February 2019. Solely teams in the Kreisliga A (English: district league) or lower were targeted because players in higher, semi-professional leagues are paid and hence financially motivated, which could have been a motivational confound in our research. We contacted teams that met this criterion by randomly selecting them from the soccer portal "FuPa," 2 . Additionally, a call was published on the same webpage. To follow up on potential participants, phone calls and in-person visits at trainings were conducted. To reach a sufficient sample size, this sampling procedure was complemented by social media calls and snowball sampling through referrals by coaches. As a financial incentive, the three teams with the most participants in the study received 50, 100, and 150 Euro team funds, respectively.

Measures
Several psychological instruments and macro-level indicators were assessed, as listed below. 3

Independent variables
Individual level Interethnic contact. Interethnic contact frequency and valence were assessed through two questions: "How often do you have contact with [for ethnic majority participants: ethnic minorities; for minority participants: ethnic Germans] while playing soccer?" (frequency) and "To which extent do you experience this contact with [for ethnic majority participants: ethnic minorities; for minority participants: ethnic Germans] during soccer practice as negative or positive?" (valence). Answers were scored on 7-point Likert scales, ranging from 1 (not at all) to 7 (very often) for contact frequency and 1 (negative) to 7 (positive) for contact valence.
Perceived diversity ideologies. The diversity ideologies that the participants perceived (a) to be held by their coach and (b) to be prevalent in the team were examined separately. For this purpose, the 4-item multiculturalism and 4-item color-blindness scales by Ryan, Hunt, Weible, Peterson, and Casas (2007) were adapted. Answers were scored on a 7-point Likert scale ranging from 1 (does not apply) to 7 (does apply). Each four items measured perception of the coach's multiculturalism (e.g., "My coach emphasizes the importance of appreciating group differences between ethnic groups"; ␣ = .77) and colorblind ideology (e.g., "My coach recognizes that all people are basically the same regardless of their ethnicity"; ␣ = .86), and the multicultural (e.g., "My soccer team emphasizes the importance of appreciating group differences between ethnic groups"; ␣ = .89) and colorblind ideologies (e.g., "In my soccer team, people recognize that all people are basically the same regardless of their ethnicity"; ␣ = .90) perceived to be prevalent in the team.
Team social capital. The four-dimensional scale by Forsell et al. (2018) measured perceived team social capital on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Friendly-acceptance was composed of eight items (e.g., "In our team, it is easy to make friends"; ␣ = .87), whereas norms of behavior (e.g., "Our team expects a high standard of behavior"; ␣ = .79), trusting-reciprocity (e.g., "Team members who help other members know the favor will be returned"; ␣ = .71), and governance (e.g., "For decisions, team members discuss issues and decide together"; ␣ = .79) were composed of four items each.

Team level
Four different diversity measures were assessed at level 2. These variables were calculated based on ethnically categorizing team members' names, reported on "FuPa.net", by using the genealogical databases "forebear.io" and "namespedia.com." Each team member was placed into one of the following ethnic heritage categories: German, Eastern European, Turkish, Arab, North/Western European, Southern European, Asian, African, other Western countries. The categories were chosen because they reflect the main immigrant origins in Germany, while also capturing a large geographical area. Coding was executed independently by the principal researcher and for a random subset by a second researcher who was unaware of the hypotheses. Inter-rater reliability was high, Ä = .87, SE = .02 (86.70 % agreement).
Based on this coding, four diversity indices were calculated (see Table 2 for details). In contrast to the traditional minority-majority dichotomous measurement of diversity, Budescu and Budescu (2012) have recommended to include multicategorical measures. Therefore, the Simpson Diversity Index (SDI; Simpson 1949), in addition to more common diversity proportion estimates (i.e., non-Western, minority), were calculated. Given that a high number of Muslim minorities often are seen as increasing the threat of terrorism in public discourse (Brettfeld & Wetzels, 2007;Molthagen & Korgel, 2009), we also wanted to examine whether the proportion of Muslims in a team would play a role for such intergroup attitudes. Each diversity indicator was scored from 0 to 1, with higher values representing more ethnic diversity within the team.

Dependent variables
In total, four dependent variables were assessed at the individual level.

Perceived intergroup threat
To measure symbolic threat, González and colleagues' (2008) 3-item scale (e.g., "German identity is being threatened because there are too many people with a migration background"; ␣ = .94) was adapted. Items were adjusted correspondingly for ethnic minority members, such that the item wordings were matched to their specific ethnic group (e.g., for Turkish participants: "Turkish identity is being threatened because there are too many Germans," for Greek participants: "Greek identity is being threatened because there are too many Germans"). The measure was selected based on Stephan et al.s' (2009) argument that immigrants can also perceive a numerically large outgroup in society (e.g., majority group members) as a threat to their identity. Indeed, previous research has successfully used this measure among majority and minority groups (see Obaidi, Kunst, Kteily, Thomsen, & Sidanius, 2018;Tahir et al., 2019). Realistic threat was also examined through three items from the same authors (e.g., "Because of the presence of people with a migration background, German people have fewer resources"; ␣ = .91) and adjusted for ethnic minority members as described previously. Responses were scored on a 7-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree).

Support for violent extremist groups
The four-item scale (␣ = .84) by Nivette and colleagues (2017) was used. Responses were scored on a 7-point Likert scale ranging from 1 (fully untrue) to 7 (fully true). Items included, for instance, "It's OK to support groups that use violence, commit attacks, or kidnap people to fight for a better world."

Personal violent extremist conviction
The 3-item contextual violence questionnaire by Finkel and colleagues (2016) was used (␣ = .88), encompassing items such as "Do you personally feel that using violence against civilians in defense of your ideological convictions can be justified?" Each item was scored on a 7-point Likert scale ranging from 1 (never) to 7 (often) for items 1 and 2, and from 1 (strongly disagree) to 7 (strongly agree) for item 3.
Step-wise, random-intercept multilevel modeling was performed. Ethnic diversity measures of the soccer teams served as the level-2 indicator, whereas individual intercultural perceptions, social capital, and contact experiences of the soccer players served as level-1 variables.
Following recommendations, level-2 variables were grand-mean centered (Peugh, 2010). Symbolic threat had a skewness of 1.04 (SE = .15) and a kurtosis of .08 (SE = .31), realistic threat a skewness of 1.19 (SE = .15) and a kurtosis of .39 (SE = .31), support for violent extremist groups a skewness of .64 (SE = .15) and a kurtosis of -.54 (SE = .31), and personal violent extremist conviction a skewness of 1.76 (SE = .15) and a kurtosis of 2.72 (SE = .30). While these skewness indicators may, according to more lenient standards (George & Mallery, 2010), still be acceptable, distributions of the residuals in the models violated the assumption of normality (SOM, Figures S3-68). Hence, log transformations were performed on all dependent variables, significantly improving normality of the residuals (SOM, Figures S7-68).
Intraclass correlation analysis justified multilevel modeling for all four outcome variables: = .06 (symbolic threat), = .09 (realistic threat), = .07 (support for extremist groups), = .06 (personal violent extremist conviction). Due to the large number of tests, family-wise Holm's correction (1979) was applied. Interaction effects were only followed-up on with simple slope calculations if they remained significant after this correction. Analyses were conducted in five steps: In Step 1, main effects of the level-2 diversity indices on the dependent variables were estimated. As these diversity indices were calculated on the same genealogical data, violating their data independence, separate models were estimated to examine the associations of each diversity index with each dependent variable. In Step 2, we assessed whether the effects of these indices would be moderated by players' minority versus majority group membership. In Step 3, effects of individual-level intercultural attitudes on the dependent variables were entered as fixed effects because the estimation of them as random effects resulted in the models not converging. Level-2 effects established in Steps 1 and 2 were dropped from these models when their effect was non-significant. Effects of the individual-level intercultural variables on the dependent variables may also depend on whether participants are minority or majority members. Thus, in Step 4, interactions between perceived diversity ideologies (level 1) and players' group membership (majority vs. minority) were added. Furthermore, in this step, the interaction between interethnic contact frequency and valence was added based on previous research (Barlow et al., 2012). In Step 5, we investigated cross-level interactions between ethnic diversity at the team level and intercultural attitudes at the individual level. Again, given the dependency of the team-level independent variables, separate models were estimated for each level-2 diversity measure (Tables 5-8, Models A-D).

Results
Means, standard deviations and correlations for the main study variables across participants (soccer players) and soccer teams are shown in Table 3. Please note that, for brevity, we only focus on effects that were significant after Holm correction in the results.
Step 1: main effects of diversity indices (level 2) As displayed in Table 4, none of the indices had a significant main effect on the outcome variables symbolic threat, realistic threat, and support for violent extremist groups. However, Muslim proportion, Non-Westerner proportion, and minority Table 3 Pearson's correlations across participants (presented above diagonal) and Spearman correlations across teams (presented below diagonal) between the main study variables. Note. a 0 = minority member, 1 = majority member. *p < .05. **p < .01. ***p < .001.

Table 4
Step 1: Main effects of the diversity (level-2) variables on the dependent variables. proportion had significant and positive main effects on personal violent extremist conviction. For model fit estimates, see SOM, Table S3.
Step 2: Diversity indices (level 2) x group membership (level 1) interactions Next, the models presented in Table 4 were extended by adding the main effect of minority vs. majority group membership and its interaction with the level-2 diversity predictor. Compared to the models from Step 1, significant model fit improvements were only observed for the outcome variable symbolic threat, ps ≤ .015 (SOM , Table S3). For all other outcome variables, model fit changes were not statistically significant, all ps ≥ .143 (SOM , Table S3). No significant interactions were observed in any model, ps > .550 (SOM, Tables S7-S10).
Step 3: Main effects of level-1 intercultural perceptions and experiences For the dependent variables symbolic threat, realistic threat, and support for violent extremist groups, none of the ethnic diversity indices were retained in Step 3 as their effects were non-significant (see Steps 1-2). For personal violent extremist conviction, we present a model (Table 8, Model 3) in which we retained the effects of the Muslim proportion variable as it was the most predictive diversity index in the previous steps. For models including the other predictive diversity indices, see SOM, Table S11.

Symbolic threat
As illustrated in Table 5 (Model 3), majority members indicated significantly more symbolic threat than minority members. Furthermore, higher scores on the social capital dimension norms of behavior were associated with significantly more symbolic threat. Both associations remained highly significant after Holm correction (Table 5, Model 3).

Realistic threat
As displayed in Table 6 (Model 3), higher norms of behavior were associated with more realistic threat, remaining highly significant after Holm correction. In contrast, perceived team color-blindness and interethnic contact frequency were both significantly related to less realistic threat after Holm correction.

Support for violent extremist groups
None of the main effects of level-1 variables remained significant after Holm correction (Table 7, Model 3).
Step 4: individual-level interactions between perceived intercultural ideologies, contact experiences and group membership In Step 4, we added interactions between the level-1 variables perceived intercultural ideologies (i.e., color-blindness, multiculturalism) and minority vs. majority group membership. Furthermore, an interaction between interethnic contact frequency and valence was added. None of the models showed significant model fit improvements compared to step 3 (SOM, Table S4). None of the interactions reached statistical significance for symbolic and realistic threat (Tables 5 and 6, Model 4). For support for violent extremist groups and personal violent extremist conviction, no interaction remained significant after Holm correction (Tables 7 and 8, Model 4).
Step 5: Cross-level interactions between intergroup perceptions, experiences and diversity indices Finally, we tested whether ethnic diversity at the team level would interact with intercultural perceptions at the individual level. Given the dependency of the level-2 independent variables (i.e., the diversity indices), separate models were again estimated for each diversity index (Tables 5-8, Models 5A-D, with each letter reflecting a model with a different diversity predictor). For brevity, we only focus on significant findings. For model fit comparisons, see SOM, Table S6.

Cross-level interactions with contact valence
For support for violent extremist groups, there was a significant negative cross-level interaction of SDI with interethnic contact valence (Table 7, Model 5A). As displayed in Figure 2, a higher ethnic diversity as reflected by higher SDI values was related to an increase in support for violent extremist groups when interethnic contact valence was low/negative, b = .48, SE = .18, t(30) = 2.59, p = .015, but to a decrease in support for violent extremist groups when interethnic contact valence was high/positive, b = -.59, SE = .20, t(30) = -2.90, p = .007.
Similarly, there was a significant positive cross-level interaction between the proportion of non-Westerners in the teams and interethnic contact valence on support for violent extremist groups (Table 7, Model 5C). A higher non-Westerner proportion in a team was significantly related to more support for violent extremist groups when contact valence was low/negative, b = .54, SE = .18, t(30) = 2.98, p = .006, but not when contact valence was high/positive, b = -.06, SE = .16, t(30) = −.35, p = .730, see Figure 2 .

Cross-level interactions with contact frequency
Significant positive interaction effects between three diversity indicators (i.e., the proportion of Muslims, non-Westerners, and minority members in the team) and contact frequency were observed on support for violent extremist groups (Table 7, Models 5B-D). Simple slopes analyses demonstrated that the proportion of Muslims in a team was not significantly associated with support for violent extremist groups when interethnic contact frequency was low, b = −.20, SE = .21, t(30) = −.92, p = .363, but was related to more such support when interethnic contact frequency was high, b = .69, SE = .22, t(30) = 3.13, p = .004, see Figure 2. Similarly, a higher non-Westerner proportion had no significant association with support for violent extremist groups when interethnic contact frequency was low, b = −.15, SE = .17, t(30) = −.90, p = .374, but was related to more support for violent extremist groups when interethnic contact frequency was high, b = .64, SE = .19, t(30) = 3.31, p = .002, see Figure 2. Finally, a higher minority proportion within teams was not significantly associated with support for violent extremist groups when interethnic contact frequency was low, b = −.24, SE = .19, t(30) = −1.23, p = .226, but was related to significantly more support for violent extremist groups when contact frequency was high, b = .60, SE = .20, t(30) = 2.96, p = .006, see Figure 2.

Discussion
Much research has argued that soccer may function as an arena of positive social change (Müller et al., 2008;Halm, 2006). Yet, little is known about the psychological and contextual factors shaping political attitudes, including extremism in this context (Körting et al., 2015;Maurer, 2017; United Nations Office on Drugs and Crime [UNODC], 2018). In the present research, we found evidence that individual-level factors independently and in interaction with contextual variables (i.e., macro-level ethnic diversity) may be associated with extremist attitudes and factors closely interrelated with such attitudes, namely threat perceptions (Obaidi et al., 2018).
Some evidence suggested that more diversity (assessed at the team level) was related to more personal violent extremist conviction but not to threat or support for violent extremist groups. Yet, a cross-level interaction between ethnic diversity and how players perceived interethnic contact in their teams was observed in terms of support for violent extremist groups. When contact quality was low, diversity was associated with more support for violent extremist groups, whereas some evidence suggested that diversity was related to less support when contact quality was high. Importantly, when contact frequency was high in models controlling for contact quality, three diversity indices (proportion of Muslims, non-Westerners, and minority members in a team) were significantly associated with more support for violent extremist groups. These findings emphasize the importance of considering the interplay between contextual-and individual-level processes when investigating the formation of intergroup and intercultural attitudes within sports teams, including their extreme forms.

The role of individual-level intercultural experiences and perceptions
Based on extremism, intergroup and intercultural relations research (Berry, 1997;Pettigrew, 1998;Putnam, 2000;Stephan & Renfro, 2002), one could have expected that individuals who (a) perceived high social capital within the team, (b) felt that the coach or the team endorsed a multicultural rather than colorblind ideology, or (c) were involved in frequent and positive interethnic contact during the training would score lower on threat and violent extremist attitudes. Mixed evidence was found for these propositions. When the team was perceived to endorse stricter behavioral norms, a factor typically equated with higher social capital (Forsell et al., 2018;Putnam, 2000), individuals displayed more symbolic and realistic threat. This finding nuances previous research, which has highlighted social capital's potential to improve intergroup and intercultural relations (Laurence, 2009;Lee, 2013;Verhagen & Boonstra, 2014) and to decrease involvement in extremism (Boehnke et al., 1998). It is possible that the stricter behavioral norms variable in the present study functioned as a proxy of conformity or contextual tightness, which is often related to more negative intergroup outcomes (Gelfand et al., 2011(Gelfand et al., , 2013Whitehouse, Yustisia, Putra, Kavanagh, & Rufaedah, 2019). In fact, Gelfand and colleagues (2011) found that individuals in tight cultures (that value strict norms and conformity) were more likely to display negative attitudes towards outgroups (e.g., immigrants) and were more concerned about outgroup threats. One explanation for the present findings may thus be that, when soccer teams are perceived as having strict behavioral norms, players perceive people from other ethnic groups as potentially norm-deviant and threatening.
Whereas no significant association with multiculturalism was found, the results showed that when individuals perceived their team to endorse a colorblind attitude, they perceived less realistic threat and showed less personal violent extremist conviction. This finding stands in contrast to some previous research (e.g., Neville et al., 2013;Richeson & Nussbaum, 2004), yet is consistent with Dovidio and colleagues (2016), who suggested that color-blindness may distract group members from group disparities. Riek et al. (2010) further demonstrated that making common identities salient can improve outgroup attitudes through less perceived threat (also see Jaśko & Kossowska, 2013;Saguy et al., 2009). Hence, when the team atmosphere is perceived to be colorblind, thus focusing on commonalities and downplaying differences, soccer players may, at least at a superficial level, perceive less threat and hold more positive outgroup attitudes.
Importantly, the associations of color-blindness with personal violent extremist conviction, in line with previous research, depended on whether participants were minority or majority members (Rattan & Ambady, 2013;Verkuyten, 2007) In the present study, these findings seemed to reflect suppressor effects and should be interpreted with caution. We nevertheless provide some discussion here. Minority members who perceived the team to endorse a colorblind ideology showed less personal violent extremist conviction, whereas the variables were unrelated among majority members. Again, following Dovidio and colleagues' (2016) rationale, it is possible that color-blindness may distract minority members from prevailing inequalities and makes them perceive intergroup relations as fairer. As such, if one understands extremism as an extreme form of collective action, it may make sense that minority members in colorblind teams may show less of these collective action tendencies.
The more frequent interethnic contact players experienced, the less realistic threat they perceived. This is in line with intergroup contact theory (Allport, 1954;Pettigrew, 1998), suggesting that when people belonging to different groups interact, discrimination and prejudice towards the outgroup may decrease (Wagner et al., 2008; for a review see Pettigrew & Tropp, 2006;Pettigrew, Tropp, Wagner, & Christ, 2011). In contrast to previous work (Barlow et al., 2012;Laurence, Schmid, & Hewstone, 2017;Van Dick et al., 2004), contact valence was not significantly associated with threat, nor was there a significant interaction between contact valence and frequency. This suggests that at least in soccer teams, both may play independent roles, but future research is needed to replicate this finding.

Interactions between the team and individual level
Team-level ethnic diversity alone was not associated with threat perceptions, nor with personal violent extremist conviction. However, it was significantly associated with support for violent extremist groups when diversity was measured as the proportion of Muslims, non-Westerners, or ethnic minority members in a team. Although these associations became non-significant when controlling for individual-level variables in later steps, this finding suggests that ethnic diversity alone cannot always be regarded as improving intergroup relations (Hjerm, 2007;Schlueter & Scheepers, 2010;Van Assche, 2018). Yet, in support of person-environment theorizing (Magnusson & Magnusson, 2013), the associations of team-level ethnic diversity with support for violent extremism were conditional on individual-level variables. Ethnic diversity was related to more support for violent extremist groups when players perceived interethnic contact to be negative or frequent, but to less support for violent extremist groups when contact quality was high. These findings suggest that ethnic diversity in sports environments may backfire if a positive intercultural environment is not in place. Also, it shows that solely increasing the frequency of contact may have unintended effects and suggests that the focus should be on achieving positive forms of contact. To be noted, our study was underpowered to test whether the interaction of ethnic diversity with contact frequency was further moderated by contact valence (i.e., as a three-way interaction). Future studies should therefore test whether the most positive intercultural outcomes can be achieved when ethnic diversity, contact frequency, and contact valence are high at the same time.
It should also be noted that whether interactions were statistically significant depended on how extremist attitudes were assessed; all significant cross-level interactions were observed for Nivette and colleagues' (2017) measure of support for violent extremist groups, but not for Finkel et al.s' (2016) measure of personal violent extremist conviction. Thus, as both our measures included the concept of violence, it seems as if what mattered in the present study was whether it involved one's active (intended) involvement in such violent action. Future research may follow up on these distinctions while also testing whether our findings may be replicated in terms of extremism in itself without its potential violent components.
Finally, our findings suggest that ethnic diversity, under negative contact conditions, may primarily be associated with group-based extremism, rather than changing people's individual extremist convictions. One reason for this may be the potential of contact to reinforce group distinctions, which in competitive (rather than collaborative) settings such as sports may lead to more negative intergroup outcomes (cf. Dovidio, Gaertner, & Kawakami, 2003).

Limitations
The present study extended the emerging literature on the Context x Individual interaction debate of ethnic diversity (Magnusson & Magnusson, 2013;Van Assche, 2018) by applying it to the context of extremism in soccer teams. Yet, some limitations should be noted. First, whereas we examined a sample of 257 players from 36 teams and controlled for multiple comparisons, thereby reducing the risk of Type-I errors, some teams consisted of only two players, which may have reduced the accuracy of some estimates or led to some tests being underpowered (Gelman & Hill, 2006;Snijders & Bosker, 1993).
Next, the study focused deliberately on one of the most diverse regions in Germany, where extremism is a growing problem (MI NRW, 2018). Yet, the minority sample was relatively small compared to the majority sample. Furthermore, both samples were non-representative, and data were cross-sectional, preventing causal conclusions.
One could argue that the lack of female participants in the study is a limitation (Bjørgo & Gjelsvik, 2015). However, given that the majority of violent right-wing (Götschenberg, 2018) and Islamist extremists (BKA et al., 2016) are male and that the number of male soccer players is more than five times higher than that of female soccer players in Germany (Statista, 2019), this was the most feasible approach for the present research. Nevertheless, future studies may profitably examine the role of gender.
The fact that the majority of the ethnic minority participants in this study identified as Christian may limit, or at least complicate, the interpretation of the results. Some may argue that, in terms of extremism, it would have been most relevant to focus on Muslim minority vs. Christian majority members. Yet, we believe that extremism is not specific to certain ethnicities or religions but is rather driven by status differences in society (Kunst & Obaidi, 2020). Indeed, both individuals with Eastern European and Middle Eastern backgrounds (the two largest groups in our minority sample, see SOM, Figure S1) experience low status in German society.
Reflecting a general limitation of most research on extremism, the present study assessed several socially undesirable attitudes, such as support for violent extremist behaviors. Due to the sensitivity of the content, participants might not always have responded entirely truthfully to the questions (Randall & Fernandes, 1991). This was one of the reasons why we also included less sensitive measures (i.e., threat perceptions) that have proven to be reliable and robust predictors of extremism (e.g., Obaidi et al., 2018). Moreover, the extremism measures only assessed violent extremist attitudes/support, not actual extremism, which tends to be challenging to assess, both practically and ethically (Bjørgo & Gjelsvik, 2015).
It should also be noted that we created diversity measures based on genealogical coding of names. While this provided an arguably more accurate proxy measure of ethnic diversity than asking players about their perceived team diversity, and despite high interrater reliability, there are many ways in which names can be categorized. For instance, what constitutes a "Western" group can be discussed (Stearns, 2008).
The independent variables in this study were explicitly framed to the soccer context, whereas the dependent variables assessed attitudes related to society more broadly. This may have limited the explanatory power of our independent variables. For future research, it may be beneficial to measure variables such as color-blindness and multiculturalism not only framed towards the team and coach but also towards society at large to tease out the unique importance of psychological processes taking place within soccer teams.
Finally, one could argue that the team level is essentially a meso rather than macro level. As such, other higher-level variables could have influenced the associations. For instance, degrees of actual inequality within participants' neighborhoods and their perception of it could have been assessed.

Societal implications
Based on the results, we recommend that stakeholders aim for a multi-factorial approach to prevent extremist radicalization and threat perceptions in sports settings. Specifically, as this study suggests, one fruitful way to capitalize on the positive potential of diversity in soccer teams may be to nurture positive contact between players. Coaches could implement this knowledge, for instance, through bonding activities (e.g., team excursions) and team-building workshops. Most importantly, increasing ethnic diversity in teams should not be seen as a blueprint for reducing or preventing extremisms. Rather, how this diversity is managed seems to be of decisive importance.