Classroom identification in ethnic minority and majority students: Effects of relationships and ethnic composition.

BACKGROUND: Many studies have shown that school belonging is crucial for students' school adjustment, but the construct has been operationalized in different ways. Moreover, most research has focused on adolescents and not compared its antecedents for ethnic minority versus majority students. AIMS: Based on Goodenow and Grady's (1993) seminal paper, we examined classroom identification as a central aspect of school belonging in minority and majority preadolescents, and predicted it from relationships with peers and teachers, taking into account classroom ethnic composition and perceived multicultural teaching. SAMPLE: Participants were 485 grade 4-6 students from 39 classrooms in Dutch primary schools. Of these children, 68 had a Turkish background, 72 had a Moroccan background, and 345 had a native Dutch background.
METHODS: Participants completed questionnaires at two waves (4.5 months apart). We used self-reports to measure classroom identification at both waves, and student-teacher relationship closeness and conflict, multicultural teaching, and peer friendship and rejection at Wave 1. We conducted multilevel analyses to predict classroom identification at Wave 2, while controlling for classroom identification at Wave 1.
RESULTS: Children of all ethnicities reported more classroom identification over time if they were less rejected by their peers and had more co-ethnic classmates. For minority children, both closeness and conflict with the teacher predicted less identification, but the effect of conflict appeared to result from their ethnic underrepresentation in the classroom.
CONCLUSIONS: Negative peer relationships can undermine classroom identification, and the student-teacher relationship has special significance for ethnic minority students.

Background

There is ample evidence that students’ school belonging has important implications for their emotional and academic adjustment (see Allen, Kern, Vella‐Brodrick, Hattie, & Waters, 2016; Osterman, 2000). A sense of connection to the school community can satisfy students’ fundamental need for relatedness, and thereby improve their psychological well‐being (Baumeister & Leary, 1995) but also stimulate the internalization of academic values and thus, the development of self‐determined academic motivation (Ryan & Deci, 2000; Vansteenkiste, Lens, & Deci, 2006).

The literature on school belonging is large and informative but limited in a few respects. First, there are variations in the exact definitions and operationalizations of the concept. For example, school belonging has been regarded as a subcomponent of school identification (Voelkl, 2012), but the terms are also sometimes used interchangeably (e.g., Allen et al., 2016). Likewise, although relationship experiences at school are typically considered an integral part of school belonging (Goodenow & Grady's 1993), researchers have examined students’ relationship perceptions as antecedents of it (Allen et al., 2016; Voelkl, 2012). Second, much of the work on school belonging has examined specific ethnic minority groups and thereby deepened the understanding of existing achievement gaps (Becker & Luthar, 2002; Booker, 2006) which is important as societies throughout the world are becoming more ethnically diverse (see Thijs & Verkuyten, 2014). Still, very few studies have explicitly compared the antecedents of school belonging for ethnic minority and majority students. For example, of the 51 studies included in a recent meta‐analysis by Allen et al. (2016), only one examined the moderating role of ethnicity. Third, most research has focused on adolescence, presumably because there is a considerable drop in school belonging during that developmental period. Still, school belonging tends to have rather strong rank‐order stability (Anderman, 2003), which makes it important to study individual differences at younger ages.

The present research addressed these limitations. We conducted a short‐term longitudinal study (two waves, 4.5 months apart) on preadolescent students (aged 9–13 years) from different primary school classes (grades 4–6) in the Netherlands. Specifically, our goal was to contribute to the understanding of ethnic group differences in school belonging by examining the link between children's classroom identification and their relationships with classmates and teachers, and by comparing these links for children of Turkish and Moroccan versus native Dutch descent. Turks and Moroccans are the largest non‐western ethnic groups in the Netherlands, and they tend to have low socioeconomic status and experience a relatively high degree of prejudice (Gijsberts, Huijnk, & Dagevos, 2012). Compared to ethnic Dutch students, Turkish‐Dutch and Moroccan‐Dutch students tend to underperform on a variety of academic indicators, such as primary school standardized test scores, enrolment in academic versus vocational tracks in secondary school, and overall educational attainment (Gijsberts et al., 2012; van de Werfhorst & van Tubergen, 2007). Our hypotheses will be described in the following, and they are summarized in Figure 1.

Figure 1

Model of expected positive (solid lines) and negative effects (dashed lines).

Classroom identification, peers, and teachers

In their seminal paper, Goodenow and Grady's (1993) equalled school belonging to a ‘psychological sense of school membership’ (p. 61). Thus, it implies a sense of identification with the school as a social group to which students have an emotional attachment (Reynolds, Lee, Turner, Bromhead, & Subasic, 2017). Remarkably, however, this sense of identification is absent from Goodenow and Grady's (1993) well‐cited definition of school belonging as ‘the extent to which students feel personally accepted, respected, included, and supported by others in the school social environment’ (p. 80). Following other researchers (e.g., Miller et al., 2017), we examined classroom identification as a core aspect of children's school belonging. The classroom is the most central aspect of students’ school environment, as they spend much time with classmates and often have the same classmates over the years. Research attests to the social and psychological importance of classmates for (pre)adolescents. For example, evidence indicates that students primarily form positive and negative peer relations within classes rather than between them (Stark, Leszczensky, & Pink, 2017) and that classroom peer groups serve a significant social reference function (Thijs & Verkuyten, 2013). From a social identity perspective (Reynolds et al., 2017), classroom identification can be defined as the cognitive and emotional significance of the classroom collective for the individual student. It can have several advantages, as it may fulfil the need for relatedness (Baumeister & Leary, 1995), but also other identity needs such as those for self‐esteem, efficacy, meaning, and distinctiveness (Vignoles, 2011). Additionally and consequentially, classroom identification may promote students’ self‐determined academic motivation (Ryan & Deci, 2000) and, ultimately, their academic achievement (Reynolds et al., 2017).

Although school involvement involves both classroom identification and social relations, these concepts are distinct. The former reflects a sense of membership, whereas the latter reflects the positive and negative relationships formed at school. According to the rejection‐disidentification model (Jasinskaja‐Lahti, Liebkind, & Solheim, 2009) – which fits well with the social identity perspective – people's identification with a group depends (respectively, negatively and positively) on the degree to which the group rejects or accepts them. To our knowledge, no research has examined classroom relationships as predictors of classroom identification, but it seems reasonable to expect comparable effects. As classmates are the very embodiment of the classroom, children's identification is presumably tied to their peer relations (cf., Jasinskaja‐Lahti et al., 2009). However, teachers can be expected to be important for children's sense of classroom identification as well. Research has shown that they can impact students’ sense of peer community through their managing of the peer social dynamics in the classroom, but also have a direct effect on children's school bonding (Gest, Madill, Zadzora, Miller, & Rodkin, 2014). Particularly when children have one or two teachers throughout the school year – which is typical for Dutch primary schools – teachers should be considered as members of the classroom collective, and in the present study, we used both peer and teacher relationships to predict children's classroom identification.

Ethnic group differences

The academic risk (or vulnerability) hypothesis (Hamre & Pianta, 2001, 2005; Roorda, Koomen, Spilt, & Oort, 2011) states that relationships with teachers are more important for students at higher risk of academic underachievement. The idea is that such students have a greater need for teacher support and should therefore be more strongly affected by student–teacher relationship quality. Risk factors identified in the literature include low SES, problem behaviour, and – given the documented achievement gaps between different ethnic groups – ethnic minority status. With respect to ethnicity, the empirical evidence for the academic risk hypothesis is limited. Although some studies report student–teacher relationship quality to be more important for the education outcomes of ethnic minority students (e.g., den Brok, van Tartwijk, Wubbels, & Veldman, 2010; Burchinal, Peisner‐Feinberg, Pianta, & Howes, 2002), meta‐analytic research has found no overall stronger effect on academic motivation and achievement in samples with more ethnic minority students (Roorda et al., 2011). Additionally, there is some indication that the student–teacher relationship can be more important for the school liking of specific minority groups (Murray, Waas, & Murray, 2008). The risk associated with ethnic minority status is not always clear and is often confounded with other factors such as low SES (Motti‐Stefanidi & Masten, 2013) and ethnic peer victimization (Verkuyten & Thijs, 2002). However, from an acculturation perspective, the risk may lie in the (experienced) cultural distance between the school and the home environment. Most schools are institutions of the dominant culture and implicitly assume the behaviours, values, knowledge, and customs of the majority group (Chiu, Pong, Mori, & Chow, 2012; Vedder & Horenczyk, 2006). Although there is ethnic diversity among teaching staff, teachers typically belong to the ethnic or cultural majority group (Thijs & Verkuyten, 2012). Minority students may therefore need support from their teachers to bridge the subsequent gap between their school and home environment.

The relative importance of teachers can also depend on students’ specific cultural backgrounds. Cultures differ in a myriad of ways, but one important comparison dimension is that of power distance, which refers to the existence and acceptance of differences in power and authority. Despite their mutual differences, both Turkish and Moroccan cultures can be considered relatively power‐distant (Hofstede, 1991). Hence, compared to their native Dutch peers, Turkish‐Dutch and Moroccan‐Dutch children are likely more oriented towards authority figures such as parents and teachers (Pels, Nijsten, Oosterwegel, & Vollebergh, 2006; Thijs, 2011). These differences in power distance do not necessarily affect the quality of the student–teacher relationship, as research has shown that the different groups of children do not differ in their perceptions of it (Thijs & Verkuyten, 2012). Yet they do lead to our expectation that the student–teacher relationship is more important for the classroom identification of Turkish‐Dutch and Moroccan‐Dutch children (Hypothesis 1). As peers are neither authority figures nor representatives of the school as a cultural institution, we did not anticipate comparable (i.e., stronger) effects of peer relations on ethnic minority students’ classroom identification.

Positive versus negative relations

Classroom relations can develop in various directions, and in the present study, we investigated the unique contributions of both their positive and negative aspects. We examined two aspects of children's perceived student–teacher relationships: closeness and conflict. Closeness involves the degree of warmth and open communication between student and teacher, and, in close relationships, children know that they can rely on their teacher for emotional support. By contrast, conflict involves the experience of mutually negative feelings and strenuous interactions (Pianta, Hamre, & Stuhlman, 2003; Verschueren & Koomen, 2012). In regard to children's peer relations, we examined both positive and negative nominations they received from their classmates to measure friendship and rejection, (Newcomb, Bukowski, & Pattee, 1993). Although these positive and negative relationship characteristics are inversely related, they are sufficiently independent to include them as separate predictors (see, for example, Koomen & Jellesma, 2015; for the student–teacher relationship, and Dijkstra, Lindenberg, & Veenstra, 2007, for peer relations).

In their seminal research review ‘Bad is stronger than good’, Baumeister, Bratslavsky, Finkenauer, and Vohs (2001) convincingly showed that negative situations have stronger and more long‐lasting effects than positive ones. In various domains of life, there is a basic tendency to be more strongly affected by negative events, and, according to the authors, this tendency can be considered as evolutionary adaptive: ‘Survival requires urgent attention to possible bad outcomes, but it is less urgent with regard to good ones’ (Baumeister et al., 2001, p. 323–370). Based on this reasoning, we expected to find, regardless of their ethnicity, that children's classroom identification would more strongly depend on the negative aspects of the relations with their peers and teacher than on the positive ones (Hypothesis 2).

Other classroom factors: Multicultural teaching and ethnic composition

For ethnic minority children, classroom identification could also be affected by their perceptions of multicultural teaching practices (Hypothesis 3). Teachers can vary in the extent to which they positively address cultural diversity and actively condemn prejudice and discrimination in their classrooms, and research has shown that this variation is associated with higher ethnic self‐esteem among minorities and positive interethnic relations (for a review, see Verkuyten & Thijs, 2013). Multicultural teaching stresses the value of different cultures and the importance of equality. Teachers can acknowledge minority students’ ethnic backgrounds and make them feel more welcome in their classroom so they are more likely to form psychological attachments to it.

Additionally, it is important to consider children's ethnic representation among their classmates (see Benner & Crosnoe, 2010). According to self‐categorization theory (Turner, Hogg, Oakes, Reicher, & Wetherell, 1987), the likelihood that people will categorize themselves and others as group members rather than unique individuals partly depends on within‐group similarity relative to between‐group similarity. A strong presence of co‐ethnic peers makes the class more homogenous and more similar to the self, and this similarity to their classmates can be expected to increase children's identification with the classroom (Hypothesis 4). Theoretically, this effect could be stronger for ethnic minority children (Hypothesis 5) as their ethnic identity tends to be more salient and important for them than for their majority peers (e.g., Verkuyten & Thijs, 2004). It can also be anticipated that the ethnic composition of the classroom moderates the effects of the other predictors but differently so for the teacher variables and the peer relationships. For children with relatively many co‐ethnic classmates, it is probably easier to ‘fit in’, and this could make the relationship with their teacher and their teacher's multicultural teaching less important for their classroom identification (Hypothesis 6). At the same time, the classroom identification of these children is probably more dependent on their peer relations (Hypothesis 7), because their peers are more likely to function as social reference group for them (see Turner et al., 1987). Indeed, earlier research has shown that children who belong to the numerical ethnic majority group in their classrooms are more strongly affected by negative peer experiences (Graham, Bellmore, & Nishina, 2009).

Method

Participants and procedure

Participants were part of a larger sample containing 888 children (aged 9–13 years) from 39 classrooms (grades 4–6) in 18 regular elementary schools in different parts of the Netherlands. Classrooms varied in ethnic diversity, with the proportion of Dutch students ranging from 0% to 100% (M = 43%, SD = 29%). Data were collected in two waves: once at halfway through the school year (January‐March) and again at the end of the school year (June–July). Passive parental consent was obtained and children participated voluntarily. At each wave, students completed questionnaires in their classrooms. All of the measures described below were assessed during the first wave, and classroom identification was assessed during both waves.

For the present study, we selected children with a Moroccan, Turkish, or native Dutch background based on children's ethnic self‐labelling and their descriptions of their parents’ ethnicity. Initially, the selected sample consisted of 526 respondents. However, we used listwise deletion in the main analyses as there were some missing values on the Wave 1 variables (<3.05%), as well as several selected children (1.3%) who did not report classroom identification at Wave 2. Little's MCAR test indicated that missingness was completely at random, χ2 (191) = 219.73, p = .08. Thus, the final sample consisted of 68 Turkish‐Dutch children, 72 Moroccan‐Dutch children, and 345 native Dutch children. This final sample included children from all 39 classrooms in the original sample from 18 schools. The ages of these 485 children ranged from 9 to 13 years old (M = 10.55, SD = 0.97), and 50.7% were girls. Most ethnic minority (i.e., Turkish or Moroccan‐Dutch) students were born in the Netherlands (92.9%), but for 78.4% of them, both parents were born abroad. All but two of the participants’ teachers self‐reported their ethnic background. Each of them self‐identified as Dutch.

Measures

Classroom identification

Students responded to two questions about their sense of classroom identification: ‘Are you proud of your class?’ and ‘Are you happy to be in your class’. Responses were scaled from 0 (No!) to 4 (Yes!). Following other researchers (e.g., Reynolds et al., 2017), these items were adapted from a previous research on group identification in children (Sierksma, Thijs, & Verkuyten, 2014). These items were significantly correlated at each wave, r 1  = .72, t 1(483) = 22.76, p < .001; r 2  = .74, t 2(483) = 24.06, p < .001. The omega coefficients were 0.6 at waves 1 (95% CI [0.51, 0.79]) and 2 (95% CI [0.53, 0.81)].

Student–teacher relationship

To assess students’ perceptions of their relationship with their teachers, they completed the closeness and conflict subscales of the Student Perception of Relationship with Teacher Scale (SPRTS; Koomen & Jellesma, 2015). Closeness was measured with six items (including ‘I feel comfortable with my teacher’ and ‘if I have a problem I can go to my teacher’) for which the omega coefficient was 0.82 (95% CI [0.79, 0.85]). The conflict subscale comprised six items (e.g., ‘I feel my teacher doesn't trust me’ and ‘I can be very angry with my teacher’), and the omega coefficient was 0.8 (95% CI [0.76, 0.84]). Students rated items from 0 (No, definitely not!) to 4 (Yes, definitely!).

Measurement invariance between the Dutch, Turkish, and Moroccan ethnic groups was tested using the semTools package in R (semTools Contributors, 2016), which simultaneously runs several multiple‐group confirmatory factor analysis models to test different levels of invariance. These analyses included the selected sample of 485 participants. Table 1 lists the fit indices for the models testing configural, metric, and scalar invariance, all of which used the same two‐factor structure including the closeness and conflict subscales. The different sample sizes for each group limit the interpretability of the chi‐square values for these models, but they can be compared using the CFI and RMSEA values (van de Schoot, Lugtig, & Hox, 2012). The fact that the changes in CFI values are <.01 and the RMSEA values for each model fall within on another's confidence intervals is evidence of metric and scalar invariance (Timmons, 2010).

Table 1

Model fit indices for confirmatory factor analyses testing measurement invariance

Model	df	RMSEA	TLI	CFI	ΔCFI	χ2	Δχ2	p
1. Configural	159	.06	.94	.95		262.33
2. Metric	179	.06	.95	.95	.000	281.31	18.98	.52
3. Scalar	199	.06	.95	.95	.003	306.80	25.49	.18

CFI = comparative fit index; df = degrees of freedom; RMSEA = root mean square error of approximation; TLI = Tucker–Lewis index.

Peer relations

Students’ friendship with and rejection by their peers was measured using peer nominations. Single‐item peer nomination measures can be highly reliable because they aggregate information from various reporters (Marks, Babcock, Cillessen, & Crick, 2013). Coie, Dodge, and Coppotelli (1982) noted the importance of distinguishing independent positive and negative peer experiences using sociometric measures in order to avoid confounding status with separate reports of acceptance or rejection. For example, positive status or popularity entails not only high levels of liking, but also low levels of disliking among peers (Coie et al., 1982, p. 558). The present study focuses on the independent effects of positive and negative peer relations measured using classmates’ nominations of friendship and rejection.

Participants nominated their classmates for each of the following categories: ‘best friends’ in the class and classmates next to whom they ‘wouldn't like to sit’. Students listed up to ten classmates for each category. Consistent with traditional practices using peer nominations (e.g., Madill, Gest, & Rodkin, 2014), proportion scores were then calculated for each student for every category. Friendship scores were calculated by dividing the number of ‘best friend’ nominations students received by the number of participants in their classrooms (minus one). Rejection scores were then similarly calculated using peer nominations for classmates next to whom they ‘wouldn't like to sit’.

Perceived multicultural teaching

Students’ perceptions of the multicultural teaching by their teacher were measured with three items that have been successfully used in several Dutch studies (see Verkuyten & Thijs, 2013): ‘Does your teacher ever say that all cultures should be respected?’, ‘Does your teacher ever say that it is wrong to discriminate?’, and ‘Does your teacher ever say that people from all cultures are equal?’. The response scale ranged from 0 (absolutely never!) to 4 (very often!). The omega coefficient was 0.77 (95% CI [0.72, 0.81]).

Socioeconomic status

The index measuring socioeconomic status (SES) included four questions about the number of cars, computers (including laptops and iPads), bedrooms, and televisions in the household (Torney‐Purta, Lehmann, Oswald, & Schulz, 2001).

Proportion of in‐group classmates

Students ‘self‐reported ethnicity was used to calculate the proportion of the participating classmates of their same ethnic group. Proportion of in‐group classmates was then calculated for each student as the number of students in their ethnic group divided by the total number of participating students in the class.

Statistical approach

Preliminary analyses

Table 2 lists the intercorrelations, means, and standard deviations among all the raw, student‐level variables. There was considerable stability in children's classroom identification over time. Teacher closeness and conflict were moderately related to classroom identification, whereas peer friendship and rejection were weakly related to it (Cohen, 1988). Next, visual analyses of the distributions for each variable revealed skewed distributions for both conflict and rejection. We therefore transformed these variables for the main analyses (square roots).

Table 2

Means, standard deviations, and intercorrelations among student‐level predictor variables

	M	SD	1	2	3	4	5	6	7	8	9	10	11
1. CIWave 1	3.03	0.91
2. CIWave 2	3.00	0.97	.56***
3. Closeness	2.72	0.83	.39***	.25***
4. Conflict	0.67	0.73	−.31***	−.23***	−.53***
5. Peer friendship	0.15	0.09	.10*	.05	.10*	−.04
6. Peer rejection	0.11	0.13	−.09*	−.10*	−.12**	.20***	−.35***
7. Multicultural teaching	2.28	1.00	.10*	.02	.22***	−.10*	.07	−.10*
8. % In‐group classmates	0.53	0.29	−.06	.08	−.01	.04	.03	.02	−.28***
9. SES	2.05	0.52	.04	.02	.02	.03	−.03	.00	−.09*	.34***
10. Age	10.54	0.97	.11*	.08	.03	.05	−.10*	−.04	.21***	−.04	.09*
11. Ethnic minority status			.08	−.07	−.07	.10*	.01	.02	.25***	−.60***	−.32***	.06
12. Gender			−.11*	−.02	.05	−.08	−.02	−.04	.03	.05	−.09	−.05	−.09*

CI = classroom identification.

% In‐Group refers to the proportion of same‐ethnic classmates for each student.

***p < .001; **p < .01; *p < .05.

We tested the variation in classroom identification associated with school‐level and classroom‐level clustering. We first specified a three‐level, unconditional means model predicting end‐of‐year classroom identification scores with random intercepts at both the school and classroom level. We then used the variance estimates from this model to calculate the conditional intraclass correlations (ICCs). The results indicated that there was no variance associated with schools (ρschool  = 0) after accounting for classroom‐level variation (ρclass  = 0.14). We then used the ICCs to compute the design effect estimates for schools (DEFFschool  = 1) and classrooms (DEFFclass = 3.94) according to the procedures outlined by Peugh (2010). These estimates reflected a need for a multilevel modelling approach accounting for clustering only at the classroom level.

Main analyses

We specified two multilevel regression models in R using the ‘nlme’ package (Pinheiro, Bates, DebRoy, Sakar, & R Core Team, 2017). Standardized results were then calculated using the ‘reghelper’ package in R (Hughes & R Core Team, 2017). To assess the effect of ethnic minority status, we used a contrast coded ‘0.5’ for minority (Turkish and Moroccan) and ‘−0.5’ for majority Dutch students. Following Enders and Tofighi's (2007) recommendations for analyses focusing on associations between Level 1 (i.e., student‐level) variables, we centred all continuous student‐level variables within classrooms – except the proportion of ethnic in‐group classmates which was grand‐mean‐centred. This technique allowed us to separate the within‐classroom (Level 1) effects from the between‐classroom effects (Level 2; Bell, Jones, & Fairbrother, 2018; Enders, 2013), which we controlled for by including classroom averages in the models. In order to accurately capture students’ classroom experiences, which are informed by all classmates, we used the data for all original participants (n = 888) to obtain these classroom averages. The within‐classroom centred scores therefore reflect student variation relative to their classmates, whereas the grand‐mean‐centred scores reflect variation relative to the sample, which we believe is more relevant only for the proportion of ethnic in‐group classmates.

Results

Model 1 predicted classroom identification at Wave 2 (the end of the school year) from the positive and negative aspects of children's peer and teacher relationships, classroom ethnic composition (proportion of in‐group classmates), perceived multicultural teaching, and ethnic minority status and its interactions with each of the previous variables. To partial out the stability of classroom identification, the models controlled for this variable at Wave 1 (mid‐year). Age, gender, and socioeconomic status were included as control variables as well. Next, the models included the classroom means for all group‐centred variables to control for classroom‐level effects, which were non‐significant and are not included in the table. Fixed effects and variance estimates are listed in Table 3.

Table 3

Student‐level fixed effects estimates (Top), variance estimates (Bottom), and fit indices for models predicting year‐end classroom identification

Parameter	Model 1		Model 2
	β	SE	β	SE
Fixed effects
Classroom identification (Wave 1)	.49***	.04	.50***	.04
Closeness	.01	.04	.01	.04
Conflict	.02	.04	.03	.04
Peer friendship	−.02	.04	−.00	.04
Peer rejection	−.09*	.04	−.08†	.04
Minority status	−.08	.09	−.06	.09
% In‐group	.07	.08	.10	.08
Age	−.02	.04	−.01	.04
Gender	.03	.04	.02	.04
SES	−.07†	.04	−.08*	.04
Multicultural teaching	−.01	.04	−.00	.04
Closeness*Minority	−.07	.04	−.10*	.05
Conflict*Minority	−.10*	.04	−.05	.05
Friendship*Minority	.06	.04	.05	.05
Rejection*Minority	.04	.04	−.01	.06
% In‐group*Minority	.13	.09	.10	.09
Multicultural teaching *Minority	.02	.04	.00	.05
Closeness*% In‐group			−.06	.05
Conflict*% In‐group			.11*	.05
Friendship*% In‐group			−.01	.05
Rejection*% In‐group			−.09	.06
Multicultural teaching *% In‐group			−.02	.05
Variance estimates
Classroom level	0.07		0.07
Residual	0.56		0.54
Model fit
Loglikelihood*‐2	1124.68		1110.74

% In‐group refers to the proportion of same‐ethnic classmates for each student. Models also controlled for classroom mean values of within‐classroom‐centred variables.

***p < .001; **p < .01; *p < .05; † p < .1.

Next to a strong effect of classroom identification at Wave 1 – which indicated considerable stability over time – there was a main effect of peer rejection (β = −.09, p < .05). This effect was consistent with our expectations and indicated that peer rejection predicted declining levels of classroom identification over time. There were no significant main effects for student–teacher closeness, conflict, or minority status. However, there was a significant interaction between the latter two variables. As illustrated in Figure 2, ethnic minority students reported greater classroom identification when they perceived lower levels of teacher conflict relative to their classmates (β = −.1, p < .05). Post‐hoc simple slopes analyses revealed a significant negative slope (p < .05) for ethnic minority students, but not for majority students, indicating that the effect of teacher conflict was absent for Dutch children.

Figure 2

The interaction between student–teacher conflict and ethnic majority (Dutch)/minority (Turkish or Moroccan) group membership.

In model 2 we examined the moderating role of classroom ethnic composition by adding two‐way interactions with closeness, conflict, peer friendship and rejection, and multicultural teaching. Including these interactions as predictors significantly improved model fit, (5) = 13.94, p < .05. There was also an unexpected negative interaction between teacher closeness and ethnic minority status (β = −.1, p < .05). As illustrated in Figure 3, closeness appeared to predict a decline in classroom identification in ethnic minority students. However, post‐hoc simple slopes analyses indicate that the negative slope for ethnic minority students was only marginally significant (p < .1), and the slope for native Dutch students was non‐significant.

Figure 3

The interaction between student–teacher closeness and ethnic majority (Dutch)/minority (Turkish or Moroccan) group membership.

As expected, the proportion of in‐group classmates interacted positively with conflict (β = .11, p < .05). Figure 4 shows that conflict predicted declines in classroom identification among students for whom this proportion was relatively low (1 SD < M) but not among those with relatively many (1 SD > M) ethnic in‐group classmates. Post‐hoc simple slopes analyses further revealed a marginally significant negative slope (p = .05) only for students with a lower proportion of ethnic in‐group classmates. Importantly, the two‐way interaction between minority status and conflict was no longer significant in Model 2, but the main effect of proportion in‐group classmates was now marginally significant and positive (as expected).

Figure 4

The interaction between student–teacher conflict and proportion of in‐group (i.e., same‐ethnic) classmates.

Discussion

The present research examined the differential effects of both teacher and peer relations, perceived multicultural teaching, and ethnic classroom composition on the classroom identification of ethnic minority and majority students’ classroom over time. Taken together, our results provide partial support for our hypotheses. To begin with, we expected that relationships with teachers would be more important for the classroom identification of minority (vs. majority) students (Hypothesis 1) and that especially the negative aspects of children's relations in the classroom would affect their identification with it (Hypothesis 2). Starting with the latter, we did not find main effects for closeness, conflict, or friendship, but, for minority and majority children alike, peer rejection was associated with less classroom identification over time. Thus, our results are partly consistent with the notion that ‘bad is stronger than good’ meaning that people are more likely to be affected by negative versus positive situations (Baumeister et al., 2001).

With respect to our first hypothesis, we found that teacher conflict was detrimental for the classroom identification of minority but not majority students. However, this effect was no longer significant when the interaction between classroom composition and teacher conflict was partialled out. Thus, unlike we originally hypothesized, this result was not driven by the difference in cultural background between the immigrant‐origin child and the majority teacher (Chiu et al., 2012; Vedder & Horenczyk, 2006) or more specifically a stronger orientation towards authority figures among Turkish‐Dutch or Moroccan‐Dutch children (Pels et al., 2006; Thijs, 2011). Instead, it was being a member of a local ethnic minority (i.e., an underrepresented ethnic group in the classroom), which made students more sensitive to conflict with their teachers, and the likelihood of this was much larger for students with a Turkish and Moroccan background compared to their Dutch peers.

Unexpectedly, we found that closeness tended to predict a decrease in classroom identification for the minority children, whereas it had no effect for the majority children. We do not have a clear‐cut explanation for this result, but it could be related to the cultural dimension of power distance. Respect for authority is an important value in power‐distant cultures (Hofstede, 1991), and research has shown that parents of Turkish‐Dutch or Moroccan‐Dutch children have childrearing beliefs that stress obedience and respect for authority (Eldering, 1995; Leseman, Sijsling, Jap‐A‐Joe, & Sahin, 1995; Pels, 1991). As a result of this cultural orientation, the classroom identification of those children might be undermined if they have a too personal and intimate bond with their teacher rather than a more formal one. However, more research is needed to support such a conclusion and address the role of specific cultural values more directly. Additionally, the lack of positive effects for closeness can be related to the relatively short interval between the two measurement occasions (4–6 months) and the considerable stability of children's classroom identification. It is important to note that closeness was positively correlated to identification at both time points, and the possibility remains that closeness might increase classroom identification over longer time periods, especially in native majority students.

Our other hypotheses concerned the (main) effects of perceived multicultural teaching and the proportion of ethnic in‐group classmates. Although we anticipated this for the Turkish‐Dutch and Moroccan‐Dutch students (Hypothesis 3), we did not find that their perceptions of multicultural teaching predicted their classroom identification. Apparently, consistent with previous research (Thijs & Verkuyten, 2012), the quality of the interpersonal relationship between minority students and their majority teacher is more important than what the latter teaches about ethnic diversity. Next, and as expected (Hypothesis 4), the proportion of co‐ethnic students tended to have positive effect on the change in children's classroom identification in our second regression model. This finding is consistent with self‐categorization theory (Turner et al., 1987) according to which the relative homogeneity of the in‐group and one's similarity to it increase the likelihood of categorizing the self as an in‐group member. This effect of ethnic composition was absent in our first model and not stronger for the ethnic minority students, as we originally anticipated (Hypothesis 5). Still, it indicates that, although school ethnic diversity can be beneficial for children's interethnic relations (Thijs & Verkuyten, 2014), it may be easier to ‘fit in’ and belong when students are ‘amongst each other’. Apparently, this fitting in did not interact with the student–teacher relationship and teachers’ multicultural teaching (Hypothesis 6), suggesting that teachers are not more important for the classroom identification of students who are the numerical minority in the classroom. Also, against our expectations (Hypothesis 7), we did not find that the impact of children's peer relations depended on the proportion of co‐ethnic classmates. Previous longitudinal research has indicated that peer rejection is more painful if one's ethnic group is in the majority, because it is more difficult to explain away this rejection to the prejudice of one's classmates (Graham et al., 2009). Apparently, such attributional processes are less relevant for classroom identification.

By focusing on classroom identification as a central aspect of children's school belonging, the present study fits well with the social identity perspective, which appears to provide an adequate and promising framework for the study of children's educational adjustment (see Reynolds et al., 2017). One important contribution of the social perspective is that it can explain how students come to adopt the goals and standards of their school environment. More specifically, self‐categorization theory, which is part of the social identity perspective, posits that individuals are more likely to act in accordance with group norms and group expectations when their group identity is salient (Turner et al., 1987). Thus, high levels of classroom identification could make academic values more self‐relevant and thereby stimulate children's self‐determined academic motivation (Ryan & Deci, 2000). We did not include such outcomes in the present study. However, our findings on the effects of teacher conflict and peer rejection are consistent with a social identity perspective on the effects of in‐group rejection, and more specifically, the rejection‐disidentification model (Jasinskaja‐Lahti et al., 2009). To our knowledge, our results are the first to show that this model applies to relations in the classroom as well.

In evaluating the present findings, some limitations should be considered. First, our operationalization of classroom identification was limited. Although the correlation between them was strong, our measure consisted of two items only. Moreover, it assessed children's emotional attachment to their classroom as well as their liking of it, but not other aspects of identification such as centrality (the importance of the classroom for one's self‐definition) or introjection (the extent to which one is personally affected by what happens to the class; see Ashmore, Deaux, & McLaughlin‐Volpe, 2004). Thus, future studies should use more extensive scales to examine children's classroom identification.

Second, we cannot rule out the possibility that the association between student–teacher conflict and classroom identification was affected by shared method variance, as we relied on children's self‐reports to measure both variables. Future research should use teachers’ reports of the student–teacher relationship and use them and others as sources for data triangulation. Still, previous research has shown considerable convergence for teachers’ and students’ relationship reports, especially for conflict (Koomen & Jellesma, 2015). Moreover, if the student–teacher relationship affects students’ classroom identification, as the present findings indicate, it most likely does so through their subjective experience of that relationship.

Third, our longitudinal design consisted of two measurement occasions only. This design can be clearly preferred over a cross‐sectional one, as it permits conclusions about the direction of the associations found, but by including three or more measurement occasions, it would have been possible to control for time‐invariant individual differences (Hamaker, Kuiper, & Grasman, 2015).

To conclude, the present study makes a unique empirical and conceptual contribution to the literature on school belonging by examining classroom relations as antecedents of classroom identification in primary school students of different ethnicities. We showed that preadolescent students of all ethnicities reported more classroom identification over time if they were less rejected by their peers and if they had more co‐ethnic classmates. In addition to this, there were effects of the student–teacher relationship but only among the ethnic minority students. The experience of conflicted interactions with their teachers predicted a decrease in identification for them, and this could be explained by their ethnic underrepresentation in the classroom. However, they also tended to report lower classroom identification if they experienced more closeness with their teacher. Although more research is needed to interpret and substantiate our findings, they do indicate that negative peer relationships can undermine children's sense of classroom identification and that the student–teacher relationship has special significance for students from ethnic minority backgrounds. Practical attempts to promote classroom identification in different groups of students could therefore focus on ameliorating negative relationships with peers and teachers, and thereby make education matter for all.