Same-Race Friendship Preference Across the Middle School Years: The Role of School Racial/Ethnic Context

Abstract

The current study examined the developmental trajectory of same-race friendship preference of racially/ethnically diverse students over the course of middle school. Participants were African American, Asian, Latinx, and White youth recruited at the start of middle school in 6^th grade (n = 4,361; M_age = 11.33 years) and followed across the three years of middle school. School racial/ethnic diversity and the racial/ethnic representation of students in their academic classes, including honors classes, were examined as predictors of friendship preferences over time. Results from latent growth curve models revealed that same-race friendship preference increased over the course of middle school and was shaped by both the school and classroom racial/ethnic context, above and beyond availability. Greater school racial/ethnic diversity predicted steeper increases in same-race friendship preference over time for all racial/ethnic groups. However, there were interactions involving race/ethnicity when the analyses focused on how students were represented in their academic classes compared to school. African American and Asian youth who were underrepresented in honors classes showed steeper increases in same-race friendship preference over time. Implications for prejudice reduction and creating more inclusive school environments are discussed.

With dramatic shifts in the racial/ethnic landscape of the U.S. school-aged population (NCES, 2019; Vespa et al., 2018), there are more opportunities than ever for schools to bring together diverse youth and create contexts in which meaningful relationships with peers from different backgrounds can develop. Although positive cross-racial interactions are necessary to fully realize the educational and psychosocial benefits of diverse schools (Garces & Jayakumar, 2014, Graham, 2018), same-race friendships are developmentally normative during adolescence and are uniquely beneficial for racial/ethnic identity development, social support, and coping with discrimination (Graham, et al., 2014; Rivas-Drake et al., 2014; Tatum, 2017). Indeed, evidence suggests that same-race friendship preference tends to increase over time, particularly during adolescence when identity negotiation and establishing peer networks are major developmental tasks (Shrum et al., 1988). Less is understood, however, about how the structural features of schools and classrooms shape the development of friendship preference above and beyond these normative tendencies. The present study examined how school racial/ethnic composition and the organization of students within academic classes affect the development of same-race friendship preference in middle school.

School Racial/Ethnic Composition and Friendship Preference

During adolescence, schools are the source for most friendships (George & Hartmann, 1996; Gifford-Smith & Brownell, 2003). Propinquity, or the availability of peers in schools, sets the stage for friendship formation. Blau’s (1977) macrostructural theory posits that as the size of one’s racial/ethnic group becomes smaller, opportunities to form same-race friendships decrease. Similarly, as racial/ethnic heterogeneity (diversity) increases, there are fewer opportunities for same-race contact, insofar as groups become relatively more equal in size. Thus, as availability of same-race peers decreases due to smaller ingroup size or greater school diversity, there should be fewer same-race friendships. Past research documents this propinquity pattern (Hallinan & Tuma, 1978; Joyner & Kao 2000).

When opportunities are taken into account, however, friendships often remain segregated based on race, even in diverse schools (Moody, 2001; Mouw & Entswisle, 2006; Quillian & Campbell, 2003; Thijs & Verkyten, 2014). This pattern highlights a second mechanism shaping friendships: homophily, or preference for similar peers (McPherson et al., 2001). In the U.S., race/ethnicity remains one of the most salient social categories on which homophily is based (Kao et al., 2019; McPherson et al., 2001; Shrum et al., 1988). How might the racial/ethnic composition of schools affect homophily processes? When composition is defined as the size of one’s racial/ethnic group, cross-sectional studies with adolescents find that even numerically small groups show more preference for same-race friendships compared to larger groups (Quillian & Campbell, 2003). For example, Wilson and Rodkin (2011) found that African American children in the numerical minority who resided in elementary school classrooms with a White majority showed more preference for same-race friendships. Contexts with a majority group may be particularly threatening to small groups, especially when a lack of numerical power coincides with racial/ethnic minoritized status.

The effects of composition and relative group size become more complex when more than two groups are present and the diversity of multiple groups needs to be captured. Unlike ingroup size, diversity measures allow us to model the number of groups in a setting and how evenly represented they are. The presence of many groups of relatively equal size is the hallmark of diversity (Budescu & Budescu, 2012). Research examining friendship preference as a function of school diversity has shown that as diversity increases, adolescents tend to prefer same-race friends (Currarini et al., 2010; McDonald et al., 2013; Moody, 2001; Mouw & Entwisle, 2006). Using Add Health data, Moody (2001) found a curvilinear association between school diversity and same-race friendship preference such that friendships became more segregated as diversity increased, peaking when schools are racially/ethnically balanced (i.e., two racial/ethnic groups of roughly equal size), then decreasing in the most diverse schools. Moody (2001) suggested that as a numerically smaller group increases in size with increasing diversity (i.e., multiple groups approaching relatively equal size), its members are more likely to find same-race peers with whom they share important attributes beyond race. It could also be that increases in the size of the numerical minority threatens the majority, creating an us vs. them dynamic. Studying friendship preferences as a function of school diversity is challenging because diversity measures are non-relational; alone, they do not distinguish between experiences based on ingroup size (e.g., being a numerical minority or majority) (Koopmans & Schaeffer, 2015). Thus, to unpack same-race friendship processes, it is clear that measures of school diversity, individuals’ own ingroup size, and their interaction are needed.

Dynamic Diversity and the Organization of Instruction

Thus far our discussion of numerical ingroup size and school racial/ethnic diversity as determinants of same-race friendship preference has focused on structural features of schools that are presumed to be relatively unchanging (i.e., time-invariant predictors). Yet there is growing recognition that contact opportunities as well as experiences of segregation vary across different social spaces in individuals’ daily lives, requiring new measures that move beyond structural features of one context at one point in time (Graham, 2018; McKeown & Dixon, 2017; Yip et al., 2019). Indeed, more dynamic features of the school environment are necessary when studying friendships given a central feature of adolescent friendship networks is their fluidity (Poulin & Chan, 2010).

One way to examine racial/ethnic composition in a more dynamic way is to use measures of ingroup size and diversity in academic classes compared to school (e.g., Echols & Graham, 2016; Juvonen et al., 2018; Kogachi & Graham, 2020). If schools were perfectly integrated, we would expect the relative representation of each student’s racial/ethnic group in their classes to match their representations at the school level. However, this is not often the case. Consider a school where African American youth make up 45% of the population. One African American student is in classes that are 70% African American – their ingroup is overrepresented based on their expected representation given the school composition. A different same-race student is in classes that are 20% African American – their ingroup is underrepresented.

We know from past research in the U.S. that differences in ingroup representation systematically vary depending on course level and tracks (e.g., Mickelson, 2015; Tyson, 2011), with White and Asian students tending to be overrepresented in higher-ability tracks and African American and Latinx students tending to be underrepresented in such tracks (e.g., Oakes, 2005). Prior friendship studies have reported that tracking at the school level (proportion of adolescents in advanced courses) predicts increases in same-race preference or concomitant decreases in cross-race preference (Hallinan & Williams, 1989; Kubitschek & Hallinan, 1988; Moody, 2001; Schofield & Sagar, 1977; Stearns, 2004). However, none of these studies has taken into account individual exposure within classrooms; nor have they looked at the effects longitudinally.

At the individual level, contact theory (Allport, 1954) suggests that differences in group representation based on course level not only restrict a student’s access to the full range of peers, but also creates a context structured such that status hierarchies are correlated with race. It may be that in such contexts same-race friendship preference is heightened. Tyson (2011) illuminates how racialized tracking processes affect the perceived link between race and achievement specifically for African American youth, and how underrepresentation in advanced or gifted programs can reinforce status differences based on race. Same-race friendships in such contexts have been shown to serve as important academic supports and resources (Riegle-Crumb & Callahan, 2009; Tyson, 2011). Thus, underrepresentation in advanced courses may lead to greater preference for same-race friendships for youth who are negatively stereotyped academically (African American and Latinx youth), especially when there are limited opportunities to form such friendships.

The Current Study

The current study was designed to examine how the school racial/ethnic composition, measured by ingroup size and overall diversity, and the organization of students in academic classes affect the development of preference for same-race friendships across the middle school years, over and above the normative developmental trajectories of friendship preference. The existing literature on the structural effects of a school’s racial/ethnic composition on friendship preference suggests that propinquity and homophily processes can evolve over time, yet past research has been almost entirely cross-sectional. We focus on the middle school years for two reasons. First, previous research on friendships suggests that cross-race friendships decline and same-race friendship preference increases during early adolescence (DuBois & Hirsch, 1990; Hallinan & Teixeira, 1987; Shrum et al., 1988), although the needed within-person longitudinal analyses have not been conducted. Second, the cognitive and social changes that occur during early adolescence shape the meaning of friendships, making this a particularly important developmental period to study who adolescents prefer as friends. Compared to childhood friendships, adolescent friendships are distinguished by their intimacy, self-disclosure, and emotional support (Berndt, 1982). Friendships in early adolescence also play a critical role in social identity formation processes, helping youth to integrate aspects of themselves and others as they develop a more complex understanding of the significance of racial/ethnic group membership (Rivas-Drake et al., 2014; Tatum, 2017).

We capitalized on a large racially/ethnically diverse sample that was followed over the three years of middle school. Students were recruited from middle schools that systematically varied in racial/ethnic diversity and the numerical representation of each of the four pan-racial/ethnic groups (African American, Asian, Latinx, and White). To capture same-race friendship preference, we used a measure that provides the probability of having a same-race friend, given the availability of both same- and cross-race peers in school (Jugert et al., 2011; Moody, 2001; Wilson & Rodkin, 2011). We also had class rosters that allowed us to measure each student’s exposure to same-race/ethnic peers in their individual classes at each time point. By calculating the difference between this more dynamic measure of each student’s individual exposure to same-race/ethnic peers and the structural measure of their group’s representation at the school level, we were able to examine the impact of under- or overrepresentation of ingroups in academic classes, including those that could be designated as advanced (i.e., honors courses).

The first aim of this study was to examine the development of same-race friendship preference during middle school and whether the school racial/ethnic context as measured by ingroup size (proportion same-race/ethnic peers in school), racial/ethnic diversity, and their interaction, affected initial levels and trajectories of same-race friendship preference. We also tested whether these associations differed for particular racial/ethnic groups. Because of the absence of longitudinal research, we took an exploratory approach. Consistent with a balance of power hypothesis (Juvonen et al., 2006), we hypothesized that youth whose racial/ethnic group is relatively small in size would show greater same-race friendship preference in schools that are less diverse (i.e., one large majority outgroup). This interactive effect may be especially true at the start of middle school when students do not know each other well, intergroup anxiety is heightened, and when characteristics that are socially meaningful like race/ethnicity may be salient for friendship choices. On the one hand, such context effects may diminish over time as youth become less anxious about interacting with outgroup members through more exposure and contact and find similarities beyond race/ethnicity on which to build friendships (Jugert et al., 2011; Turner & Cameron, 2016). On the other hand, the salience of race in friendships choices during adolescence may become more context-dependent (Raabe & Beelmann, 2009), with socially meaningful identities like race/ethnicity continuing to determine friendships in particular contexts based on the school racial/ethnic composition. Only longitudinal research can shed light on that possibility.

The second main goal of the current study was to better understand the structural and dynamic contributions of the racial/ethnic context on the development of same-race friendship preference over time. Of particular interest were the changing associations between underrepresentation in advanced academic classes experienced by different racial/ethnic groups. Taking into account this more dynamic racial/ethnic context variable, change can be modeled at both within-person and between-person levels. Within-person analyses assess yearly changes in representation in academic classes compared to school (e.g., does a student in advanced academic classes who is more underrepresented than usual in a given year show more same-race friendship preference than usual?). Between-person analyses assess stable individual differences in the racial/ethnic context predictors (e.g., do students who are most underrepresented in honors classes show the most increases in same-race friendship preference over time?). In past research with this sample, we found that the between-person analysis captures more of the structural features of school such as the organization of instruction, whereas the within-person analysis captures the psychological experiences of dynamic changes in the racial/ethnic context (Kogachi & Graham, 2020). Few longitudinal studies of friendship preference have tested these different effects. We hypothesized that between-person differences in underrepresentation in advanced academic classes would predict greater increases in same-race friendship preference trajectories over time, particularly for groups who are negatively academically stereotyped (i.e., African American and Latinx youth). Given that testing within-person effects is less common, we are reluctant to propose strong hypotheses. However, it could be that dynamic diversity – experiencing a significant change in representation in academic classes from year to year – has a similar effect on friendship preference. If adolescents experience more underrepresentation than usual in academic honors courses, they may be especially likely to gravitate toward same-race friends.

Most research on the impact of underrepresentation in advanced courses has focused on African American and Latinx youth as members of academically marginalized groups. Indeed, our hypotheses center on these two groups. However, we draw from a rich multiracial sample of Black, Latinx, Asian, and White youth that allowed us to examine the effects of structural and dynamic diversity on friendship preferences among more academically privileged racial/ethnic groups as well.

Method

Participants

The sample for this study was drawn from the UCLA Diversity Project an ongoing longitudinal study of 5,991 adolescents recruited in the fall of 6^th grade from 26 urban middle schools in California and followed during the three years of middle school. The schools were selected to represent a variety of racial/ethnic compositions. Six schools were racially/ethnically diverse such that no single racial/ethnic group represented a numerical majority and members of each of the four major racial/ethnic groups (i.e., African American, Asian, Latinx, and White) were present; 9 schools had two large and relatively equal racial/ethnic groups (e.g., Latinx and Asian) with very few members of other racial/ethnic groups; and 11 schools had a clear numerical majority racial/ethnic group (either African American, Asian, Latinx, or White) with a smaller number of members from each of the other racial/ethnic groups. Thus, the size of each racial/ethnic group varied systematically across schools from numerical minority to majority. To avoid confounding race/ethnicity and socioeconomic status (SES) in school selection, the sample was restricted to lower-middle and lower-SES communities. This was based on the percentage of students receiving free or reduced lunch and census data (e.g., median income, number of people in the work force) for neighborhoods in which schools were located. Schools with average enrollments of 900–1200 students and reading and math achievement (40th to 60th percentile on standardized tests) were selected.

The racial/ethnic composition of the sample was based on student self-report. Students were asked to select their race/ethnicity from the following options: American Indian, Black/African American, Black/other country of origin, East Asian, Latinx, Mexican/Mexican American, Middle Eastern, Pacific Islander, South Asian, Southeast Asian, White/Caucasian, Multiethnic/Biracial, and Other. For the present study, some groups were combined (Black/African American and Black/other country of origin, East Asian and Southeast Asian, and Latinx and Mexican/Mexican American). Based on student self-report, the sample was racially/ethnically diverse and included 31% Latinx, 19% Whites, 13% East/Southeast Asians, and 11% African Americans. The remaining 26% of participants who self-reported their race/ethnicity as either American Indian, Middle Eastern, Pacific Islander, South Asian, Multiethnic, or Other were excluded from analyses as they were either too small or are not recognized as racial/ethnic categories in school demographic data available at the California Department of Education (CDE). The analytic sample was therefore limited to the four major racial/ethnic groups.

As with most longitudinal studies, not all participants had complete data at each assessment wave. The majority of adolescents participated in all four waves (n = 3,377, 74%). An additional 640 (14%) participated in three, 444 (10%) in two, and 115 (3%) in one wave. There were few systematic differences in these participation rates based on the demographic characteristics of the sample. Our retention rates are similar to or better than other large longitudinal cohort studies of adolescents (e.g., Gutman et al., 2017; Harris et al., 2019). The final analytic sample consisted of 4,576 participants at the first wave of data collection (51% females; M = 11.33 years). The racial/ethnic composition of this sample was 41% Latino, 26% White, 17% East/Southeast Asian, and 16% African American.

Procedure

Institutional Review Board (IRB) approval was obtained from the University of California, Los Angeles (UCLA), for the UCLA Middle School Diversity study (IRB#11–002066-CR-00004). Participants were recruited in three cohorts in the fall of 6^th grade and were surveyed in the fall and spring of 6^th grade, and in the spring of 7^th and 8^th grade. To increase return rates of parental consent forms, two ipods were raffled in each school for students who returned the form. Across the 26 schools, 83% of the consent forms were returned with parents granting permission to participate. Surveys were group administered and read aloud by a trained graduate student researcher. Participants answered questions in survey booklets as a second research assistant circulated around the classroom to help individual students as needed. Students received honoraria of $5 for each 6^th grade survey, $10 for 7^th grade, and $10 for 8^th grade.

Measures

Time-Invariant Predictors

Race/Ethnicity.

Self-reported race/ethnicity was dummy coded such that Whites were the reference group.

Racial/Ethnic Ingroup Size.

To measure size of participants’ racial/ethnic group, the proportion of same race/ethnic peers in school was calculated using school-level race/ethnicity data collected from the CDE (2014). CDE data were aggregated into four primary racial/ethnic categories: African American, Asian (East/Southeast Asian), Latinx, and White. The proportion of students at school that matched students’ racial/ethnicity category was used. The values of this measure ranged from 0 to .68, indicating substantial differences in the relative size of racial/ethnic groups across the schools.

School Racial/Ethnic Diversity.

Objective school ethnic diversity was measured based on school-level race/ethnicity data collected from the California Department of Education (CDE) and using Simpson’s diversity index (1949):

$$D_S=1-{\sum _{i-1}^g{P}_i}^2$$

where P is the proportion of students in the school who are in racial/ethnic group i. This proportion is squared (p_i²), summed across g groups, and then subtracted from 1. D_S gives the probability that any two students randomly selected from a school will be from different racial/ethnic groups. Values can range from 0 to approximately 1, where higher values indicate greater diversity (i.e., more ethnic groups that are relatively evenly represented with no clear numerical majority). Thus, Simpson’s index captures the number of different groups in a setting and the relative representation of each group. D_S ranged from, .48 to .77, (M = .64, SD = .08) indicating moderate to high diversity.

Time-Invariant Covariates

Gender. Students self-reported their gender. Gender was binary-coded, with females assigned values of 1 and males assigned values of 0. Parent Education. The parent or guardian with whom the student lived was asked to complete a questionnaire about their highest level of education. An average score using educational level was computed, with higher scores indicating higher educational attainment. This measure ranged from 0 to 5 (0 = elementary or junior high school to 5 = graduate degree) (M = 3.86, SD = 1.59). Generational Status. Generational status was determined with three questions asking whether the student and his or her parents were born in the U.S. Students born abroad were considered first generation. Second- and third-generation students (U.S. born students with one foreign-born parent or both U.S.-born parents) were collapsed and assigned values of 0 (N = 3,906). First-generation students were assigned values of 1 (N = 515). Proportion Free or Reduced Priced Meals (FRPM). Proportion of students receiving FRPMs was included as a school-level covariate. FRPM served as a proxy for school SES.

Time-Invariant and Time-Varying Predictors

When constructs are measured at each time point like racial/ethnic ingroup representation in academic classes and academic honors, both between-person effects (the time-invariant average of each measure across waves) and within-person effects (time-varying deviations from the average) should be modeled (Curran & Bauer, 2011). The within- and between-person effects were disaggregated using centering recommendations outlined by Curran and Bauer (2011). Time-invariant measures were grand-mean centered (i.e., centered at the sample mean). Time-varying measures were group-mean centered (i.e., centered at each individual’s average level of each construct across middle school) to estimate youths’ deviations from their own average across each wave of data.

Racial/Ethnic Ingroup Representation in Academic Classes Compared to School.

To calculate academic classroom racial/ethnic ingroup exposure, classmate information from student transcripts was obtained (Echols & Graham, 2016). Because there was a high rate of participation within schools (M = 83%), this was considered to be a good estimate of students’ actual exposure to ingroup peers. Although students typically had four academic classes each day, there were few differences in the average proportion of same-race/ethnic classmates represented across these classes at any of the waves. Individual students’ racial/ethnic representation across academic courses was therefore averaged. Participants (n = 720) who were in more than one class that had less than 7 students in the sample (2 SDs below the mean of 21 students) were removed for that wave from analyses. For participants in this analytic sample, the proportion of same-race/ethnic peers in the school was calculated based on CDE data.

To calculate academic classroom racial/ethnic ingroup exposure, school proportion same-race/ethnic peers was subtracted from average classroom proportion same-race/ethnic peers. Positive scores indicate more same race/ethnic peers in academic classes compared to school (i.e. overrepresentation), while negative scores indicate fewer same race/ethnic peers in one’s classes compared to school (i.e., underrepresentation). Because classroom composition was measured at each wave of data collection, we could compute a time-invariant and time-varying measure. The time-invariant measure was computed by taking the average of this measure for each student across all waves of middle school, then grand-mean centered (M = −.01, SD = .13, range = −.48–.52). The time-varying levels of racial/ethnic ingroup representation in academic classes compared to school from each time point was group-mean centered, that is, the measure at each wave was centered around each participants’ time-invariant average across middle school (M_range = −.01 – −.03, SD_range = .11–.14).

Academic Honors.

Based on school transcripts, each student’s academic courses were coded such that honors courses were given a 1 and all other courses were given a 0. Codes were aggregated with scores of 3 or higher (i.e., student took 3 or more honors courses) re-coded as 1 to indicate academic honors designation. We chose this cutoff because of the bimodal distribution of honors course-taking within and between schools. Most students in honors courses took at least 3 courses with that designation. Students who were in accelerated academic programs (e.g., magnet, gifted) were also automatically coded as 1 to indicate academic honors. Across the four waves of data, 12.7%–28% of students were coded as academic honors. All other scores were coded as 0 to indicate not honors.

Time-Invariant and Time-Varying Covariate: Academic Grade Point Average (GPA)

Transcripts were obtained from school records data for each year of the study. Grades of all academic core courses (English, math, science, social studies) were coded on a scale ranging from 4.00 (A/A+) to 0.00 (F), with increments of .33 to indicate a grade that included a plus or minus. The average academic GPA was then computed by including the grades of all participants in each of these core classes at each wave of data. To measure the time-invariant control for mean levels of achievement across middle school, the average GPA for each participant across all waves of data was computed (M = 2.82, SD = .92). The time-varying measure was also included (M_range = 2.80–2.90, SD = .94–1.01).

Outcome: Same-Race Friendship Preference

Friendships were measured using a well-established sociometric procedure utilized in prior research (Graham et al., 2014). At each wave, students were asked to list the students in their grade who were their “good friends”. A list of the names of all students in their grade regardless of whether those students received parent consent, was provided upon request for spelling purposes. Students were allowed to make an unlimited number of nominations. Unilateral, as opposed to reciprocal, friendship nominations were used in this study because we were interested in who students considered to be their friend, regardless of whether the other person also viewed them as a friend. To determine if a friendship tie was same- or cross-race, each student’s self-reported race/ethnicity was used and matched to the target nominator’s own self-reported race/ethnicity. Friendships were considered same-race if they came from the same racial group as the nominator. Although the analyses focused only on African American, East/Southeast Asian, Latinx, and White participants, students from all racial/ethnic groups in the 26 middle schools were included to compute cross-race friendships. To assess preference for same-race friends relative to cross-race friends, we used the invariant log odds measure that computes same- and cross-race friendship nominations, net the opportunities for same- and cross-race contact in school (Charles & Grusky, 1995; Moody, 2001; Wilson & Rodkin, 2011). The odds of nominating a friend of the same race relative to the odds of nominating a friend of a different race was computed using the following formula:

$$ɑ=\text{AD}/\text{BC}$$

where A is the number of same-race friends, B is the number of cross-race friends, C is the number of same-race peers not nominated as a friend, and D is the number of cross-race peers not nominated as a friend. An advantage of using this measure is that it does not depend on the total number of nominated friends and the total number of same- or cross-race peers in the school (Moody, 2001). Given the distribution of odds ratios is highly skewed, ɑ was log-transformed. Values of the log-transformed ɑ can range from −∞ to +∞, with positive values indicating preference for same-race friends, and negative values indicating preference for cross-race friends. Zero indicates neutrality (M_range = 1.01–1.38, SD_range = 4.04–4.23)

Analytic Plan

A series of latent growth curve models (LGCMs) was conducted to address the aims of this study. The first aim was to examine the level of same-race friendship preference at the start of middle school, the development of same-race friendships over time, and whether school racial/ethnic composition and race/ethnicity predicted different initial levels and trajectories of same-race friendship preference. To do this, an unconditional model was first estimated to establish the initial levels of same-race friendship preference at the start of middle school and its trajectory to model rate of change over time. A series of conditional models was then estimated to examine the effects of the school racial/ethnic composition on both initial status and change in friendship preference, using proportion of same-race/ethnic peers in school, racial/ethnic diversity, and their interaction. The analytic model of the first study aim is depicted in Appendix A. We tested whether effects varied by youths’ race/ethnicity. Demographic and achievement indicators were controlled for in each model.

The second aim of this study was to test the effect of racial/ethnic ingroup representation in academic classes compared to school on the development of same-race friendship preference, and whether these effects varied by academic honors designation and student race/ethnicity. The analytic model of this aim is depicted in Appendix B. We tested both the between-person (time-invariant) and within-person (time-varying) effects of ingroup representation in academic classes on change in friendship preference simultaneously. To do this, the intercept and slope of friendship preference were regressed on the between-person (time-invariant) racial/ethnic ingroup representation in academic classes, academic honors designation, race/ethnicity, and their interaction terms. Additionally, friendship preference at each time point was regressed on the measure of ingroup representation in academic classes at the same time point to capture within-person fluctuations in the academic context. By including time-varying racial/ethnic ingroup representation in academic classes and academic honors, we can control for the dynamic short-term deviations from the overall growth trajectory of outcomes and allow participants to serve as their own control. This eliminates unmeasured time-invariant confounds as alternative explanations, providing stronger inferences regarding the associations of interest (Raudenbush & Bryk, 2000).

In constructing the final LGCMs for both study aims, the reference groups were rotated to test each ethnic group comparison. Significant interactions were probed at 1 SD above and below the mean for continuous variables. Nonsignificant interaction terms were removed (Aiken & West, 1991). All analyses were conducted in Mplus (Muthén & Muthén, 1998, version 7.3) using TYPE=COMPLEX to account for nesting of students within schools. The estimation procedure MLR was specified for handling missing data through full-information maximum likelihood (FIML), allowing for the inclusion of all available data in the analyses by fitting the covariance structure model directly to the observed raw data for each participant (Enders, 2010).

Results

Intercorrelations and descriptive statistics of the main study variables are shown in Table 1. Same-race friendship preference showed moderate stability over time and was on average significantly and negatively correlated with school racial/ethnic diversity, positively correlated with ingroup size at school, and positively correlated with racial/ethnic ingroup representation in academic classes compared to school.

Table 1.

Intercorrelations and Descriptive Statistics for Main Study Variables

	1	2	3	4	5	6	7
1. Same-Race Friendship Preference F 6th Gr	–
2. Same-Race Friendship Preference S 6th Gr	.59**	–
3. Same-Race Friendship Preference F 7th Gr	.44**	.49**	–
4. Same-Race Friendship Preference F 8th Gr	.32**	.41**	.54**	–
5. School Racial/Ethnic Diversity	−.15**	−.18**	−.15**	−.14**	–
6. School Racial/Ethnic Ingroup Size	.20**	.19**	.19**	.16**	−.01	–
7. Time-Invariant Ingroup Representation in Academic Classes	.28**	.29**	.27**	.24**	−.06**	−.16**	–
N	4,227	4,213	3,732	3,373	4,511	4,576	4,381
Mean (Uncentered)	1.01	1.08	1.21	1.38	0.63	0.39	0.02
Std. Deviation	4.23	4.06	4.08	4.04	0.08	0.16	0.13

Note.

^*p < .05

^**p < .01

^***p < .001.

Unconditional Growth Model

To examine the overall change over time in same-race friendship preference, an unconditional LGCM was estimated. Factor loadings of the trajectory’s intercept or initial status at the fall of 6^th grade were set to 1 to represent the starting point. Factor loadings of the second latent factor represented the trajectory’s slope or rate of change. To define the slopes as linear, the loadings were set to 0, .5, 1.5, and, 2.5 reflecting the data collection schedule. To test the possibility of nonlinear trajectories, a freed loading model in which only two parameters for the slope factor were fixed with time-codings was estimated and the rest were freely estimated (Bauer & Cai, 2009). Chi-square difference tests were computed to compare model fit and estimates of the coefficients were examined to determine if nonlinear growth should be modeled. The freed loading model did not result in significant improvements of model fit indicating that modeling a linear trajectory fit was appropriate.

The unconditional model for same-race friendship preference fit the data well, χ² (5, N = 4,537) = 18.90, p = .002; CFI = 1.00; RMSEA=.03, 90% CI [.01, .04]. The mean level of same-race friendship at the start of sixth grade was .31 reflecting that adolescents on average preferred same-race friendships. On average, adolescents also showed greater same-race friendship preference over time (b = .10, p < .01). There was significant variation in initial status and growth indicating that adolescents start at different levels of same-race friendship preference (b = 11.04, p < .001) and grow subsequently at different rates (b = 1.53, p < .001). The initial level of friendship preference at the start of sixth grade was negatively correlated with its slope factor (b = −.49, p < .001) revealing that adolescents who show more same-race friendship preference at the start of middle school showed less steep increases in same-race friendship preference over time compared to those who showed less same-race friendship preference at the start of middle school who have steeper increases over time. This may capture a “ceiling effect” given those who report higher initial levels have a greater opportunity to decrease over time.

Effects of School Racial/Ethnic Composition on Same-Race Friendship Preference

To examine the effect of school racial/ethnic composition and race/ethnicity on the initial status of and change in same-race friendship preference, conditional models including school diversity, ingroup size in school, race/ethnicity, and their interactions were tested in a stepwise process. To test the possibility of a curvilinear effect of school racial/ethnic diversity on same-race friendship preference (Moody, 2001), the squared term for school diversity was included to predict both the intercept and slope. However, the effects were non-significant and not included in the final model.

The final conditional model fit the data well χ² (27, N = 3,901) = 93.14, p < .001; CFI = .99; RMSEA=.02, 90% CI [.00, .02]. As shown in Table 2, African American youth showed less same-race friendship preference compared to White youth (b = −.76, p < .05), and steeper increases in same-race friendship preference compared to White youth over time (b = .24, p < .05). There were no other racial/ethnic differences.

Table 2.

Coefficients From Final Conditional LGCM Examining School Racial/Ethnic Composition

	Same-Ethnic Friendship Preference
Parameter	Standard. (S.E.)
Predicting Intercept Factor
Gender	.65 (.15)***
Parent Education	−.09 (.05)
Generational Status	.49 (.28)
African American	−.76 (.38)*
Asian	.32 (.30)
Latinx	−.01 (.27)
Overall GPA	−.19 (.10)
Overall Honors designation	−.08 (.32)
School Ingroup Size	4.76 (.92)***
School Diversity	−4.31 (1.26)**
School % Free/Reduced Meals	.03 (.01)***
School Diversity X Ingroup Size	17.56 (2.04)*
Predicting Slope Factor
Gender	.−.18 (.06)**
Parent Education	−.01 (.03)
Generational Status	.02 (.09)
African American	.24 (.13)
Asian	.15 (.12)
Latinx	.23 (.17)
Overall GPA	.02 (.08)
Overall Honors designation	−.02 (.12)
School Ingroup Size	−.49 (.34)
School Diversity	1.35 (.68)*
School % Free/Reduced Meals	.00 (.00)
School Diversity X Ingroup Size	−4.22 (2.95)
Intercept	1.08 (.45)*
Slope	.02 (.27)
Intercept/Slope Covariance	−1.88 (.01)***
Intercept Variance	9.28 (.53)***
Slope Variance	1.45 (.11)***
χ2(20, N = 3,901)	93.14 (p < .001)
CFI	.99
RMSEA (90% CI)	.02 (.00, .02)

Note.

^*p < .05

^**p < .01

^***p < .001.

Turning to the effects of school composition, there was a significant effect of the interaction between ingroup size and school racial/ethnic diversity predicting the intercept of same-race friendship preference (b = 17.56, p < .05), but no significant effect of the interaction on the slope (b = −4.22 p = .15). Furthermore, this two-way interaction did not vary by race/ethnicity for either the intercept (African American: b = −1.29, SE = 3.58, p = .72; Asian: b = 6.07, SE = 3.38, p = .07; Latinx: b = −4.00, SE = 4.01, p = .57) or slope (African American: b = 1.90, SE = 1.41, p = .18; Asian: b = .12, SE = 1.50, p = .94; Latinx: b = 2.75, SE = 1.83, p = .13), and therefore 3-way interactions were removed. Reference groups were rotated and no other group differences were found.

As depicted in Figure 1, tests of simple slopes revealed that the effect of ingroup size on friendship preference was not significant for youth in schools that were less diverse (b = 1.37, p = .55). That is, in less diverse schools (i.e, schools with a clear numerical majority and minority), all youth preferred same-race friendships regardless of ingroup size. However, in more racially/ethnically diverse schools (i.e., schools with more groups present that were relatively more equal in size), ingroup size significantly predicted same-race friendship preference (b = 6.95 p < .001) such that as ingroup size increased, same-race friendship preference increased, with numerically smaller groups showing less preference for same-race friendships. Thus, in support of our hypothesis, numerically small groups showed less preference for same-race friends in more diverse schools compared to less diverse schools.

Figure 1. Interaction Effect of School Racial/Ethnic Diversity and Ingroup Size Predicting Initial Levels of Same-Race Friendship Preference and Regions of Significance.

Note. *p < .05 **p < .01 ***p < .001.

Analysis of the slope of same-race friendship preference revealed only a significant main effect of school racial/ethnic diversity. Consistent with previous cross-sectional research, greater school diversity predicted steeper increases in same-race friendship preference over time (b = 1.29, p < .05).

Effects of Ingroup Representation in Academic Classes and Honors Designation

To address the second aim of the study, the effect of the organization of youth in academic classes, academic honors designation, and race/ethnicity on the initial level and change in same-race friendship preference was tested. Time-invariant (between-person) average levels of racial/ethnic ingroup representation in academic classes compared to school, academic honors designation, race/ethnicity, and their interactions were added to both the intercept and slope factors (see Table 3). There were no significant 3-way interactions predicting the intercept. However, there was a significant 3-way interaction predicting the slope involving African American (b = −6.67, p < .001) and Asian (b = −10.91, p < .001) youth compared to White youth. Reference groups were rotated and no other group differences were found.

Table 3.

Coefficients From Final LGCM Examining Classroom Ethnic Context

	Same-Ethnic Friendship Preference
Parameter	Standard. (S.E.)
Predicting Intercept Factor
Gender	.57 (.14)***
Parent Education	−.08 (.06)
Generational Status	.41 (.31)
African American	.10 (.30)
Asian	.21 (.24)
Latinx	−.39 (.28)
Overall GPA	−.20 (.13)
Overall Honors designation	−.17 (.69)
Ingroup Representation in Academic Classes	12.33 (1.63)***
Ingroup Representation in Academic Classes X African American	−1.77 (2.90)
Ingroup Representation in Academic Classes X Asian	−4.22 (2.07)*
Ingroup Representation in Academic Classes X Latinx	−1.75 (2.36)
Ingroup Representation in Academic Classes X Honors	4.24 (2.90)
Honors X African American	2.60 (.74)***
Honors X Asian	−.20 (.91)
Honors X Latinx	.65 (.91)
Ingroup Representation in Academic Classes X Honors XAfrican American	8.44 (4.69)
Ingroup Representation in Academic Classes X Honors X Asian	−6.14 (5.34)
Ingroup Representation in Academic Classes X Honors X Latinx	−5.41 (3.62)
School Ingroup Size	7.92 (.86)***
School Diversity	−6.35 (1.41)***
School % Free/Reduced Meals	.015 (.01)
Predicting Slope Factor
Gender	−.18 (.07)*
Parent Education	−.02 (.03)
Generational Status	−.02 (.12)
African American	.27 (.15)
Asian	.22 (.10)*
Latinx	.35 (.15)*
Overall GPA	.04 (.08)
Overall Honors designation	.13 (.39)
Representation in Academic Classes	−2.36 (1.17)*
Representation in Academic Classes X African American	2.08 (1.29)
Representation in Academic Classes X Asian	2.75 (1.43)
Representation in Academic Classes X Latinx	1.69 (1.33)
Representation in Academic Classes X Honors	3.15 (1.64)
Honors X African American	−.12 (.44)
Honors X Asian	.44 (.54)
Honors X Latinx	−.27 (.38)
Representation in Academic Classes X Honors XAfrican American	−6.67 (2.51)***
Representation in Academic Classes X Honors X Asian	−10.91 (2.85)***
Representation in Academic Classes X Honors X Latinx	−.64 (1.93)
School Ingroup Size	−.72 (.39)
School Diversity	1.89 (.62)**
School % Free/Reduced Meals	.01 (.00)
Time-Varying Controls
Representation in Academic Classes	2.45 (.83)**
W2 GPA	.10 (.12)
W2 Honors designation	−.64 (.37)
W3 GPA	−.27 (.17)
W3 Honors designation	−.53 (.14)***
W4 GPA	−.16 (.13)
W4 Honors designation	.49 (.20)**

As depicted in Figure 2, tests of simple slopes revealed that for African American (β = .38, p < .001) and Asian (β = .53, p < .01) youth, being underrepresented in academic honors classes predicted significant increases in same-race friendship preference over time. In contrast, there were no changes in same-race friendship for Whites who were underrepresented in honors classes (β = .16, p = .52). Rather, for White youth, being underrepresented in non-honors courses was predictive of steeper increases in same-race friendship preference over time (β = .29 p < .05). No other slopes showed significant change in same-race friendship preference.

Figure 2. Interaction Between Racial/Ethnic Ingroup Representation in Academic Classes, Honors, and Students’ Race/Ethnicity Predicting Trajectories of Friendship Preference.

Note. *p < .05 **p < .01 ***p < .001.

Turning to the within-person effects of ingroup representation in academic classes compared to school, there was a significant within-person effect (b = 2.45, p < .01) revealing that in years when there were more racial/ethnic ingroup peers in academic classes than usual, youth showed more same-race friendship preference than usual. There were no significant interactions between time-varying ethnic ingroup representation in academic classes with academic honors designation or ethnicity. These non-significant interactions were not included in the final model.

Discussion

Do early adolescents’ preference for same-race friendships increase over the middle school years? Does exposure to other racial/ethnic groups in diverse schools slow down this rate of growth? These questions that draw on principles of homophily (similarity) and propinquity (availability) in friendship formation have not been well studied in the friendship literature, in part because most studies of same-race friendships have been cross-sectional and have not systematically studied the school racial/ethnic context as multifaceted, multilevel, and dynamic over time. The current study took a longitudinal approach to investigate the development of same-race friendship preference over the course of middle school, and how the school and classroom racial/ethnic context shape such preferences over and above normative developmental trends. The study drew from a unique sample of racially/ethnically diverse adolescents whose numerical ingroup representation varied systematically within schools that also varied by school racial/ethnic diversity. By studying variables not previously examined in the friendship literature, our research offers new insights into the conditions under which early adolescents seek close ties with peers who share their same race.

Effects of School Composition on Same-Race Friendship Preference

Our first aim was to examine the overall trajectory of same-race friendship preference and whether the school racial/ethnic context as measured by ingroup size, racial/ethnic diversity, and their interaction, affected initial levels and slopes of same-race friendship preference. On average, all youth tended to prefer same-race friendships at the start of middle school, which is consistent with past research (e.g., Thijs & Verkyten, 2014). Over the course of middle school, same-race friendship preference increased. These normative increases in same-race friendship preference coincide with a time in development when race/ethnicity takes on added significance and when youth become more aware of their racial/ethnic group membership (Graham & Echols, 2018).

Is homophily a more powerful determinant of friendship preference than propinquity as the above findings might suggest? The answer to this question becomes more complex when we examined the school contextual effects of ingroup size and school diversity on preference beyond availability. Two schools can have the same diversity score, as calculated by Simpson’s index, but the numerical size of the different racial/ethnic groups within those schools can vary quite a bit (Koopmans & Schaeffer, 2015). We were therefore especially interested in testing the interaction between diversity and group size in order to distinguish how the broader context of diversity impacted individual students based on their ingroup’s size. Findings revealed that at the start of middle school when youth might not know each other well, the effect of ingroup size on friendship preference depended on school diversity. In schools that were more racially/ethnically diverse (i.e., when outgroups were made up of multiple, more equally represented racial/ethnic groups), youth showed less same-race friendship preference as ingroup size decreased. In contrast, in less racially/ethnically diverse schools (i.e., where there was one relatively large outgroup), all youth – including numerically small groups – preferred same-race friendships. Focusing on numerically small groups, school diversity mattered in unique ways. Homophily (preferring same-race friends) was only evident in less diverse schools.

These findings are consistent with past research showing that numerically small groups “hunker down” in the face of a large outgroup and seek solidarity with ingroup friends (Quillian & Campbell, 2003; Wilson & Rodkin, 2011). Extending these findings, the results from this study suggest that concerns about power dynamics and processes of threat or competition for numerically small groups may be specific to less diverse school contexts where there is one large racial/ethnic majority group. Greater school racial/ethnic diversity, on the other hand, may serve as a buffer to turning inward at the start of middle school for numerically small groups. Past research has shown that school diversity can reduce social vulnerability and self-blaming attributions when experiencing peer harassment, as well as foster less rigid and more inclusive norms that allow all students to better fit in (Graham et al., 2009; Juvonen et al., 2018). Our results suggest that diversity can also affect friendship preferences, fostering greater crossing of racial/ethnic boundaries for youth who may feel marginalized due to being numerically small in size. Thus, in less racially/ethnically diverse schools, the start of middle school may be a time to target feelings of vulnerability or marginalization for youth who have fewer same-race peers.

Numerically larger groups, on the other hand, showed relatively high levels of same-race friendship preference regardless of school diversity. Thus, the pull of racial homophily appeared to be particularly strong. This may be because larger groups have more opportunities to find friends who match on other important attributes within their racial/ethnic group and do not need to seek friendships with outgroups (Moody, 2001). We know that similarity in a critical determinant of friendships at all ages and especially in adolescence (Laursen, 2017; McPherson et al., 2001). What features of similarity in addition to (other than) race attract adolescents to one another is an important topic for future research.

The interaction between school diversity and group size discussed above was documented at the beginning of middle school. When examining the trajectories of same-race friendship preference, the pattern of findings differed. Consistent with past cross-sectional research that has documented that preferences for same-race friendships tend to increase with greater school racial/ethnic diversity except in the most diverse schools (Currarini et al., 2010; McDonald et al., 2013; Moody, 2001; Mouw & Entwisle, 2006), greater school racial/ethnic diversity predicted steeper increases in same-race friendship preference over time, above and beyond initial levels and developmental increases of friendship preference over time. Neither ingroup size nor its interaction with school diversity significantly predicted change in friendship preference. There were also no racial/ethnic group differences in the effects of school diversity. Thus, race/ethnicity appears to become increasingly important for friendship selection in more diverse middle school contexts, which may provide enough opportunities to find friends within one’s racial/ethnic group (intraracial diversity) (Moody, 2001).

Classroom Organization and Dynamic Diversity

Why might school diversity predict increases in same-race friendship preference over time? In prior research with this sample, we found that race/ethnicity-related outcomes may be most affected by the organization of students within academic classes (Juvonen et al., 2018; Kogachi & Graham, 2020). Indeed, the effect of school racial/ethnic diversity may not fully reflect the racial/ethnic realities that adolescents encounter day-to-day across the middle school years and as youth become more familiar with their peers. Rather, the development of friendship preference over time may be a response to racialized experiences within academic classes.

To further unpackage these racialized experiences, the second main goal of the current study was to examine how the organization of youth in academic classes shapes same-race friendship preference during middle school. We examined the effect of racial/ethnic ingroup representation in academic classes, academic honors status, and race/ethnicity. Instead of using aggregated data by racial/ethnic group at the course level (e.g., over or underrepresentation of racial/ethnic groups in honors classes within a school), we focus on the individual level representation of youth’s racial/ethnic group within their unique set of academic classes. Compared to Whites, African American and Asian youth who were underrepresented in their honors classes showed steeper increases in same-race friendship preference over time. Latinx youth were not significantly different from White youth and showed no such changes over time.

These interactive effects differed from our original hypotheses that underrepresentation in honors classes would be most impactful for groups who are negatively academically stereotyped. Underrepresentation in academic classes can create a threatening environment for these youth and same-race friendships may serve as important supports (Graham & Morales-Chicas, 2015; Riegle-Crumb & Callahan, 2009; Tyson, 2011). As such, the findings for African American youth were expected.

We did not hypothesize that Asian youth would show similar patterns. However, Asians are also academically stereotyped albeit in a positive direction. Past research has documented the unique challenges associated with the model minority stereotype for Asian students, including increased performance pressures (e.g., Cheryan & Bodenhausen, 2000) and poorer mental health (e.g., Lee & Zhou, 2015; Chen & Graham, 2018). Positive stereotypes are complex and are not often well-received. For example, Asian adults who were targets of positive stereotypes from outgroup members showed increased vigilance around being viewed based on stereotypes as opposed to individuating information, and believed outgroup members who held positive stereotypes also held negative views about Asians (Siy & Cheryan, 2016). Although not tested in this study, the unique processes and stressors associated with the model minority stereotype may be heightened in contexts of underrepresentation in honors classes for Asian youth, where race/ethnicity as well as achievement stereotypes are more salient. This in turn may lead Asian youth to increasingly seek same-race friendships as a way of coping.

The model minority stereotype also perpetuates inequality and status hierarchies by minimizing racial discrimination against Asians, and by diminishing the significance of race and discrimination on achievement outcomes for other racial/ethnic minoritized groups, particularly African Americans (Lee & Zhou, 2015; Wu, 2014). Developmental research suggests that Asian adolescents report more peer discrimination and race-based peer victimization due to the stereotypes about their achievement and sociability (e.g., Greene et al., 2006; Rosenbloom & Way, 2004). In contrast, African American youth report more institutional discrimination from adults in school. It could be that discrimination experiences for Asian and African American youth increase in contexts where achievement is salient and when one is underrepresented. We hope to pursue the long-term adjustment consequences of underrepresentation in different classes for racial/ethnic groups who carry the burden of academic stereotypes, whatever their content. Deeply entrenched achievement disparities in this country (and in our sample) suggest that a focus on African American and Asian youth is a good first step.

Limitations and Future Directions

Despite its contributions, this study had several limitations. First, because we were particularly interested in the longitudinal processes of same-race friendship preference development and how the racial/ethnic context and race/ethnicity play a role, latent growth curve modeling was used rather than social network analysis. Therefore, network mechanisms that may be contributing to greater same-race friendships were not controlled for. For example, the tendency to become friends with friends of friends can lead to greater same-race friendships over time; such processes do not reflect preferences (e.g., Moody, 2001). Thus, same-race friendship preference may be overestimated in this study. We are also not able to capture broader friendship networks at the school level. The degree to which adolescents’ own friendship networks are integrated within the school friendship network, and how interconnected the overall school friendship network happens to be, could have important moderating effects on the impact of same-race friendship preference on developmental outcomes (Cappella et al., 2017; Gest et al., 2011). School and classroom racial/ethnic diversity do not tell us who is hanging out with who and how segregated or integrated the overall friendship patterns are. As researchers studying school-based friendship networks, we can broaden and enrich our conception of racial/ethnicity in the network literature by incorporating different school level variables that change over time (e.g., degree to which a school’s friendship network is segregated based on racial/ethnicity) to predict changes in friendship dynamics.

A second limitation is that we do not have measures of academic tracking. The phenomenon of racialized tracking is multidimensional and requires moving beyond ingroup over- and underrepresentation. The degree of segregation of outgroups within classes must also be taken into account. For example, does underrepresentation in honors courses for African American and Asian youth lead to greater same-race friendship preferences if outgroups in those classes are more racial/ethnically diverse? Do these associations depend on the relative representation of a particular racial/ethnic outgroup (e.g., Whites vs. Latinxs)? Understanding the meaning of school racial/ethnic diversity and its association with perceptions of the racial/ethnic climate and social and academic hierarchies will be important future directions in shedding light on the ways racial/ethnic diversity in both schools and classrooms can affect intergroup relations (Graham & Morales-Chicas, 2015; Saenz et al., 2007).

Third, we focus on the school and academic classroom contexts, but did not examine extracurricular activities. Extracurricular activities have been shown to be an especially important way schools can promote cross-race friendships (Knifsend & Juvonen, 2016; Moody 2001). However, to the degree that extracurricular activities are segregated and racialized which is often the case (Clotfelter, 2002), similar patterns in friendships as those found with academic classes may occur. Thus, future research can extend this work to that context.

Fourth, the same-race friendship preference measure was limited in that it does not measure whether friendships were with peers in academic classes. Nor does it capture the stability of each unique friendship tie. Having same-race friendships may reflect a need to seek out support and solidarity regardless of whether that friend was in academic classes, or maybe especially when that friend shared those classes. An important question for future research is how the organization of youth in schools might affect the relational stability and quality of same-race friendships both in and outside of the classroom (Lessard et al., 2019).

Finally, our data were collected in California, one of the most racially and ethnically diverse states in the U.S. California has the largest Latinx, Asian, Multiracial/ethnic, and Native American populations of any state (U.S. Census Bureau, 2019), along with the presence of smaller racial/ethnic groups (e.g., Middle Eastern, South Asian). We were not able to include these groups due to their small size in our schools and/or that the CDE does not disaggregate its data by these groups. However, given the robust findings of homophily, we would expect to find comparable levels of preference for similar others across different racial/ethnic groups particularly given that the contextual effects were powerful for numerically small groups in our study. Whether our findings can be replicated in other parts of the U.S. that are less diverse or that are experiencing a more recent immigration history than California will be important for future research. However, the findings from this study may foreshadow friendship preferences among an increasingly diverse school-age population in other parts of the country.

Conclusion

The nature of intergroup relations in the U.S. is evolving as racial/ethnic groups change in relative size and status (Vespa et al., 2018). These demographic shifts are most dramatic among the U.S. school-aged population (Hussar & Bailey, 2020), making it a critical time to be studying the effects of school and classroom composition on intergroup relations. Like past research, the results from the present study suggest that adolescents prefer same-race friendships across the middle school years. This is good news. Same-race friendships are developmentally normative and provide support and positive ingroup identity, which are important during adolescence (Graham et al., 2014; Tatum, 2017). Over and above these normative trajectories, however, both the structural and organizational features of schools play an important role in shaping same-race friendship preference in middle school that can reinforce rigid boundaries between groups along racial/ethnic lines. Thus same-race friendship preference, or turning inward, may at times be a reaction to school practices such as racialized tracking. It would be misguided to support such practices as a mechanism for promoting ingroup solidarity. Instead, we argue that reducing social and academic hierarchies based on race/ethnicity due to the ways that ingroup size and school diversity interact, and the degree to which race/ethnicity is correlated with academic honors, can diminish the salience of race for friendships, beyond the normative needs for ingroup affiliation. That also would be good news. The results of this study suggest that school racial/ethnic diversity can buffer marginalization and a need to turn inward at least at the start of middle school when social networks are beginning to form. However, racial/ethnic diversity is not enough to counteract the school structural forces that shape same-race friendship preference over time, particularly if classrooms remain racially/ethnically stratified based on achievement.

Appendix A

Analytic Model of Aim 1.

Appendix B

Analytic Model of Aim 2.