Conditions for cultural belonging among youth of immigrant descent in Germany, the Netherlands, Sweden and the United Kingdom. Comparative analysis of intergroup experiences and classroom contexts

ABSTRACT Intergroup contexts of classrooms have an influence on mainstream and ethnic belonging of youth of immigrant descent. Based on the normative reference group theory, a higher level of ethic identification is expected in classrooms with a higher proportion of co-ethnic youth and, based on constrict theory, a lower level of mainstream identification is expected in ethnically more diverse classrooms. However, these relations between ethnic composition and cultural belonging may differ between countries due to differences in multicultural policies and colonial histories. Using two waves of data from CILS4EU survey, we conduct a cross-national comparison of the relation between intergroup contexts of classrooms and cultural belonging. The results imply that both theories cannot be generalized across national contexts, and other factors vary as well across countries. Therefore, research and policy in this area should pay more attention to the national context in terms of policies and histories.


Introduction
Children and youth of immigrant descent develop a potential dual cultural sense of belonging or identity: It reflects how strongly they identify with the mainstream culture (dominant culture in the receiving country) and their heritage (ethnic) culture or with both at the same time (Berry 2011;Faas 2009). These cultural identities are important not only for the individual development of youth but also for the society in question. On the one hand, they influence individual psychological wellbeing and school performance (Brunsma 2005;Miller-Cotto and Byrnes 2016;Schachner, van de Vijver, and Noack 2018b;Schotte, Stanat, and Edele 2018). On the other hand, for society as a whole, a certain level of mainstream identification is considered an essential dimension of social cohesion (Schiefer and van der Noll 2017).
Schools and especially classrooms are particularly important microsystems in this regard: The classroom level is very salient for European youth, given that they usually spend most of the school day with the same peers within their classroom. Consequently, current research focuses on processes that unfold within classrooms (Syed, Juang, and Svensson 2018). Ethnic diversity increases in many Western schools, changing the ethnic composition of schools and classrooms and increasing opportunities for contact with members of different ethnic groups. Indeed, classroom ethnic composition is considered a particularly important structural classroom intergroup context influencing cultural identity development (Agirdag, van Houtte, and Van Avermaet 2011;Schachner et al. 2018a;Umaña-Taylor 2004).
Studies on classroom ethnic composition distinguish between ethnic diversity and the proportion of immigrant descent students (Rjosk et al. 2017). Previous research shows for immigrant descent youth mixed results for the relations between ethnic diversity and cultural identification between different countries (Agirdag, van Houtte, and Van Avermaet 2011;Schachner et al. 2016). It is therefore likely that the impact of classroom composition on the cultural identification of the individual systematically varies from country to country: There are notable differences in the broader intergroup climate and context (Guimond, de la Sablonnière, and Nugier 2014) which are due to differences in the country-level political, historical and social context (Faas 2009;Janmaat 2014;Suárez-Orozco et al. 2018;Heizmann 2016;Heizmann and Ziller 2020).
Yet we know little about between-country differences in how classroom context relates to cultural belonging, since the mostly single-country study findings from different country contexts are difficult to synthesize because of divergent research methods (Phalet and Baysu 2020;Syed, Juang, and Svensson 2018).
A number of country comparative studies empirically show the importance of multicultural or integration policies and histories for students' belonging (Ham, Yang, and Cha 2017;Schachner et al. 2017), or behaviour (Veerman 2015). Moreover, the relationship between the proportion of immigrant descent youth and inclusive views on immigrants was found to be positive in old immigration nation states but non-related in new immigration states (Janmaat 2014). Taken together, these prior findings point to migration or colonial histories on the one hand and the political framing of diversity on the other as primary potential sources of cross-national variation. By including both into our analytical framework we address central structural and political aspects of immigration in these countries.
The aim of this paper is to provide a cross-national comparison on the relations between classroom intergroup contexts and the cultural belonging of immigrant descent youth. We take an interdisciplinary approach and discuss possible between-countries differences based on theoretically proposed mechanisms of multicultural policies (MP) and migration histories (MH). While previous research comprised mostly single country studies with divergent research methods, we will conduct a cross-national comparison with a uniform research method. The data from "The Children of Immigrants Longitudinal Survey in Four European Countries" (CILS4EU) is very suitable for this due to the combination of uniform questions regarding cultural belonging in different countries and the longitudinal study design. As we are limited to an investigation of four countries, we will only formulate expectations regarding differences in classroom-level effects between these countries. This also implies that our comparative approach will be a comparison of result patterns, as with four countries we cannot actually test countrylevel factors and cross-level moderation processes. Nonetheless, the four countries investigated cogently represent the four logically possible combinations of MP and colonial histories.

Theory
Experiencing a sense of belonging is a basic human need and an important condition for motivation and well-being (Baumeister and Leary 1995). A sense of belonging may be experienced at different levels, such as in school or through identifying as a member of a cultural group, and can be an important resource specifically for immigrant descent youth ). As mentioned above, the cultural identity of people from immigrant descent comprises at least some mainstream and ethnic identity components, i.e. identification with these two cultural contexts, including their communities and institutions. A number of studies show that the simultaneous development of both mainstream and ethnic identity is important for the well-being of immigrant descent youth (see also Suárez-Orozco et al. 2018, for a review). In examining the strength of identification with mainstream and ethnic cultures, such studies address ethnic and mainstream belonging, ethnic and mainstream orientation or identification and often these terms are used interchangeably (Schachner et al. 2016;Thijs, Keim, and Geerlings 2018).
Drawing on contextual models of development and the adaptation of immigrant descent youth (Bronfenbrenner and Morris 2006;Suárez-Orozco et al. 2018), we distinguish three levels of intergroup experiences and contexts that can affect cultural identification and belonging among young people from an immigrant descent: individual experiences, classrooms, and countries as higher-level contexts (see also Schachner, van de Vijver, and Noack 2018b, for a review). In this study, we will mainly focus on the classroom and country context, while taking into account relevant individual-level factors and their country variations as well.

Classroom intergroup context: ethnic composition
For the classroom intergroup context, ethnic composition is considered an important structural aspect relating to cultural identification (Agirdag, van Houtte, and Van Avermaet 2011;Edele et al. 2020;Schachner et al. 2016;Syed, Juang, and Svensson 2018). Ethnic composition can be measured with various indicators: the proportion of co-ethnics, the proportion of non-immigrant descent students, and diversity in terms of the number and relative size of ethnic groups present in the classroom (Rjosk et al. 2017).
Whereas contact theory (Allport 1954; Van der Meer and Tolsma 2014) and conflict theory (Blalock 1957) focusses mostly on attitudes towards persons from other groups, normative reference group theory (NRGT) (Merton 1968) and (2) constrict theory (CT) (Putnam 2007) outline processes of collectivity or cultural identification of one's own group (Agirdag, van Houtte, and Van Avermaet 2011). Therefore, we will elaborate these last two theories below. Both theories spring from the idea that the ethnic composition of a given context determines the basic level of opportunities for engaging with specific groups (Blau 1974). These opportunities can generate further consequences within the classroom context. The two theories differ in that NRGT focusses on the proportion of co-ethnic or non-immigrant descent students, and CT on ethnic diversity.
NRGT postulates that experiencing a group context with its constant interactive reproduction of norms can be considered normative for the cultural identification of individuals (Merton 1968). NRGT therefore underlines the importance of the dominant identity in the group for cultural identification. Following this theory, a higher proportion of co-ethnics should facilitate opportunities for engaging with one's ethnic culture and thereby enhance ethnic identification (Syed, Juang, and Svensson 2018). Conversely, a higher proportion of non-immigrant descent youth provides more opportunities for immigrant descent youth to engage with the mainstream culture. In a class composed mainly of non-immigrant descent youth, mainstream identification should be higher than in a class with many immigrant descent youth.
Yet, if the group of immigrant descent youth in the classroom is very diverse in terms of the ethnic groups present, children lack a clear focal ethnic group. CT explains the effect of a group composition on cultural identification based on the absence of a dominant identity (Agirdag, van Houtte, and Van Avermaet 2011). According to CT (Putnam 2007), the lower opportunity for shared social norms in ethnically diverse contexts stimulates an overall decline in social cohesion, i.e. all groups present within a given context are adversely affected by ethnic diversity. More precisely, CT states that ethnically diverse settings counteract the formation of social cohesion, operationalized as social relations and trust in both ingroup and outgroup and between them due to anomie (a lack of common identities and values) (Van der Meer and Tolsma 2014). Given that social cohesion in educational settings is regarded as a multidimensional construct that includes social relations, belonging and orientation towards the common good (Veerman and Denessen 2021), less social relations and trust are expected to reduce belonging. Following the reasoning of anomie, in the short term Agirdag and colleagues (2011) similarly expect a decline in both ethnic and mainstream belonging in ethnically diverse classrooms.
The expectation based on CT of lower ethnic identification in a classroom with higher diversity seems contrary to NRGT, because a higher diversity consists of more and lager ethnic subgroups, leading to a correlation between ethnic diversity and the proportion of co-ethnics. However, these two indicators refer to two different mechanisms that could occur at the same time. While ethnic diversity refers to the ethnic makeup of the whole class, the proportion of co-ethnics could differ between ethnic groups within the class. Therefore, proportion of co-ethnics refers to influences of focal identities of ethic subgroups in the class, and the ethnic diversity refers to the influence of the lower likelihood of a focal identity of the whole class.
A range of empirical studies found relations between the ethnic composition and cultural identification of youth (Agirdag, van Houtte, and Van Avermaet 2011;Edele et al. 2020;Schachner et al. 2016;Syed, Juang, and Svensson 2018). Some empirical studies in Belgium and Germany have produced findings in line with NRGT. They show an increase in mainstream orientation or identification among immigrant descent youth in classrooms with a higher proportion of non-immigrant descent youth (Agirdag, van Houtte, and Van Avermaet 2011;Edele et al. 2020;Schachner et al. 2016). However, regarding the proportion of co-ethnics, NRGT is not confirmed in a number of countries. A negative relationship between the share of co-ethnics and ethnic identification was found in the United States (Umaña-Taylor 2004), while no relationship between the share of co-ethnics and both mainstream and ethnic identities was found in Germany (Edele et al. 2020). This may reflect the fact that on the classroom level, even in a classroom with relatively large ethnic groups, the ethnic group of non-immigrant descent has more power to attain normativity than any other ethnic group from the immigrant descent students, because it reflects the normativity of the mainstream population as such.
For the classroom intergroup context, drawing on NRGT, we expect youth exposed to a higher proportion of non-immigrant descent youth to perceive higher levels of mainstream belonging (H1a). Moreover, immigrant descent youth exposed to more co-ethnic youth in their classroom are expected to perceive stronger ethnic belonging (H1b).
Empirical studies on the relationship between ethnic diversity and cultural identification have so far produced mixed evidence for different countries. The findings of Schachner et al. (2016) and Edele et al. (2020) do reject the CT on the relation between ethnic diversity and mainstream orientation in German secondary schools. Contrary to these findings, Agirdag, van Houtte, and Van Avermaet (2011) partly confirm the CT in Belgian primary schools, with lower mainstream identification found in ethnically more diverse schools, at least amongst immigrant descent students. Agirdag, van Houtte, and Van Avermaet (2011) state that it is important to consider mediators: more diversity, for example, also leads to more cross-ethnic friendships, which in turn promote social cohesion and more inclusive mainstream identifications. They empirically show in their study that the negative relation between ethnic diversity and mainstream belonging for immigrant descent students is partly suppressed by the friendships with non-immigrant descent students. These findings are in line with the review of Graham (2018), who concludes that cross-ethnic friendships are an important mediating mechanism to explain relations between structural school features (diversity) and attitudes of students. Therefore, it is important for the interpretation of effects whether a study includes or excludes cross-ethnic friendships in the analysis.
The mixed findings could have several causes, as studies vary in research method (longitudinal versus cross-sectional; focus also not all on explaining mechanisms such as cross-ethnic friends), the age and educational level of youth (primary versus secondary education), and again the country intergroup context, i.e. youth country of residence (Germany versus Belgium). This underlines the need for a comparative approach. In the next section we therefore address the country as an intergroup context and how it may influence the relationships mentioned above.

Country intergroup context: multicultural policies and migration histories
For the country intergroup context, both MP and MH are expected to affect the cultural orientations of immigrant descent youth (Castles and Miller 2009;Faas 2009;Suárez-Orozco et al. 2018). Such policies and histories are likely to have a direct impact on the cultural belonging of immigrant descent youth, but they may also explain differences between countries in the relations between ethnic diversity as an indicator of the classroom composition and mainstream belonging and cultural belonging. Focusing on these two contextual factors has several conceptual and empirical advantages. Combining the historical structure of immigration and the political approach regarding ethnic diversity covers a large bundle of differentiating aspects. Moreover, the second advantage is that our countries feature all possible combinations of both factors. Finally, as we will see, both factors are highly relevant for intergroup relations and identity formation processes.
MP in education focuses, for instance, on heritage culture and language training. More generally, it supports people with immigrant descent in maintaining cultural customs and conventions of their respective heritage culture, for example by allowing certain dress code exemptions or funding ethnic group organizations. It therefore demonstrates a distinct political view on the relationship between minorities and the majority. Sweden and United Kingdom have more extensive MP when, compared to Germany and the Netherlands (Banting and Kymlicka 2013).
Arguments based on the Common Ingroup Identity Model (CIIM) (Gaertner and Dovidio 2000) posit that MP can serve to redefine mainstream society, exemplified by Canada's commitment to diversity becoming a constitutive factor of that society. Therefore, it can be said that this plurality-based political stance has become an integral part of Canadian identity. Under such conditions, MP may also foster a mainstream orientation. Therefore, given the appreciation of diversity embodied in these policies, it is expected that they facilitate heritage culture orientation, though not necessarily an exclusive one. In seeming contrast to the CIIM, critics suspect that such policies facilitate separation or "parallel societies" (Barry 2001;Koopmans 2010), whereas proponents argue that multicultural rights are of fundamental importance for the integration of people with immigrant descent (Igarashi 2019;Kymlicka 1995;Vertovec and Wessendorf 2009) or positive for integration and mixed for non-immigrant descent people (Ward et al. 2018).
In line with the argument from the CIIM, a recent empirical international comparison shows that the mainstream orientation of immigrant descent youth is stronger in countries with a stronger MP (Yagmur and van de Vijver 2012) and the identification gap to non-immigrant descent youth of first generation non-European adult immigrants is smaller in countries with stronger MP (Igarashi 2019). Yet, a strong mainstream identification is also often paired with a strong ethnic identity in these countries and both contribute to positive outcomes (Schachner et al. 2017). Moreover, Berry et al. (2006) suggest that policymaking should encourage at least some retention of heritage culture in conjunction with an open national society to achieve the most desirable psychological outcomes for immigrant descent youth.
In short, if MP appears to buffer the negative relationship between diversity and mainstream belonging (i.e. the effect is weaker), we expect that the constrict hypothesis can be rejected for Sweden and the United Kingdom (H2a) with their relatively strong MP.
Besides these policies, countries differ with respect to MH, which in turn have shaped their general posture towards immigration and attendant policies (Janmaat 2014;Veerman and Dronkers 2016). This posture can also explain differences in the experiences of youth. One major axis of division between countries is the extent to which each has its own colonial history (Castles and Miller 2009;Drouhot and Nee 2019), which may produce differing views on immigration-related diversity, but also on the process of integrating immigrant descent youth themselves (Veerman and Dronkers 2016). The United Kingdom and the Netherlands have a much more extensive colonial history than Germany or Sweden (Conrad 2013;Drouhot and Nee 2019;Naum and Nordin 2013). The CIIM (Gaertner and Dovidio 2000) again suggests that a country with a long colonial past can more effectively claim a common superordinate group due to its international ties with former colonies. I.e. the notion of being British or Dutch may be more inclusive, also including citizens of and migrated from former colonies, whereas being German or Swedish may be more exclusively reserved for people who are perceived to be ethnically German or Swedish. Guerra et al. (2016) present for the impact of a colonial history empirical evidence that people with Brazilian and African descent in Portugal claim a common superordinate group. Furthermore, Verkuyten and Martinovic (2015) argue that people with Surinamese descent in the Netherlands perceive themselves as more prototypical for the superordinate group than people with Turkish and Moroccan descent, and they attribute this to the colonial history between the Netherlands and Surinam. However, a common ingroup identity does not entail an equal status between the groups; an overarching identity can emerge nonetheless, as evidenced empirically by these studies.
Given that a part of the immigrant descent children in ethnically diverse classrooms is more likely to identify with this common superordinate group, the proposed lack of a shared identity in ethnically diverse context is expected to be less likely to appear in ethnically diverse classrooms in countries with a colonial history. Therefore, in line with the CIIM (Gaertner and Dovidio 2000) we expect positive relations or non-relations between ethnic diversity and mainstream belonging in countries with a colonial history. More precisely, if colonial MH appear to buffer the negative relationships between diversity and mainstream identification (i.e. a weakened effect), the constrict hypothesis will be rejected for the Netherlands and United Kingdom (H2b) with their relatively extensive colonial histories.

Data
To test the hypotheses we have used the CILS4EU data (Kalter et al. 2016). The CILS4EU provides valuable data on the cultural belonging of immigrant descent youth, because schools with a higher proportion of immigrant descent youth are oversampled. The number of participating classes varies between 214 in England and 271 in Germany. The CILS4EU contains information on three annual waves in secondary education. Most youth turned 14 during wave 1. We used only wave 1 (2010/2011) and wave 2 (2011/ 2012), because some youth in the Netherlands left secondary education after wave 2. Ethical approval was obtained by three ethical committees (see for instance Regional Ethics Committee, Stockholm, 2010/1557-31/5). Informed consent was obtained from school principals, parents and youth participants.
Youth answered questions about mainstream and ethnic belonging and individual intergroup experiences in both waves. As we have used restricted data from that survey, we performed our analyses in the Secure Data Centre of GESIS. 1 The full CILS4EU sample in wave 1 provides information on 18,716 youth. We calculated our classroom intergroup context variables using the full sample on whom data concerning their country of origin were available. Of the 8,557 immigrant descent youth (including first, second and third generation) who participated in wave 1, between 18.9 and 26.2 percent dropped out in wave 2. Due to missing data on the variables, our sample contains 5,553 youth in 869 classrooms in four countries.

Classroom intergroup context measures
We constructed all classroom intergroup context variables using the country of origin and generation information provided by the CILS4EU (Dollmann and Kalter 2014). The share of co-ethnics was computed by dividing the number of co-ethnic youth in the class minus one by the number of youth in the class. The share of non-immigrant descent youth was constructed by counting the number of youth per class with no migration background minus one divided by the total number of youth per class minus one.
Ethnic diversity is measured with an inverse Herfindahl index. This indicator takes into account both the number of different ethnic groups 2 and the relative group size. Given that we are interested only in the diversity of immigrant descent youth in the class, we calculated the index, like earlier studies in Germany and Belgium, as follows: 1 − (proportion ethnic group 1) 2 + (proportion of ethnic group 2) 2 + … + (proportion ethnic group n) 2 .

Country intergroup context measures
The Multicultural Policy Index 2010 (MCP) 3 was used as an indicator for the country multicultural intergroup context. Research on European colonial migration flows regards the Netherlands and United Kingdom as former colonial powers (Hansen 2004, 26). Although, following the United Kingdom and the Netherlands, Germany had the fourth largest colonial empire in the 1880s, German colonial history lasted for only 30 years, a relatively short period compared with the Dutch and British empires (Conrad 2013). The Swedish number of overseas colonies is regarded as relatively few and short-lived (Naum and Nordin 2013, 14). We therefore indicate only the Netherlands and United Kingdom as countries with an extensive colonial past.

Individual level measures
Five measures are considered important at the individual level: perceived discrimination, socio-economic status, gender, first generation and the number of non-immigrant descent friends.
One of the main individual intergroup explanations for an increase in ethnic identification and a decrease in mainstream identification is discrimination. People who experience discrimination on ethnic grounds are more likely to develop a reactive ethnic identity (Rumbaut 2008). It is reactive because it does not emerge spontaneously from the ethnic group of the immigrant descent students but is an answer to perceived exclusionary tendencies in the society. These perceptions lead to the membership of one's own ethnic group (Schachner et al. 2018a) being emphasized and heightened in a compensatory fashion. The perceived discrimination covariate was constructed using information on youth answers to the question how often they feel discriminated against or treated unfairly in schools, public transport or shops, stores and cafés and by police or security guards, with a four-point Likert scale with answers ranging from 1 (always) to 4 (never). Based on Dollmann (2021) we calculated the mean of the four answers on perceived discrimination. Analyses of the differences in Cronbach Alpha's between the four countries in our selection of the dataset, indicate that the alpha's vary from 0.51 to .68, with only Germany below .66. 4 A higher socio-economic status is regarded as another important individual indicator for the adaptation of immigrant descent youth (Suárez-Orozco et al. 2018). It is also likely to influence the cultural belonging of youth (Brunsma 2005). Correspondingly, being a member of a marginalized, lowstatus group can be perceived as a barrier to identification with the powerful, more privileged majority (Branscombe, Schmitt, and Harvey 1999). But there is also contrary evidence (Verkuyten 2016) and studies that find no relationship with mainstream identification (Agirdag, van Houtte, and Van Avermaet 2011;Edele et al. 2020). Both information on the number of books and the occupational status of the parents was collected, but the occupational status question had a higher number of missing cases than the number of book question. Although the number of books variable is a crude indicator (Engzell 2021), it is useful as children find it hard to answer more precise questions about their parents' professions and education (e.g. Schachner et al. 2017). A five-point scale ranging from 1 (0-25) to 5 (more than 500) was used.
Third, we use the gender of students as covariate, because female immigrant descent youth identify more strongly with the mainstream identity (Agirdag, van Houtte, and Van Avermaet 2011;Edele et al. 2020).
The fourth covariate, first generation, distinguishes first from second and third generation students, as the latter have more experience with the host country and thus a different outlook on ethnic and mainstream orientations (Edele et al. 2020;Platt 2014;Schulz and Leszczensky 2016).
Finally, as noted above, the number of cross-ethnic friends is expected to be an important mechanism (Graham 2018) that suppresses the negative impact of ethnic diversity on mainstream belonging (Agirdag, van Houtte, and Van Avermaet 2011). Moreover, the number of non-immigrant descent friends is, based on the NRGT, likely to be a mechanism that partly explains the relation between the proportion of non-immigrant descent youth and mainstream belonging. Comparable to Agirdag, van Houtte, and Van Avermaet (2011), we operationalized the number of non-immigrant descent friends as indicator of cross-ethnic friends for immigrant descent students.

Dependent variables
The CILS4EU contains data on youth concerning their cultural belonging, including mainstream belonging and ethnic belonging as our two dependent variables. Mainstream belonging is based on youth reporting how strongly they feel that they belong to the country where they live. For instance, youth in the Netherlands were asked: "How strongly do you feel Dutch?" Ethnic belonging was measured with two questions. First students were asked after the mainstream belonging question: "Some people feel that they belong to other groups too". Students were provided with the option to mention ethnic groups or no other group. For the ethnic group youth specified as they feel they belong to: "How strongly do you feel that you belong to this group?" The dependent variables were measured on a fourpoint Likert scale ranging from 1 (not at all strongly) to 4 (very strongly). Students who answered that they don't feel that they belong to any other group were coded not at all strongly.

Analytical procedure
We used multilevel ordered logit structural equation models for each country separately, including only youth of immigrant descent. Given the two dependent variables a structural equation model is the most appropriate specification. We implemented this specification using the gsem command in Stata 15. Due to interrelations between contexts, the models were built in a stepwise fashion. We included our three ethnic composition variables: proportion of co-ethnics, proportion of non-immigrant descent youth, and ethnic diversity. Moreover we included the following covariates at the individual level from wave 1: perceived discrimination, socio-economic status, gender and first generation as covariates. Table 1 shows the descriptive statistics for the variables and Appendix A the correlations between variables, separately for each of the four countries. Given that we are interested in how being exposed to a specific ethnic composition predicts cultural belonging, our dependent variables are from wave 2. The time points where each variable was assessed therefore also reflects the assumed direction of effects. 5 Following these first analyses, we carried out a second analysis, focussing on the number of non-immigrant descent friends, which is one of the important mechanisms that explains the relation between classroom composition and belonging. This second analysis helps to gain insight into a potential mechanism behind the relation between diversity and belonging, and it shows to what extent the proposed relationships unfold independently of friendship formation. Moreover, these results then are more easily compared to studies that also included friendships. The results of the second analysis are provided in Figure 2. Table 1 shows that the highest mainstream orientation is to be found in the Netherlands, at both wave 1 and wave 2. A stronger ethnic belonging seems not to be clustered by MP, given that the mean ethnic belonging is relatively low in Sweden. The mostly comparable means and standard deviations of our ethnic composition indicators show that possible differences between countries should not be driven by large differences in the ethnic composition of schools. Mean perceived discrimination is lowest in both Sweden and the Netherlands.

Main results
However, for a more accurate test of our hypotheses, we have to control for confounding variables and simultaneously test effects of the independent, context variables on identity outcomes. For this purpose, we conducted multilevel ordinal logistic structural equation modelling. The results of our two model are presented in Figures 1 and 2.

Normative references
The relation between the percentage of non-immigrant descent youth and mainstream belonging is significantly positive in all countries. We accordingly accept the non-immigrant descent student normative reference hypothesis (1a). Figure 2 shows that the relationship between the proportion of non-immigrant descent youth and mainstream belonging becomes non-significant in the Netherlands and United Kingdom, when we add the mechanism of non-immigrant descent friends at wave 1. Therefore, the number of nonimmigrant descent friends is a mechanism that explains the relation  between the proportion of non-immigrant descent youth and mainstream belonging. This suggests that the presence of that group not only serves as a normative reference point for immigrants and their descendants, but also as a source of friendships that have a significant bearing on their mainstream and ethnic orientation. In Figure 1, all countries except Sweden show significant positive parameter estimates for the relation between proportion of coethnics and ethnic belonging. We therefore accept the co-ethnic student normative reference hypothesis (1b) for Germany, United Kingdom, and the Netherlands, but reject it for Sweden. The relationship between the proportion coethnic students and ethnic belonging declines in all counties and becomes non-significant in the Netherlands when we control for the mechanism of non-immigrant descent friends. This again implies that the formation of friendship indeed is an important mechanism behind classroom composition and belonging. Finally, regarding our covariates Appendix B shows negative significant relations between perceived discrimination and mainstream belonging and no significant association between perceived discrimination and ethnic belonging in all countries, except the Netherlands. We found a significant positive relation between socio-economic status and mainstream belonging in Germany and Sweden. As expected, first generation students belong more strongly to their ethnic group and less strongly to the mainstream. Results for boys and girls are mainly comparable. Only in Germany girls feel that they belong to their ethnic group more strongly than boys.
Multicultural policies and colonial histories as moderators of the diversity-belonging link Next we turn to the country-level differences in these findings. Figure 1 shows significant negative associations between ethnic diversity and mainstream belonging in the Netherlands and the United Kingdom and non-significant associations in Germany and Sweden. We consequently reject both the MP buffer hypothesis (2a) and the colonial histories buffer hypothesis (2b) for ethnic diversity. In our four-country sample it rather appears that colonial histories strengthen the negative impact of ethnic diversity. It may for example be that on the classroom level there are opposing processes related to nonimmigrant descent students and their views regarding the colonial past, as the findings in Table 1 do not imply a lower overall mainstream belonging in countries with a colonial past. The results show no significant associations between ethnic diversity and ethnic belonging. While our main country-level hypotheses could not be confirmed for ethnic diversity, there are numerous country-level differences in Figure 1 that are worth exploring further. We will examine them in the corresponding discussion section below. Figure 2 shows stronger negative relations between ethnic diversity and mainstream belonging when we add the mechanism of non-immigrant descent friends at wave 1. This means that having non-immigrant descent friends partly suppresses the negative impact of ethnic diversity on mainstream belonging, but not to a large extent. This may be due to the fact that such friendships are more likely when diversity is higher ceteris paribus.

Discussion
The primary focus of our study was on classroom-level factors and potential cross-national variations. For the classroom, the distinction between the proportion of non-immigrant descent youth and ethnic diversity follows two lines of theoretical reasoning. First, following NRGT (Merton 1968), namely that immigrant descent youth in classrooms with more non-immigrant descent youth will increase their mainstream belonging, and second, CT (Putnam 2007), which posits that youth in more ethnically diverse classrooms are more likely to decline in their mainstream belonging. Our results reveal evidence in favour of NRGT in all countries and confirm CT only for mainstream belonging for the United Kingdom and the Netherlands. The evidence contrary to CT for ethnic belonging is in line with previous research in Belgium (Agirdag, van Houtte, and Van Avermaet 2011). Whereas CT states that diversity plays a role both for the social cohesion of the ingroup and outgroup, it does not play a role for the sense of belonging towards the ethnic ingroup. Therefore, both earlier and our results reveal that the CT cannot be generalized toward ethnic belonging of students. Moreover, these mixed results for mainstream belonging show that CT cannot be confirmed in all countries in European educational settings, replicating similar findings from research in European neighborhoods (Van der Meer and Tolsma 2014).
Regarding ethnic belonging, NRGT postulates that a higher proportion of co-ethnics facilitates this type of belonging. We found in our analysis empirical evidence for this line of reasoning in all countries except Sweden. However, our analysis revealed that the relationship between the proportion of non-immigrant youth and ethnic belonging suggests that the presence of non-immigrants not only has normative implications but also generates identification through friendships.
Negative relationships were apparent between perceived discrimination and mainstream belonging in all countries. This underscores the detrimental effect of discrimination for mainstream identification and social cohesion which was also found in other studies (Edele et al. 2020;Schachner, van de Vijver, and Noack 2018b), regardless of the specific national context.

Multicultural policies and colonial histories as moderators
As expected, the results for the relation between ethnic diversity and cultural belonging amongst youth of immigrant descent vary from country to country and these differences are visible for the classroom intergroup contexts. We consider how these differences can be related to differences in the country intergroup context (MP and colonial history). We will explore how the mixed findings for the proportion of co-ethnics and our individual level covariates could be related to differences in country contexts.
Given that United Kingdom and the Netherlands have a more extensive colonial history than Germany and Sweden, the cluster of comparable findings for ethnic diversity in the United Kingdom and the Netherlands suggests that the constrict hypothesis holds only for mainstream belonging for countries with a colonial history. If we also take into account the negative relationship between ethnic diversity and mainstream belonging that was found in previous research in Belgium (Agirdag, van Houtte, and Van Avermaet 2011) with its history of colonization, the colonial histories argument seems a feasible line of reasoning to explain this. The findings are contrary to the Common Ingroup Identity Model (CIIM) which posit positive or nonrelations in countries with colonial histories (Gaertner and Dovidio 2000). Therefore, it should be doubted whether a longer history of diversity and more experience in dealing with diversity automatically leads to more inclusive mainstream identities, as a colonial past could often also be connected to a longer history of structural racism and inequality. Differences in relations between countries with and without a colonial history could also be partly explained by a different make-up of ethnic diversity in countries with a colonial history. Especially in countries with a colonial past, an ethnically diverse student body could include not only young people issuing from the two main migration flowsguest workers and refugeesbut also youth with a colonial background or belonging to groups with such histories (Berry 2011).
Turning to our reflection on the proportion of co-ethnics and individual level covariates, it appears that a combination of extensive MP and an absence of a colonial histories ameliorates the group-level processes that link co-ethnic proportions to ethnic belonging in Sweden. However, given our limited possibilities on the country level, this interpretation cannot be claimed with certainty, and therefore warrants further examination.
We found relationships between perceived discrimination and ethnic belonging only in the Netherlands where higher perceived discrimination was associated with higher ethnic belonging. It could be argued that the combination of a colonial past and relatively low respectively scaled back MP makes perceived discrimination a more virulent issue, possibly resulting in reactive ethnic belonging as a coping mechanism (Rumbaut 2008).
We found a positive relation between socio-economic status and mainstream belonging in Germany and Sweden. Given that these are the two countries that share a more limited colonial history, it is likely that such a history indeed has a bearing on integration prospects (see also the comparison between the United Kingdom and Sweden by Kesler 2010). Successful socio-economic integration is more important for mainstream belonging when the immigration history of the country is less markedly defined by a long-standing colonial past. Findings in all four countries indicate that socio-economic barriers in the sense of a lower socio-economic status are met with higher levels of ethnic belonging. This suggests another kind of reactive ethnicity, which is related to structural inequality rather than subjectively perceived discrimination.

Conclusion
Empirical studies show divergent relations between classroom contexts and cultural belonging between countries, using divergent research methods and focussing on children and youth at different ages. This creates the need for comparative research efforts that provide lines of reasoning to explain these differences. MP and MH are regarded as potential explanations for differences in intergroup contexts (Castles and Miller 2009;Janmaat 2014;Suárez-Orozco et al. 2018). We therefore set out to test these issues with a set of four European countries known to differ in MP and MH. First, Sweden and the United Kingdom have stronger MP. Second, both the Netherlands and the United Kingdom are known for their more extensive colonial history. As a consequence, these four countries are well-suited for such a comparison.
We were able to show that there are differences in result patterns between countries. Yet, the relationships with the historical and policy contexts were complex and not entirely in line with our initial expectations. While their relationship with the consequences with ethnic diversity appears to be less straightforward, a number of other factors appear to be buffered by these two country-level factors or a combination of them, for example the non-significant SES effects in the two countries with a colonial history. While we cannot claim to have examined this level of analysis exhaustively, it is quite evident from these divergent results that theories such as the ones we investigated need to be regarded in context. Even though we only investigated European countries, it is very clear that there are cultural and political underpinnings to theorized processes that warrant further scholarly attention. Ideally, an investigation of such factors tests them against other structural differences between countries. This of course critically depends on there being more cross-national data collection efforts such as the CILS4EU with a larger country sample. This would also facilitate tests of interactions between such factors.
Moreover, we were able to show that the classroom context effects have both a normative but also a structural component in that the addition of friendships with non-immigrants have a bearing on these results. The classroom is not only a socialization agent in which societal norms are processed, it also is an important locus of interactions, and acknowledging both analytically is important.
Although our study is a step towards gaining knowledge on cross-country differences in the relation between classroom contexts and cultural belonging, the presented line of reasoning of such differences is based on only four countries and on two specific diversity-related factors that set these countries apart, namely their colonial past and MP. Other selection processes also differentiate the countries under investigation. Moreover, processes of cultural belonging are likely to differ between different immigrant heritage groups (Schulz and Leszczensky 2016) and these groups may not be entirely comparable between countries. For example, the foreign-born population in the United Kingdom had the highest percentage of highly educated people, with Sweden and the Netherlands occupying the middle ground and Germany having the smallest proportion with regard to this indicator (Thoreau and Liebig 2012). Similarly, there are vast differences between the four countries with regard migration motivations, such as Sweden mostly admitting immigrants for humanitarian reasons (Cangiano 2012). In sum, this sets Sweden apart as a country where open multicultural and integration policy is coupled with specific integration challenges, which may result in less potential for the effects of multiculturalism explaining the impact of diversity on mainstream belonging.
Besides this, as was just outlined, other factors at the country level might intervene here, but a broader investigation of these would necessitate a larger number of countries in order to actually test alternative factors against one another. Our study provides results of four countries that were all destination countries in the 1950s (Van Mol and De Valk 2016). This historical component is likely to influence the specific countries of origin in the ethnic makeup of classrooms in different countries. A larger number of countries with variation in periods when the countries were destination countries would provide the opportunity to search for the possible influence of this historical component. Moreover, given that we only used indicators of policies at a single time point, no claims on causality can be made, and we cannot test the influence of specific policy changes. Therefore, any implications of our findings for policy reform should be treated with caution. The divergent findings between countries underpin that policymakers who intend to influence the cultural belonging of immigrant descent youth should hesitate before affecting the ethnic composition of schools as an instrument to increase cultural belonging based on research in countries other than the one where the policy would be implemented. Moreover, there may also be within country differences. Therefore, it may be better to target other aspects of the school context, such as school diversity policies or school diversity climate (Celeste et al. 2019;Schachner et al. 2016).
The theoretical background of this study is interdisciplinary, drawing on widely used theories in sociology and social psychology. Based on these theories, this study takes a next step in showing empirically with comparable data that classrooms contexts relate differentially with cultural belonging in different countries. A focus on policies and on histories as used in comparative sociology combined with social psychological theories and models presents one avenue to approach the embeddedness of classroom-and individual-level dynamics that bridges gaps between disciplines (Pettigrew 2018). These new directions of research can provide insight in the different mechanisms concerning cultural belonging of immigrant descent youth that occur for different specific immigrant heritage groups with different intergroup experiences in different classroom-level contexts in the different country contexts, responding to calls for more specificity in acculturation research (Bornstein, 2017). Notes 1. https://www.gesis.org/sdc 2. Although our indicator of the groups refers to country of origin instead of the ethnicity of the students in our data, we used the concept of ethnic diversity and not that of country of origin diversity because the first concept is more common in sociological, psychological and educational science research on composition effects. 3. https://www.queensu.ca/mcp/ 4. Based on our aim for a cross-national comparison with a uniform research method we decided to use the same discrimination scale for all countries. Given the low alpha in Germany, no conclusions for the discrimination variable were made for this country. Finally, perceived discrimination is not of focal importance for our primary research questions and our central hypotheses. Analyses with each specific discrimination indicator can be found in Appendix C. 5. We also ran a model where we added identification at wave 1, however models with these covariates did not converge. This may be due to the little time between wave 1 and wave 2 which provides a small possibility of changes between the waves. Moreover, the prior measures were collected when students were at least a few months in that class, therefore cultural identification from the first wave could already have been influenced by classroom ethnic composition of that year.