Procedure
Data was collected in the Netherlands for the first wave (2010/2011) of the Children of Immigrants Longitudinal Survey in Four European Countries (CILS4EU; Kalter et al. 2016). The CILS4EU project focuses on the intergenerational integration of children of immigrants. Early adolescent immigrants and their majority peers around the age of 14 to 15 (3rd grade of secondary school) were the target population. A three-stage sampling method was used.
First, secondary schools were selected based on a probability proportional to size as well as the proportion of students with an immigrant background (referring to students who were themselves, or had at least one parent who was, born in a non-Western country). Schools were excluded from the school sample when the total number of students was less than 2% of the total target population and the target classroom size was smaller than one quarter of the average classroom size, or when it were special schools for students with cognitive, emotional, or physical disabilities. 34.9% of the 109 originally targeted schools participated. To increase the number of participating schools, a replacement strategy was implemented in which replacement schools were selected which matched the non-participating schools based on proportion of students with an immigrant background and school type. After this replacement strategy was implemented, 100 schools participated. Second, in most schools, two classrooms were selected. In schools with more than 60% immigrant students, as many classrooms as possible were selected to increase the number of participating immigrants. Of the selected classrooms, 94.5% (222 classrooms) participated in the study. Third, only students not being able to respond to the questionnaire in the language of the host country were excluded from the sample. Exclusions within schools were negligible. In total, 91.1% (4363 students) of the selected students participated in the study.
The CILS4EU survey was administered using paper-and-pencil questionnaires. Extensively trained students assistants were selected to administer the survey. In order to ensure privacy, questionnaires were identified with a unique ID number, which was linked to the specific student on a separate class list.
Sample
This study investigated the relation between aggression toward and being befriended and rejected by classmates specifically for the two largest ethnic groups in the Dutch CILS4EU data, namely non-immigrant Dutch adolescents and adolescents with an immigrant background from Turkish origin. Of these adolescents, those who had at least one Dutch and one Turkish classmate were selected. This resulted in a sample of 1042 adolescents (917 Dutch and 125 Turkish adolescents) in 85 classes in 56 schools. Adolescents in the sample were on average 14.9 years old (SD = 6.8 months) and there were about as many boys (51.4%) as girls. On average, Dutch adolescents’ socio-economic background was higher than that of Turkish adolescents (based on the international socio-economic index of occupational status, ISEI, ranging from 0 to 100; Ganzeboom et al. 1992; average ISEI for Dutch adolescents = 51.8, average ISEI for Turkish adolescents = 35.6). Only one participant in the sample had a missing value on one of the study variables. For this Turkish adolescent, information on sex was missing. As the exclusion of this participant will not likely influence the results, it was decided to exclude the participant from the analyses.
Measures
Immigrant background
Immigrant background was assessed using information on students’, their parents’, and their grandparents’ country of birth and country of origin as provided by the student as well as their parents. Students who themselves, their parents, and grandparents were born in the Netherlands, were classified as ‘Dutch’. Students who themselves, their parents or grandparents were born outside of the Netherlands were classified as having an immigrant background. A bottom-up approach was used to define immigrant adolescents’ ethnic origin in which information at the grandparent level was used first, followed by information about the parents and the adolescent. For the grandparent and parent levels, two decision rules were applied to define adolescents’ ethnic origin. First, the majority rule indicated that if the majority of (grand)parents was born in a certain country, this information was used to define adolescent’s ethnic origin. Second, the priority rule indicated that if there was no majority, priority was given to the country of birth of the (maternal grand)mother (for details on the definition of adolescents’ ethnic origin: Dollmann et al. 2014).
Friendship
Participants responded to the question: “Who are your best friends?”. They received a list of names and random identification numbers of all of their classmates (also non-participating classmates) and were asked to nominate up to five classmates as their best friends. To measure the extent to which students were befriended by their Dutch and Turkish classmates, the number of nominations for best friends they received from each of the groups was calculated. Furthermore, the proportion of nominations for best friends students received from each of the groups compared to the total number of nominations they could have received from these groups was calculated.
Rejection
Participants were also asked to nominate classmates they did not want to sit close to (“Who would you not want to sit by?”), which is taken as a proxy for rejection. Participants could nominate up to five classmates. To measure how often students were rejected by their Dutch and Turkish classmates, the number and proportion of rejection nominations students received from each of these groups were calculated.
Aggression
Participants were asked to nominate an unlimited number of classmates they perceived as being mean to them (“Who is sometimes mean to you?”). To measure students’ aggression toward their Dutch and Turkish classmates, the number and proportion of nominations students received from each of these groups were calculated.
Control variables
Self-reported sex was included to control for possible sex differences in being befriended or rejected by classmates (girls were coded as 0 and boys were coded as 1). In addition, to control for the availability of same- and cross-ethnic classmates, the number of Dutch and Turkish classmates adolescents have were included in the analyses.
Analytical Strategy
To account for the availability of Dutch and Turkish classmates, proportions of nominations are discussed in the descriptive statistics. In the confirmatory analyses, the absolute number of nominations is used while accounting for the number of Dutch and Turkish classmates.
Main analyses
Poisson regression models in Stata version 15.1 (StataCorp 2017) were used to examine the relation between aggression, friendship, and rejection because of the non-negative count characteristic of the dependent variables (Cameron and Trivedi 2013). Given that adolescents are more likely to be befriended or rejected by Dutch or Turkish classmates in classrooms with, respectively, more Dutch or Turkish students, the number of Dutch or Turkish classmates was used as an exposure variable, indicating the number of nominations an adolescent could have received. That is, the number of Dutch or Turkish classmates was used as on offset to account for opportunity differences (see Long and Freese 2006). Furthermore, robust standard errors for the parameter estimates were examined to control for mild violation of underlying assumptions (Cameron and Trivedi 2009).
Before conducting the regression models, it was checked whether the dependent variables’ variance exceeded the mean, i.e., whether there was an issue of overdispersion. This was checked by comparing the full regression model using a regular Poisson distribution to a model using a negative binomial distribution, which accounts for overdispersion. The difference between the model was tested using a likelihood ratio test of the overdispersion parameter alpha. For being befriended by Dutch or Turkish classmates, overdispersion was not found to be an issue (Χ2= 0.94, p = 0.17; negative binomial models could not be estimated for being befriended by Turkish classmates due to convergence problems). For being rejected by Dutch or Turkish classmates, however, overdispersion was found (Χ2 = 363.37, p < 0.001; Χ2 = 8.75, p = 0.002). Therefore, Poisson regression models were used to examine the extent to which adolescents were befriended by Dutch and Turkish classmates, whereas for their rejection by Dutch and Turkish classmates negative binomial regression models were used.
Students were nested in classrooms which were nested in schools. Given the low number of selected classrooms in each school, school-level variation was comparable to classroom-level variation. The intraclass correlation at the classroom-level was high for being befriended (0.34 and 0.27) and rejected by classmates (0.13 and 0.27) and the design effect was larger than 2 indicating the importance of accounting for the two-level structure in our data (Muthen and Satorra 1995). Moreover, likelihood ratio tests comparing the full regression model using a single-level model to a two-level model showed that the two-level Poisson regression model was more appropriate to examine the extent to which adolescents received friendship nominations from Dutch and Turkish classmates (Χ2= 28.66, p < 0.001; Χ2= 4.67, p = 0.02) and the two-level negative binomial models were more appropriate to examine adolescents’ rejection by Dutch and Turkish classmates (Χ2= 15.15, p < 0.001; Χ2= 11.58, p < 0.001). To account for the two-level structure, multilevel Poisson and negative binomial regression models were used.
The regression analyses were conducted separately for being befriended by and rejected by either Dutch or Turkish classmates as the dependent variables. Each regression analysis consisted of three models. In Model 1, we added a dummy indicating adolescents’ ethnic background to examine whether being befriended by (hypothesis 1) or rejected by (hypothesis 2) classmates differed between cross- and same-ethnic peers. In Model 2, we added aggression toward Dutch and Turkish peers as independent variables to examine the association between aggression toward each of these groups and being befriended or rejected by (hypothesis 3) classmates from each of these groups. In Model 3, we added interactions between adolescents’ ethnic background and their aggression toward Dutch and Turkish peers to examine whether the relation between aggression toward each ethnic group and being rejected (hypothesis 4) or befriended by (hypothesis 5) classmates differed between cross- and same-ethnic peers. All models controlled for adolescents’ sex and availability of Dutch or Turkish classmates. All continuous independent variables were grand-mean centered.
Additional analyses
In addition to the analyses among Dutch and Turkish adolescents, analyses were conducted for two other immigrant groups: Dutch and Moroccan (810 Dutch and 117 Moroccan adolescents in 75 classes), and Dutch and Surinamese (1099 Dutch and 121 Surinamese adolescents in 91 classes) adolescents. The measures and analytical strategy for these additional analyses are similar to the analyses for Dutch and Turkish adolescents. Results of the additional analyses can be found in Appendix 1.
Disclosure statement
All information regarding the selection of the sample, data exclusions, manipulations, and measures in the study are reported.