Identity threat and identity multiplicity among minority youth: Longitudinal relations of perceived discrimination with ethnic, religious, and national identification in Germany.

The notion that ethnic and religious minority identities are inherently incompatible with the national identities of European immigrant-receiving societies is popular in public discourse. Although findings documenting such negative associations seemingly support this claim, other research shows that the intergroup context matters for the extent to which minorities' ethnic and religious identities are conflicting (i.e., negatively associated) or compatible (i.e., positively associated) with European national identities. However, previous research relied on cross-sectional data and therefore could not capture the dynamic process through which minority youth come to develop compatible or conflicting identification patterns. We extend this work with a longitudinal approach by capturing developmental trajectories of identity multiplicity among ethnic minority early adolescents in Germany over three waves with 9-month intervals. At each measurement point, participants reported their ethnic, religious, and (German) national identification and their experiences with discriminatory treatment. We estimate a cross-lagged panel model to study how identification relates to perceived discrimination and how this affects (changes in) associations between ethnic, religious, and national identification of minority youth. Our results show prevalent positive associations between ethnic, religious, and national identification across minority youth in the sample. Those who report more frequent discrimination, however, lower their (German) national identification over time, which in turn predicts increased minority identification. We conclude that identity threat indeed triggers a development of more conflicting identification patterns.

Background

Large‐scale immigration to Europe has led to increasing ethnic and religious diversity, particularly among youth. In Germany, where the present research is situated, 23% of the population has a migration background, and this percentage increases to almost 30% among 15‐ to 35‐year‐olds (Destatis, 2016). Europe's schools today thus are attended by diverse student bodies including children from many different origin countries.

During adolescence, youngsters of immigrant origin acquire a sense of ethnic identity from interactions with parents, other family members, and co‐ethnic peers (Umaña‐Taylor et al., 2014). These youngsters are also bound to develop a sense of belonging to the country in which they are raised, or have been born and spent most of their lives (Verkuyten & Martinovic, 2012). In addition, religious identity is an important facet of minority youths’ social identity as religion often is strongly entwined with migrants’ heritage culture and is considered among the core cultural elements that migrant parents aim to transmit to the next generation (Phalet, Fleischmann, & Hillekens, 2018). Although the importance of these different social identities for minority youth is widely recognized and previous work studied their development among youth (e.g., Umaña‐Taylor et al., 2014), little research examines the development of these social identities in conjunction. We therefore know little about the repercussions of development of one social identity for the identification with the others. Moreover, previous work has highlighted the importance of identity threat, and more specifically perceived discrimination, for the compatibility of minorities’ multiple social identities, but it is as yet unknown how the intergroup context impacts on the associations between ethnic, religious, and national identification during adolescence.

Against this background, this study examines (1) how ethnic, religious, and national identification are associated among early adolescents in ethnically diverse schools in Germany, (2) how these identifications relate to experiences of discrimination, and (3) whether associations between identifications change over time as a function of perceived discrimination. We improve upon earlier cross‐sectional research that could not show whether the associations between minorities’ multiple identification are stable over time or undergo significant changes (e.g., Fleischmann & Phalet, 2016). Particularly among adolescents, who are in the process of developing their identities (Meeus, 2011), taking such a snapshot approach is problematic. Yet, there are no studies that examined how different identifications are associated during adolescence and whether these associations change over time. Although individuals hold multiple social identities, most previous empirical work among minority youth focuses on the development and psychological correlates of a single social identity at the expense of other relevant identities (often ethnic or racial identity; Umaña‐Taylor et al., 2014) or, in the context of immigration, on the combination of immigrants’ origin and host national identities (e.g., Berry, Phinney, Sam, & Vedder, 2006).

Identity multiplicity: An intergroup perspective in ethnically diverse societies

We study identity multiplicity among minority youth from a social identity perspective (Reicher, Spears, & Haslam, 2010; Tajfel & Turner, 1979). This theoretical approach is concerned with the process through which individuals come to consider themselves as part of a social group (Turner, Hogg, Oakes, Reicher, & Wetherell, 1987). It also recognizes that people are members of multiple categories and groups and therefore have multiple social identities (Verkuyten, 2018). The extent to which individuals perceive themselves as members of a certain group, and how this group membership affects their behaviour vis‐à‐vis members of in‐ and outgroups, has been the topic of a large body of research that convincingly established the contextual nature of social identity (e.g., Brown, 2000). Thus, it is clear that the nature of intergroup relations (e.g., more hostile vs. more friendly) and structural characteristics of the intergroup context, that is, the stability and legitimacy of intergroup relations and the permeability of group boundaries, affect how individuals position themselves vis‐à‐vis the multiple identity options they encounter (e.g., Verkuyten & Reijerse, 2008).

An important driver of the dynamics of social identity processes is identity threat (Branscombe, Ellemers, Spears, & Doosje, 1999). Given the ethnic hierarchy in many migrant‐receiving societies that put ethnic and religious minorities at the bottom and the non‐migrant population at the top (e.g., Verkuyten, Hagendoorn, & Masson, 1996), immigrants and their offspring are susceptible to recurrent threats to the positive value of their specific ethnic or religious minority identity. Unfair or hostile treatment due to one's ethnic or religious group membership signals to minority youth that they lack recognition as full‐fledged members of society (Maliepaard, Gijsberts, & Phalet, 2015; Schulz & Leszczensky, 2016). The rejection–identification model (RIM) suggests that such experiences of discriminatory treatment will affect the level of identification with the minority ingroup. For example, in response to racial discrimination, African Americans were found to increase their identification with their racial group, and this increased group commitment buffered against the detrimental effects of discrimination on well‐being (Branscombe, Schmitt & Harvey, 1999). Among immigrants in Europe, higher levels of perceived discrimination have similarly been related to increased ethnic and religious identification (e.g., Verkuyten & Yildiz, 2007). More recently, the RIM has been extended with the notion that perceived discrimination will also lower minorities’ identification with the majority group – who is mostly conceived as perpetrator of the unfair treatment (Fleischmann & Phalet, 2016; Jasinskaja‐Lahti, Liebkind, & Solheim, 2009) – or even motivate disidentification with the national category (e.g., Maliepaard & Verkuyten, 2018; Verkuyten & Yildiz, 2007).

These opposing expectations regarding the effects of perceived discrimination on minority youth’ identification with their minority ingroup on the one hand and the majority group on the other have implications for the study of identity multiplicity. In the German case, we would expect that ethnic and religious identifications increase in response to perceived discrimination, whereas national identification should decrease. However, we are not merely interested in changes in the strength of single social identities, but in the question of how perceived discrimination affects the compatibility of minority youth’ ethnic, religious, and national identification. We examine this by studying the associations between the three identities, and how they develop in minority youth over time.

This focus on associations between identification with multiple social categories differs from the study of identity multiplicity in terms of the strength of identification with an explicit dual identity (e.g., identifying as British Muslim; cf. Fleischmann & Verkuyten, 2016).1 Our approach is similar to the acculturation literature, which conceives of minorities’ group identifications as conceptually independent from each other (Berry, 1997; Hutnik, 1991), such that multiple combinations of high and low attachment to ethnic and/or religious and host national communities are possible. Following previous comparative research across ethnic groups and cities (Fleischmann & Phalet, 2016), we describe identification patterns as conflicting when minority and national identification are negatively correlated and as compatible when they are positively correlated. Fleischmann and Phalet (2016) found the entire spectrum of identity compatibility (e.g., in Brussels), identity conflict (e.g., Amsterdam and Stockholm), and non‐significant associations (e.g., Rotterdam, Antwerp) between ethnic and religious identification on the one hand and national and city identification on the other. Such variation across localities notwithstanding, a pattern of identity conflict rather than compatibility seems to be more common in Europe than in North America (Berry et al., 2006). Given the particular problematization of religious diversity in the European context (Foner & Alba, 2008), such negative associations are even more prominent for minorities’ religious than their ethnic identities. Indeed, several studies find a negative association between Muslims’ religious identification and their identification with the European nation of residence (e.g., Maliepaard & Verkuyten, 2018) and levels of national identification are even lower among Muslim than non‐Muslim minority youth (Fleischmann & Phalet, 2018).

Many of the studies on identity compatibility using the associational approach also examined how perceived discrimination relates to identity compatibility but were limited by their cross‐sectional approach. Several studies found that ethnic and/or religious minority identification is more negatively associated with national identification among minorities in Europe the more discrimination they perceive (Fleischmann & Phalet, 2016; Kunst, Tajamal, Sam, & Ulleberg, 2012; Verkuyten & Martinovic, 2012). However, it is as yet unknown through which process these negative associations between identities emerge when levels of perceived discrimination are high. Building on the expectations above that perceived discrimination would enhance minority youth’ identification with their ethnic and religious group, but decrease their national identification, we expect that these identities are more negatively associated the more discrimination youth perceive.

A longitudinal approach to identity multiplicity

We take a longitudinal approach and capture the developmental trajectories of identification patterns among ethnic minority early adolescents. Most previous research on identity multiplicity in the immigrant context used cross‐sectional data and was unable to trace developments in the associations between minorities’ social identities over time. Longitudinal research has been called for to provide further explanations for the process through which multiple social identities relate to each other, and how these associations change over time (e.g., Fleischmann & Phalet, 2016). In the context of immigration, it has been suggested that dual ethnic–national identification could be achieved when a strong national identity is added onto an earlier developed ethnic identity (Fleischmann & Verkuyten, 2016).

Indeed, in line with Phinney's (1990) model of ethnic identity development, research on the development of ethnic or racial identity shows significant age effects on the levels and forms of identification with the ethnic or racial ingroup (e.g., Huang & Stormshak, 2011). But because previous developmental research generally focuses on a single social identity, it is as yet unknown whether identification with multiple social identities develops in parallel, or whether developments diverge for different categories of one's social identity. It could be, for instance, that the attachment to one's ethnic and religious identity is strengthened during adolescence if youngsters become increasingly involved with co‐ethnic and co‐religious peers (Maliepaard & Phalet, 2012), or that ethnic majority friends increase ethnic minority peers’ national identification (Leszczensky, Stark, Flache, & Munniksma, 2016). If national identification does not develop in the same direction or to the same extent as ethnic or religious identification over the same period, for instance in response to experiences of discrimination, this would imply a lowering of initially positive associations between ethnic or religious identities and national identity. Studies using samples with different age ranges suggest that the association between Muslims’ religious and European national identity is more positive in pre‐adolescence and early adolescence than in late adolescence (Spiegler, Güngör, & Leyendecker, 2016; Verkuyten, Thijs, & Stevens, 2012). These discrepant findings across age groups call for longitudinal research that follows youngsters throughout adolescence and allows the simultaneous study of identity development for more than one social identity. In the absence of research regarding developmental trajectories of identity multiplicity, our study takes an explorative approach with regard to the longitudinal examination of associations and investigates to what extent the associations between ethnic, religious, and national identification are stable over time. Based on the expectations derived from RIM and its extension to disidentification, however, we expect that the associations between national identification on the one hand and ethnic and religious identification on the other will become more negative due to a lowering of national identification and increases in ethnic and religious identification in response to perceived discrimination.

Data and method

Participants

We use three waves of data from the study ‘Friendship and Identity in School’ that surveyed more than 2,000 students in ethnically diverse schools in Germany (Leszczensky, Pink, & Kalter, 2015). The data were collected in the 5th, 6th, and 7th grades of nine schools in nine towns in North Rhine‐Westphalia, one of the most ethnically diverse federal states. Lower secondary, intermediate secondary, and comprehensive schools with high shares of foreign students were sampled. The participating schools were randomly chosen within pre‐defined strata regarding different shares of foreign students. The intervals between the three waves were 9 months each; the first wave was collected in May 2013 when students were about 13 years old (M = 12.8; SD = 1.1).

The starting cohorts and intervals were chosen to make sure that most participating students would not undergo major transitions in their school career and stay in the same classroom during the period of observation. Participation in the study was voluntary and required written parental approval. Researchers instructed and supervised students, who completed paper‐and‐pencil questionnaires during lessons. Students’ participation rate was 76.5% in Wave 1, 83.3% in Wave 2, and 86.6% in Wave 3 (N wave1 = 1,668, Nwave2 = 1,860, and N wave3 = 1,889). Of all 1,668 Wave 1 participants, 1,250 also participated in Waves 2 and 3; of all 1,860 Wave 2 participants, 1,582 also participated in Wave 3. A total of 410 new participants were added in Wave 2 and 199 in Wave 3. Overall, we observe 1,250 participants three times, 640 participants two times, and 387 participants at one time‐point.

Due to the sampling design, about two‐thirds of the participating students had a migration background; that is, they or at least one of their parents or grandparents was born outside Germany. Typical for school‐based research on minority populations in Europe (Dollmann, Jacob, & Kalter, 2014), the ethnic composition of these minority students is diverse. 37.5% state Turkey as their family's country of origin, 10.0% Poland, 6.4% the Russian Federation, and 6.1% Italy. No single other origin country or region makes up more than 3% of the minority sample. Most minority students were born in Germany to foreign‐born parents and thus belong to the second generation (62.5%). For first‐generation participants, the average length of stay in Germany is 9.83 years, meaning that most arrived before the start of compulsory schooling (at age 6). Students without a migration background were not of interest for our research and therefore excluded from our analysis (N wave1 = 599, N wave2 = 644, and N wave3 = 651).

Measures

Identification

All items were rated on Likert scales ranging from 1 ‘completely applies to me’ to 5 ‘does not apply to me at all’. We reversed the scales so that higher values indicate stronger identification. The exact wording of the four items for the three identities can be found in Table 1. The items measuring national identification referred to Germany, those used for ethnic identification to one's family's country of origin, and religious identification to the religion that participants indicated in the survey. The measure of national identification is equivalent for different immigrant generations, across age groups ranging from 9 to 17 years, and among native and immigrant‐origin youth (Leszczensky & Gräbs Santiago, 2015). The ethnic identification items were presented only to students who indicated that their parents and/or grandparents were born outside Germany. These students were asked to write down their family's country of origin at the top of the page containing the ethnic identification items.2 The same four items were selected to measure national, ethnic, and religious identification in a comparable manner.3

Table 1

Descriptive statistics by wave: means (SE) or percentages

	Wave 1 N = 1067	Wave 2 N = 1215	Wave 3 N = 1237
National identification
Belonging to Germany is an important part of myself	3.27 (1.31)	3.12 (1.27)	3.06 (1.26)
It bothers me if somebody speaks ill of Germany	2.76 (1.33)	2.61 (1.26)	2.52 (1.25)
Germany is dear to me	2.93 (1.26)	2.81 (1.24)	2.75 (1.23)
I feel like I am part of Germany	3.09 (1.31)	3.01 (1.32)	2.92 (1.31)
Ethnic identification
Belonging to my family's country of origin is an important part of myself	4.39 (0.94)	4.28 (1.03)	4.19 (1.09)
It bothers me is somebody speaks ill of my family's country of origin	4.37 (1.08)	4.26 (1.08)	4.17 (1.12)
My family's country of origin is dear to me	4.48 (0.85)	4.38 (0.93)	4.28 (1.03)
I feel like I am part of my family's country of origin	4.35 (0.98)	4.23 (1.03)	4.16 (1.11)
Religious identification
My religion is an important part of myself	4.13 (1.20)	4.09 (1.26)	4.07 (1.26)
It bothers me if somebody speaks ill of my religion	3.98 (1.32)	3.94 (1.34)	3.90 (1. 40)
My religion is dear to me	4.16 (1.21)	4.13 (1.23)	4.08 (1.28)
I feel like I am part of my religion	4.19 (1.16)	4.14 (1.22)	4.07 (1.27)
Perceived personal discrimination
Due to ethnic background	1.61 (0.75)	1.59 (0.72)	1.59 (0.73)
Controls
Gender: girl	48%	n.a.	n.a.
Parental SES	0.37 (0.23)	n.a.	n.a.
Migrant generation: 1st	26%	n.a.	n.a.
Migrant generation: 2nd	62%	n.a.	n.a.
Migrant generation: 3rd	12%	n.a.	n.a.
Religious group: Muslim	46%	n.a.	n.a.
Religious group: non‐religious	14%	n.a.	n.a.
Ethnic group: Turkish	38%	n.a.	n.a.
Ethnic group: Eastern European	23%	n.a.	n.a.
Ethnic group: Other European	18%	n.a.	n.a.
Ethnic group: Other Muslim	6%	n.a.	n.a.
Ethnic group: Other non‐Western	15%	n.a.	n.a.

n.a. = not applicable.

Since all control variables are time‐invariant, we only consider their Wave 1 values in the analysis.

Perceived discrimination on the grounds of ethnicity or origin country was assessed at each measurement occasion and was indicated on a frequency scale ranging from 1 ‘never’ to 4 ‘often’. Students indicated how often German‐origin youth (1) talked badly about them, (2) insult them, and (3) treat them badly or unfairly because of their family's origin country. We computed the average across the three items to construct an observed indicator of the frequency of perceived discrimination.4

Control variables

Students indicated their gender (reference: male). They further stated their religious group, being able to choose between Catholic, Protestant, Muslim, or none; they also could write down other denominations. We included dummies for Christians and Muslims, using non‐religious and other participants as reference category. Parental socio‐economic status is captured by parents’ occupation, which was coded according to the ISEI classification, using the highest parental value and normalizing the scale between 0 and 1. We further distinguish between migrant generations, differentiating the first generation (foreign‐born student, our reference category) from the second generation (at least one foreign‐born parent) and the third generation (at least one foreign‐born grandparent). Regarding ethnic origin, we distinguished between Turkish, Eastern European, other European, other Muslim, and other non‐Western groups.

Method

We conducted structural equation modelling using Mplus version 8 (Muthén & Muthén, 1998–2015). This approach and software package imputes missing values on dependent variables using full information maximum likelihood (FIML), which allows us to retain a large sample for our longitudinal analysis.5 We first assessed the equivalence of the three latent measures of identification over time. We treat national, ethnic, and religious identification as first‐order factors, which are composed of four items each. Since all twelve items were measured at three occasions, we included error correlations between each item measured at different time‐points to take into account the interdependence of these observations, which are nested within individuals over time (Little, 2013). Furthermore, because our participants are nested within school classes, we estimated a two‐level model with classes as clustering variable and using the robust estimator (MLR).6 After establishing longitudinal measurement equivalence, we estimated a cross‐lagged panel model of national, ethnic, and religious identification and perceived discrimination, all measured at three time‐points (Little, 2013). This cross‐lagged model includes both autoregressive paths (i.e., constructs predicted by the same construct at a previous measurement occasion) and all possible crossed paths between adjacent waves (i.e., lagged effects of all constructs on all other constructs at the next time‐point). We included our (time‐invariant) control variables on all concepts of interest at all three time‐points. Finally, we specified correlations between national, ethnic, and religious identification to assess the associations between these three identities. We implemented tests of equality constraints, using the Wald test of parameter estimates, to compare regression coefficients across time‐points. We also repeated the analysis for specific religious groups contrasting Muslim with Christian minority youth (these results are found in Tables S1–S3).

Results

Descriptive results

Table 1 presents the means of the identification items by wave. For ethnic and religious identification, all means are significantly above the neutral mid‐point of the scale (all p's < .001), indicating rather strong identification. The means for German national identification are closer to, and sometimes not significantly different from, the neutral mid‐point. Consistent with previous research (e.g., Berry et al., 2006), national identification among ethnic minority youth is lower than ethnic and religious identification. Nevertheless, the mean values of national identification still indicate that this identity is substantively relevant to the adolescents in our study.

Regarding change over time, Table 1 shows a slight downward trend for all items, which we further examine below when testing for measurement invariance over time. Furthermore, we observe that the means of perceived discrimination are skewed towards a low frequency of hostile treatment and do not change substantially over time. In terms of socio‐demographics, the sample is gender‐balanced (48% female) and the mean occupational status is below average. Most participants were born in Germany to foreign‐born (62%) or German‐born (12%) parents. Muslims make up slightly less than half of the sample (46%).

Longitudinal measurement invariance

We use the 12 identification items in the upper panel of Table 1 to examine the structure and fit of the latent variables national, ethnic, and religious identification, and to assess longitudinal measurement equivalence. The unconstrained model, which allows factor loadings and intercepts to vary across time‐points, has a very good fit: χ2(522 df) = 1,165.544, p < .001, CFI = .972, TLI = .967, RMSEA = .029. We subsequently introduced constraints to test increasingly strict levels of invariance over time (cf. Kline, 2005). The Wald test of parameter constraints confirms that all loadings are invariant over time (p = .485), thus providing support for metric equivalence (Kline, 2005). A similar Wald test that additionally imposes equality constraints on all 12 intercepts, however, shows a significant p‐value (p < .001).7 The result of this test implies that full scalar equivalence is not supported. We therefore tested all 12 intercepts individually to examine whether the lack of full scalar invariance was due to single items not having equivalent intercepts, or due to changes between specific waves. When examining all 12 items individually, scalar invariance could be established for each single item as long as the other intercepts were allowed to vary (all p's > .05). Because we could not identify a specific intercept or transition that caused the lack of full scalar equivalence and because the model assuming full scalar equivalence still had a very good fit to the data (χ2 (564 df) = 1297.333, p < .001, CFI = .968, TLI = .965, RMSEA = .030), we continued with the assumption of full scalar equivalence over time to keep the model as parsimonious as possible.

The (unstandardized) loadings and intercepts of the fully constrained model are shown in Table 2. All items load significantly on the relevant latent construct and the loadings are close to one, indicating close associations between the four indicators. Similarly, the intercepts are close to each other, suggesting that the three latent constructs are well identified by these four indicators and that all indicators contribute equally to the latent variable.

Table 2

Loadings and intercepts of national, ethnic, and religious identification

	Loading	Intercept
National identification
Belonging to Germany is an important part of myself	1.000	3.189
It bothers me is somebody speaks ill of Germany	1.060	2.649
Germany is dear to me	1.250	2.905
I feel like I am part of Germany	1.190	3.082
Ethnic identification
Belonging to my family's country of origin is an important part of myself	1.000	4.397
It bothers me is somebody speaks ill of my family's country of origin	0.896	4.352
My family's country of origin is dear to me	1.040	4.485
I feel like I am part of my family's country of origin	1.122	4.335
Religious identification
My religion is an important part of myself	1.000	4.127
It bothers me if somebody speaks ill of my religion	0.958	3.967
My religion is dear to me	1.047	4.149
I feel like I am part of my religion	1.012	4.163

Loadings and intercepts were constrained to be equal in our final model; hence, only one solution is shown that applies to all three waves. All loadings and intercepts are significant with p < .001. Unstandardized loadings and intercepts are shown.

The three‐wave cross‐lagged panel model

To examine longitudinal relations between perceived discrimination and identification, we estimated a cross‐lagged panel model across three measurement occasions (Little, 2013). In this model, each (latent) variable is predicted by itself at the preceding time‐point. Perceived discrimination is further assumed to predict national, ethnic, and religious identification at subsequent time‐points, and the three identifications are assumed to predict levels of perceived discrimination and the other identifications at subsequent time‐points. Figure 1 schematically depicts a simplified version of this model.

Figure 1

Schematic representation of cross‐lagged panel model of national, ethnic, and religious identification and perceived discrimination across three waves. Note: Solid lines show paths that turned out to be statistically significant. Dashed paths are included in the model, but turned out to be statistically insignificant. For readability, only latent factors of national, ethnic, and religious identification are shown and their indicator variables are left out. Similarly, the control variables, which were related to all three identities and to perceived discrimination at each time‐point, are not shown here.

We included our control variables gender, parental socio‐economic status, migrant generation, religious group, and ethnic group (reference categories are first‐generation non‐religious and non‐Turkish minority boys).8 We first estimated an unconstrained structural model and found that it has a good fit to the data: χ2 (856 df) = 1,687.938, p < .001, CFI = .963, TLI = .956, RMSEA = .029. When inspecting the autoregressive and cross‐lagged regression coefficients of this unconstrained model, we observed that paths that were estimated repeatedly (e.g., autoregressive paths between time‐point 1 and time‐point 2, and time‐point 2 and time‐point 3, respectively) had similar estimates. Therefore, we implemented a series of equality constraints to arrive at the most parsimonious model. We first tested and confirmed that autoregressive paths could be constrained to be equal across the two transitions. The Wald test for these four parameter constraints is non‐significant at p = .606. This means that the changes in national, ethnic, and religious identification and perceived discrimination are substantially the same during both 9‐month intervals that separate our three waves of panel data. Subsequently, we tested whether cross‐lagged paths could be constrained to be the same over time, and also found support for this assumption. For these 12 parameter constraints, the Wald test also yields a non‐significant outcome of p = .125. This implies that the relationships between identification and discrimination are stable over time and do not change with increasing age or due to period effects during the data collection.9 Based on the results of these tests, our final model assumes that all autoregressive and cross‐lagged paths are stable (i.e., they are constrained to be equal at both transitions). This model fits the data well: χ2 (872 df) = 1,701.513, p < .001, CFI = .963, TLI = .957, RMSEA = .029.

Associations between national, ethnic, and religious identification

We first examined the cross‐sectional associations between national, ethnic, and religious identification at each time‐point. The correlations (i.e., standardized covariances) from the final model (including incoming paths from perceived discrimination and controls) are shown in Table 3. Given our modelling strategy, these associations represent the average correlations across all participants, whereas the cross‐lagged paths that we present below capture changes in trajectories of identification. Cross‐sectionally, national identification is weakly positively correlated with both ethnic and religious identification at the first time‐point, while ethnic and religious identification are more strongly positively correlated. The correlations change in magnitude over time10 and, as a result, the weak correlations of national with ethnic and religious identification are not significant in each wave. Ethnic and religious identification are always significantly positively related, but the magnitude of the association also decreases over time. Given these associations, ethnic and religious identification are on average strongly compatible among minority youth in Germany, and national identification is on average mostly compatible, and never conflicting, with ethnic and religious identification. Controlling for lagged effects of perceived discrimination and socio‐demographic composition, we thus do not observe a pattern of identity conflict where identifying with German national identity comes at the cost of minorities’ attachment to their ethnic or religious group identities. To investigate the psychological process behind these patterns of associations, we now turn to the relations between identification and perceived discrimination over time.

Table 3

Correlations between national, ethnic, and religious identification at the three time‐points

	National & Ethnic identification	National & Religious identification	Ethnic & Religious identification
Wave 1	0.138**	0.131**	0.443***
Wave 2	0.098n.s.	0.227***	0.292***
Wave 3	0.167**	0.077n.s.	0.216***

n.s. = non‐significant.

**p < .01; ***p < .001.

Correlations are calculated as standardized covariances between the three first‐order factors. These correlations are controlled for autoregressive and cross‐lagged paths, as well as the effects of the control variables as shown in Table 4.

Table 4

Parameter estimates (unstandardized coefficients and SEs) of cross‐lagged panel model

	National identification			Ethnic identification			Religious identification			Perceived discrimination
	Wave 1	Wave 2	Wave 3	Wave 1	Wave 2	Wave 3	Wave 1	Wave 2	Wave 3	Wave 1	Wave 2	Wave 3
National identification	n.a.	0.658 (.024)	0.658 (.024)	n.a.	−0.041 (.021)	−0.041 (.021)	n.a.	−0.063 (.025)	−0.063 (.025)	n.a.	−0.009 (.023)	−0.009 (.023)
Ethnic identification	n.a.	0.002 (.034)	0.002 (.034)	n.a.	0.761 (.040)	0.761 (.040)	n.a.	0.004 (.048)	0.004 (.048)	n.a.	0.026 (.025)	0.026 (.025)
Religious identification	n.a.	0.024 (.020)	0.024 (.020)	n.a.	0.043 (.022)	0.043 (.022)	n.a.	0.694 (.040)	0.694 (.040)	n.a.	0.043 (.022)	0.043 (.022)
Perceived discrimination	n.a.	−0.071 (.022)	−0.071 (.022)	n.a.	−0.024 (.020)	−0.024 (.020)	n.a.	0.016 (.030)	0.016 (.030)	n.a.	0.484 (.030)	0.484 (.030)
Second generation	0.225 (.060)	0.109 (.057)	0.064 (.058)	−0.174 (.057)	−0.085 (.047)	−0.092 (.050)	−0.210 (.086)	−0.124 (.059)	0.004 (.054)	−0.101 (.068)	−0.042 (.057)	−0.075 (.052)
Third generation	0.430 (.090)	0.227 (.088)	0.119 (.079)	−0.407 (.120)	−0.309 (.118)	−0.258 (.087)	−0.822 (.133)	−0.314 (.122)	−0.130 (.118)	−0.292 (.093)	−0.106 (.079)	−0.201 (.089)
Girl	0.027 (.056)	−0.037 (.050)	−0.031 (.050)	0.019 (.055)	−0.008 (.043)	0.104 (.043)	0.044 (.066)	0.023 (.052)	−0.003 (.054)	−0.162 (.048)	−0.054 (.042)	−0.094 (.041)
ISEI	−0.001 (.001)	0.000 (.001)	0.002 (.001)	−0.004 (.001)	0.000 (.001)	−0.002 (.001)	−0.003 (.002)	0.001 (.001)	−0.001 (.002)	−0.001 (.001)	−0.001 (.001)	0.000 (.001)
Christian	0.123 (.086)	−047 (.074)	−0.065 (.058)	0.246 (.080)	−0.057 (.054)	−0.028 (.063)	0.561 (.142)	−0.005 (.113)	−0.141 (.122)	0.087 (.071)	−0.063 (.070)	0.005 (.065)
Muslim	0.058 (.101)	0.017 (.092)	−0.080 (.079)	0.651 (.106)	0.007 (.078)	0.184 (.072)	1.577 (.142)	0.418 (.109)	0.379 (.139)	0.151 (.087)	−0.101 (.079)	−0.061 (.081)
Turkish	−0.164 (.090)	−0.093 (.079)	0.014 (.071)	−0.130 (.080)	0.070 (.055)	−0.087 (.054)	0.041 (.083)	0.086 (.066)	0.043 (.070)	−0.064 (.079)	0.119 (.066)	0.011 (.067)

Coefficients that are printed in bold font are significant at p < .05 or higher. Autoregressive paths and cross‐lagged paths between identifications and discrimination were constrained to be equal over time, resulting in identical parameter estimates for Waves 2 and 3 in the first four lines of this table.

The interplay of identifications and perceived discrimination

Table 4 presents the parameter estimates of our final cross‐lagged model. All three identifications show substantial autoregressive paths, meaning that those who start with higher levels of national, ethnic, and religious identification than their peers, respectively, are also likely to score relatively high on these measures during later time‐points. Perceptions of discrimination also show significant stability over time, but the autoregressive path is considerably smaller, indicating that there is more variation over time in the extent to which youth experience discrimination. There are few statistically significant cross‐lagged paths between the three identifications, which implies that, by and large, change in identification with one category was not related to change in another category at a later time‐point. However, in line with previous cross‐sectional research regarding the associations between national, ethnic, and religious identification (Fleischmann & Phalet, 2016), there was a tendency for higher religious identifiers to increase their ethnic identification over time, while higher national identifiers were more likely to decrease their religious identification over time.11 In terms of patterns of associations, this implies that minorities’ ethnic and religious identifications are rather compatible as they reinforce each other over time (even though the reinforcement seems to be one‐sided and not mutual as the path from ethnic to religious identification is not statistically significant). National and religious identification, by contrast, are more conflicting as an increase in one goes together with a decrease in the other. Finally, national and ethnic identification have no significant cross‐lagged associations, once we take their associations with religious identification into account.

Regarding the relations between perceived discrimination and identification, consistent with the notion of rejection–disidentification, higher levels of perceived ethnic discrimination predicted lower levels of national identification 9 months later. In contrast, the reverse pathway from national identification to perceived discrimination is not statistically significant.12 Neither ethnic nor religious identification is predicted by prior experiences of ethnic discrimination, nor do they in turn predict perceived discrimination; thus, the RIM receives little direct support in our data. In line with our expectation about the effects of perceived discrimination on the associations between national, ethnic, and religious identification, higher levels of discrimination are related to more negative associations between national and minority identifications. This is due to the direct negative path from discrimination to national identification, and the subsequent negative path from national to religious identification, which implies an indirect positive effect of perceived discrimination on religious identification. This specific indirect path (from perceived discrimination at t1 to religious identification at t3 via national identification at t2) is marginally significant at p = .072. It provides partial confirmation for our expectation that national and minority identities are less positively associated as a consequence of perceived discrimination.13 Perceptions of discrimination are thus related to lower national but higher religious identification in subsequent waves. In line with our expectation, a pattern of identity conflict between national identification on the one hand and religious identification on the other therefore indeed is more likely to occur among participants who more frequently perceived discriminatory treatment.

Socio‐demographic differences

Regarding the control variables, we first describe the relations with variables of interest in Wave 1 as these capture mean differences at the first measurement occasion. Since we control for prior values of identification and discrimination, the effects of the controls at later time‐points reveal the effects of socio‐demographic characteristics on changes in identification and discrimination rather than absolute differences. Table 4 shows that national identification increases with each immigrant generation, while both ethnic and religious identification decrease.14 The third generation perceives significantly less discrimination than the first; this trend is already apparent in the second generation, but the difference with the first fails to reach significance. The only significant gender difference in the starting values of the variables of interest is that girls reported less discrimination than boys. Participants with higher socio‐economic status have lower ethnic and religious identification at the start. Finally, compared to non‐religious participants, Christian and Muslim participants identify more strongly with their religion and their ethnic group. We find no significant effects of being Turkish in our final model. Once socio‐demographic differences at the first time‐point are taken into account, there are only few differences in the changes of national, ethnic, and religious identification and perceived discrimination over time according to immigrant generation, gender, socio‐economic status, religion, or ethnicity. Where such effects are significant, they go in the same direction as the effects found in Wave 1 and thus further exacerbate the compositional differences at baseline.

Discussion

This study longitudinally examined the associations between ethnic, religious, and national identification among minority youth in Germany, as well as the effects of perceived discrimination on these three identifications and their associations. Concerning the development of compatibility of minorities’ multiple social identities, our longitudinal findings provide further evidence for the detrimental role of identity threat, more specifically perceived discrimination. Across the entire sample, German national identification was either weakly positively associated with, or unrelated to, ethnic and religious minority identification, indicating that identity compatibility prevailed among these early adolescents and national identification was never conflicting with ethnic and religious identifications. However, we found significant variation in the development of the three identifications in response to perceived ethnic discrimination. In line with our theoretical expectations derived from the recent extension of the RIM to identification with the majority group, adolescents who more frequently perceived discrimination due to their ethnicity decreased their German national identification over time. Lower national identification, in turn, was related to increases in minority identifications. Our additional analyses show that among Christian (and non‐religious) but not among Muslim youth, higher national identification was in turn related to lower religious identification. Moreover, when cross‐lagged paths were allowed to vary over time, we found that national identification at the second wave also was related to lower ethnic identification in the third wave in addition to its parallel negative association with religious identification. Compared to their peers with lower levels of perceived discrimination, youngsters who report more frequent discriminatory experiences therefore developed a more conflicting pattern of identification.

Regarding change over time in patterns of identification, we found only modest variation across our three waves of panel data, which covered a period of one and a half years. Measurement invariance tests confirmed that, despite a slight downward trend in the mean levels of all three identifications, the structure and meaning of national, ethnic, and religious identification are rather stable across the period of observation. On the one hand, this finding limits our possibility to empirically examine the causes for changes in identification patterns in this sample of ethnic minority youth in Germany. On the other hand, it is substantively important as it suggests that the levels of and associations between three important and distinct identifications do not change dramatically during early adolescence. We have to be careful, though, to draw general conclusions from this first longitudinal evidence regarding identification patterns. It would be premature to declare that cross‐sectional approaches to identification patterns are non‐problematic based on the findings of this study, because the stability, as well as the finding of prevalently non‐significant and positive associations between identifications, might be related to the specific age group under study. Other research also found more positive associations between heritage and host culture identities in early adolescence (Spiegler et al., 2016), as opposed to more frequent negative associations in mid‐adolescence to late adolescence (Verkuyten et al., 2012) and early adulthood (Fleischmann & Phalet, 2016). A relevant avenue for future research therefore would be to take a longer time frame and examine at what ages initially positive associations between minorities’ identifications turn into negative associations, and why.

The national context of our study might also be related to discrepancies with earlier findings. Our study (as the one by Spiegler et al., 2016) was conducted in the German state of North Rhine‐Westphalia. The studies by Verkuyten et al. (2012) and Fleischmann and Phalet (2016) were situated in other European countries. Yet it is unlikely that the German context provides a more friendly intergroup climate as comparative studies that include Germany along with other European immigrant destinations reveal it to be among the least inclusive in terms of minorities’ identification with the nation (Fleischmann & Phalet, 2018). Similarly, Martiny et al. (2017) also found ethnic identification of Turkish‐origin mid‐adolescents to be negatively related to their German national identification.

The ethnic and religious heterogeneity of our sample is both a strength and a weakness. It is a strength because it more accurately reflects the reality of an increasing amount of today's classrooms in European immigrant‐receiving societies (Dollmann et al., 2014) than studies focusing on a single minority group, which necessarily exclude significant proportions of the relevant population. On the other hand, we know that acculturation processes differ between minority groups as a function of their migration history, the perceived cultural distance from the majority, and the majority's reaction to them, among others (cf. Berry et al., 2006). But given our sample size and the complexity of our model, it was not feasible to analyse ethnic groups separately, though we replicate our findings among different religious groups (see the Supporting Information). Similarly, we were unable to analyse migrant generations separately, not only due to the small size of the first generation, but also because the heterogeneity of our minority sample means that generational status, ethnicity, and religious affiliation often substantially overlap. For instance, more established Turkish minorities are more likely to belong to the second and third generation and are primarily Muslims, whereas more recent migrant groups from Poland and the former Soviet Union are primarily Christians. We tried to tease out the effects of these compositional differences by controlling for generational status, religious affiliation, and ethnic origin, but our controls are not able to capture the entire diversity of our sample. Studies that include larger samples of specific minority groups are better suited to examine whether the patterns of identification we find apply in a similar way to different minority groups.

Moreover, we were unable to incorporate multiple dimensions of social identity and of identity threat in our research. We agree with previous authors that future research needs to go beyond single‐item and unidimensional approaches to identification when studying minorities’ identity multiplicity (Fleischmann & Verkuyten, 2016; Wiley & Deaux, 2010). Such an approach could yield new insights into the ways in which specific dimensions of specific identities are related to the same and other dimensions of other identities. Given the lack of theorizing in this field and data limitations that did not allow us to model several dimensions of identification for all three identities, it was beyond the scope of this study to address this question. But it would be interesting for future research to examine how different dimensions of ethnic minorities’ multiple identities are related, and which ones are more easy and which ones more difficult to combine (cf. Verkuyten & Martinovic, 2012).

A similar argument can be made for various forms of identity threat. We focused on perceived discrimination in this study as it has previously been related to identification with minority identities (Branscombe, Ellemers, et al., 1999) as well as distancing from national identities (Jasinskaja‐Lahti et al., 2009). Low public regard as an alternative measure of identity threat would be interesting to take into account as well, not only as additional predictor of identification, but also in interaction with perceived discrimination. Previous research found that low public regard can buffer the negative effects of perceived discrimination on well‐being (e.g., Sellers, Copeland‐Linder, Martin, & L'Heureux Lewis, 2006). Thus, it would be interesting to examine if such an interaction also occurs with respect to the development of multiple identification. Although our data do contain indicators of public regard for participants’ ethnic and religious groups, the limited variation in identification precluded us from reliably estimating such interactions in our sample.

Conclusion

To conclude, our longitudinal study showed that the ethnic, religious, and national identification of minority youth in Germany are either unrelated or positively related, and these associations do not change substantially during early adolescence. A development towards increasing identity conflict occurred in response to more frequently perceived discrimination, which was related to lower national identification, and, in turn, higher religious (and ethnic) identification. Our findings therefore add longitudinal evidence for the notion that conflict between minorities’ social identities results from individuals’ perceptions of hostile intergroup relations.

Table S1. Parameter estimates (unstandardized coefficients and SE's) of cross‐lagged panel model, Muslim participants only (N = 516).

Table S2. Parameter estimates (unstandardized coefficients and SE's) of cross‐lagged panel model, Christian participants only (N = 510).

Table S3. Parameter estimates (unstandardized coefficients and SE's) of cross‐lagged panel model, Christian and non‐religious participants (N = 686).

Table S4. Parameter estimates (unstandardized coefficients and SE's) of cross‐lagged panel model without stationarity.

Click here for additional data file.