SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

Objectives. One of the hopes of having diverse campus environments is that the daily interaction with students from different backgrounds will promote interracial understanding and friendship. However, it is not clear to what extent interactions and friendships are multiracial. This article examines the impact of college characteristics, social distance felt toward other groups, and precollege friendship diversity on the formation of interracial friendships in the first year of college.

Methods. Using data from the National Longitudinal Survey of Freshmen, I examine how college characteristics, social distance felt toward other groups, and precollege friendship diversity affects the formation of interracial friendships in the first year of college.

Results. The results show that while precollege experiences and initial attitudes do have an impact on the formation of interracial friendship in college, campus racial/ethnic diversity is also important in predicting friendship heterogeneity. Minorities have higher predicted friendship diversity than whites, but this difference nearly disappears in the most diverse schools due to the interactive effects of school diversity on friendship diversity for white students.

Conclusions. This research provides evidence of the social benefits of assembling a diverse student body, particularly for white students, and can add to the debate over the continuation of affirmative action policies.

Colleges and universities have been at the forefront of discussions about the importance of racial and ethnic diversity. Many have actively tried to recruit minority students to fill their ranks, arguing that the presence of such students benefits not only the individual, but also the university as a community. Because of racial residential segregation, most students entering postsecondary education have grown up in neighborhoods and attended schools that are predominantly composed of others of the same racial and ethnic background. The university environment, therefore, is often the most racially and ethnically diverse environment to which most students have been exposed. This is particularly true for white and Asian students, for whom white-dominated schools, neighborhoods, and friendship groups were nearly always the norm prior to college.

Thanks to several decades of affirmative action policies and other efforts designed to bolster minority enrollments, college campuses have largely been successful in assembling a racially and ethnically diverse student body. Although these policies have come under frequent attack, there is much research to suggest that there are numerous benefits to having racially and ethnically diverse educational settings for both majority and minority students. For instance, Gurin (1999) reports in her testimony prepared for the Supreme Court that students in diverse educational environments are more civically and intellectually engaged and exhibit greater gains in intellectual and academic skills than those in less diverse settings. There is also evidence that this exposure to diversity has longer-term benefits to students. Gurin (1999) finds that students exposed to greater diversity in college report more diversity in their friendship, neighborhood, and work settings nine years after college matriculation. Other studies have similarly found that adults who interacted in interracial settings (especially schools and neighborhoods) as children were more likely to have close interracial friends as adults than their counterparts (Emerson, Kimbro, and Yancey, 2002; Sigelman et al., 1996).

This article furthers the research on the impact of diversity in higher education by examining whether diversity in college fosters meaningful interactions and the development of relationships across racial and ethnic lines, something that has rarely been studied previously. Using data from the multicampus National Longitudinal Survey of Freshmen (NLSF), this study explores two main questions. First, how much does the racial diversity of the student body predict racial diversity in individuals' friendship networks? Of particular interest is whether current contact plays a more important role in friendship composition than prior experiences with and orientations toward other groups. The second part of the article focuses on factors that influence the choice of out-group friends from particular groups by comparing whether structural factors, such as the school's racial composition, have the same relationship across out-groups. For instance, does school racial/ethnic heterogeneity have the same effect on predicting out-group friendships with Hispanic students as with blacks?

Explaining Friendship Patterns

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

Prior research on friendship formation has found that homophily, propinquity, status similarity, and reciprocity are consistently associated with interpersonal attraction. Of these, propinquity and homophily are usually the most salient factors: spatial proximity defines the set from which friends can be chosen and within that space, individuals tend to become friends most readily with those who are most like themselves. This tendency, sometimes referred to as homophily bias, has been shown to be a consistent predictor of friendship choice and helps explain why friendship groups tend to be more racially and ethnically homogeneous than the larger population from which friends could be selected (McPherson, Smith-Lovin, and Cook, 2001). However, even though individuals may not have friendship networks that perfectly reflect the racial/ethnic composition of the setting, proximity to greater diversity does often translate into a more diverse friendship network (Duncan et al., 2003; Hallinan and Teixeira, 1987a, 1987b; Joyner and Kao, 2000; Moody, 2001; Quillian and Campbell, 2003; Sacerdote and Marmaros, 2005; Sigelman et al., 1996).

Joyner and Kao (2000), for instance, examined the impact of school racial composition on the likelihood of having an interracial friend using data from Add Health on students in Grades 7 through 12. Although there were differences by race/ethnicity in the likelihood of having an interracial friend, the authors found that school racial composition accounted for a substantial proportion of this variation. The probability of having an interracial friend declined significantly as the percent of same-race students in the school rose such that students in groups representing at least 50 percent of the school population had a less than 20 percent predicted probability of having a different-race friend (see Joyner and Kao, 2000:Figure 1). Using the same data set, Quillian and Campbell (2003) also find that students in schools that are more racially and ethnically diverse have an increased likelihood of forming a cross-group friendship, but that in-group preferences intensify when students are a small racial minority in a school.

image

Figure 1. In-Group Friendship Preferences by Race/Ethnicity by School*

Download figure to PowerPoint

Other research has pointed to the importance of organizational factors in the formation of interracial friendships. For instance, Moody (2001) examines how organizational aspects of high schools, such as the degree of integration in extracurricular activities and tracking, and demographic aspects of a high school, such as racial composition, size, and socioeconomic status of most students, affect the likelihood that friendships will be segregated by race. He finds that the likelihood of friendship segregation is the highest in schools that are moderately heterogeneous, but declines at higher levels of heterogeneity (Moody, 2001). He also finds evidence that how a school is structured in terms of racial composition within academic tracks, within grade friendships, and interracial mixing in extracurricular activities has implications for friendship integration (Moody, 2001). These results imply that it is not just the overall composition of a school that matters, but also the organizational structure that may promote interracial friendships. Schools or, more generally, contexts, which are structured such that there is interracial contact among relative status equals seem to encourage cross-group friendships. Organizational factors, such as academic tracking, have been found in other research to be important predictors of racial attitudes (Goldsmith, 2004) and interracial friendships among schoolchildren (Hallinan and Teixeira, 1987a; Hallinan and Williams, 1989).

Contact Hypothesis

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

Many researchers have drawn on Allport's (1958) contact hypothesis as a framework for understanding the sociostructural conditions under which out-group relationships form given spatial propinquity. The basic premise of the contact hypothesis is that negative stereotypes about other groups arise through lack of personal contact and interaction between groups. From this perspective, the spatial proximity to sizable out-groups sets the stage for the possibility of interaction between minority and majority groups (Blau, 1977) that could dissolve these prejudices. However, contact alone may not be sufficient for breaking down the barriers of prejudice between groups that are typically highly segregated from one another in U.S. society. Allport (1958) recognized the importance of the quality of contact on prejudice reduction, including the relative status equality of participants, cooperative interdependence of actors, and the express support for cross-group interaction by persons in authority (Forbes, 1997; Pettigrew, 1998). These latter conditions point to the relevance of organizational factors in the formation of interracial friendships, above and beyond any effects of racial composition. The reduction of prejudice through intergroup contact is therefore conceptualized as the mediating factor between social structural factors and the formation of intergroup friendships.

Empirical research has generally supported the premise that intergroup contact leads to improved racial attitudes (Dixon and Rosenbaum, 2004; Duncan et al., 2003; Jackman and Crane, 1986; Pettigrew and Tropp, 2006; Powers and Ellison, 1995; Sigelman and Welch, 1993), although the studies vary in their findings of the specific type, intensity, and duration of contact that has the most impact on attitudes. Furthermore, intergroup contact may have a different effect on attitudes and outcomes such as friendship depending on the specific groups involved (as well as the context of the contact). For instance, groups that have greater feelings of social distance toward each other may be more resistant to forming cross-group friendships than members of two groups that are less socially distant.

A common problem in research testing the contact hypothesis is the possibility of self-selection bias (Dixon, 2006; Pettigrew, 1998; Pettigrew and Tropp, 2006; Powers and Ellison, 1995). The selection bias problem is rooted in the idea that persons who are the most prejudiced might simply avoid intergroup contact altogether, while those who have more intergroup contact may have chosen these situations precisely because they are less opposed to such contact. The impact of contact could therefore be exaggerated in studies that do not contain appropriate controls for selection bias. Pettigrew and Tropp (2006) compare studies in their meta-analysis in which participants either have or do not have choice in their interracial contact. As expected, studies in which participants choose a diverse environment have larger mean effect sizes for contact, but contact is still an important factor in studies in which participants have no choice in contact (Pettigrew and Tropp, 2006). Studies that have considered self-selection bias in their designs have concluded that contact generally outweighs any impact of selection.

In addition to informing the debates over the benefits of diversity in higher education, this study provides several unique contributions to the contact hypothesis and friendship literatures. First, very few of the studies on interracial friendships take place in the college or university context, which is surprising because for many students this is the first significant contact they have with other groups. Furthermore, it is unlikely that students choose a college based on its racial/ethnic composition, reducing the likelihood of self-selection bias. In addition, residential colleges and universities have a number of other unique features that make them ideal for looking at issues of contact and friendship development, such as a clearly defined population from which friends can be chosen, the potential for interracial contact across a variety of formal and informal settings, including the sharing of living spaces, and official expressions of support for interracial diversity by campus authorities. This study is also one of only a few to consider the friendship patterns of multiple racial and ethnic groups. Much of the prior research is built on black/white dichotomies, which do not reflect the increasingly multicultural nature of our society. This study also has the advantage of including students from multiple colleges and universities that range in racial/ethnic composition and other characteristics, which is important for seeing patterns in the effects of campus racial composition on friendship composition. In addition, many studies examining the impact of diversity on friendship choice do not have information on attitudes and experiences prior to interracial contact and therefore may suffer from problems of reverse causality and selection bias (Pettigrew, 1998). This study uses rich background information on students prior to coming to college and on their attitudes upon entering college to provide important controls for selectivity in studying the relationship between campus racial composition and individual friendship composition. The next section discusses in greater detail the sample and measures used to explore friendship diversity on campus.

Data and Measures

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

The data used in this article come from the National Longitudinal Survey of Freshmen, which is a national probability sample of nearly 4,000 white, black, Hispanic, and Asian first-time college students who matriculated into 28 selective colleges and universities in 1999. The first wave of data was conducted via in-person structured interviews that took place in the fall of the first year. Students were asked detailed questions about the neighborhood, school, and family environments they experienced in early childhood (age 6), early adolescence (age 13), and as seniors in high school. Students were also asked about their expectations and goals regarding education, their feelings of self-worth and self-efficacy, and detailed questions about their racial attitudes. A second wave of data was conducted in the spring of the first year to collect information about students' academic and social experiences in college, with updates being collected every spring since. The data for this article come from the first two survey waves collected at the beginning and end of the first year of college. A more extensive description of the study design and instruments can be found in The Source of the River (Massey et al., 2002).

Measuring Friendship Diversity

The measures of friendship composition in this article come from questions in the Wave 2 survey administered in the spring of the freshman year in which students were asked to list how many of their 10 closest friends since coming to college were white, black, Hispanic, or Asian. I turned these into percentages, which are shown in Table 1. Both white and black students on average report that the majority of their friends belong to the same race/ethnic group. Whites have the highest percentage of in-group friends at 76 percent, while for blacks the average student reports that 58 percent of his or her closest friends in the first year are also black. Asians show less in-group preference, with an average of 36 percent of closest friends being from the same group, while Hispanic students are the least homophilious, averaging only 19 percent of friends from their own group. Both Asian and Hispanic students reported about half of their new college friends were white, which is not surprising given that all the campuses in this analysis are majority white. However, black students reported that only a little over a quarter of their first-year friends were white.1 The racial composition of first-year friendship networks roughly approximates the patterns of racial preference and social distance between these racial/ethnic groups that are reported both here and elsewhere (Massey et al., 2002).

Table 1.  Means by Race/Ethnicity for Friendship Heterogeneity Analyses
 WhiteAsianHispanicBlack
  1. Source: National Longitudinal Survey of Freshmen.

First Year Friendship Composition
 Percent white75.94850.89155.60826.796
 Percent Asian12.34736.42112.0518.383
 Percent Hispanic4.6894.69919.4856.964
 Percent black7.0177.98812.85557.857
 Heterogeneity of college friends (0–1)0.3220.4060.4420.373
Demographic Characteristics
 Female %52.40556.51758.07964.379
 Foreign-born parent %10.52163.08750.32820.383
 Parent income ($75,000+) %64.93055.37041.92136.831
Prior Orientations and Experience
 Social distance to whites8.62410.87210.87014.873
 Social distance to Asians13.3029.60514.57616.501
 Social distance to Hispanics14.79115.41210.05513.409
 Social distance to blacks13.74114.65013.8718.522
 Average distance to out-groups (0–30)13.92713.63513.14714.948
 HS white friends (1 or more) %99.591.087.171.9
 HS Asian friends (1 or more) %49.984.146.537.7
 HS Hispanic friends (1 or more) %27.928.867.936.7
 HS black friends (1 or more) %34.835.943.990.6
 Heterogeneity of HS friends (0–1)0.1920.2130.2310.193
School Characteristics
 Percent white59.76860.04858.84460.134
 Percent Asian21.44921.11422.30820.964
 Percent Hispanic8.1418.18.7138.139
 Percent black10.64210.73910.43510.764
 Homophily bias (−1 to 1)0.3850.1980.1180.527
 School racial/ethnic heterogeneity (0–1)0.5340.5330.5420.533
 Private college/university64.83063.29563.75563.269
 All-female college52.40556.51758.07964.379
 Urban campus29.55928.36331.00429.566
 Total student population11,07711,36511,55311,212
 N998959916991

Following Moody (2001), the racial composition of friendships can be summarized into a single measure of friendship heterogeneity that is more useful for cross-group comparisons. This measure of heterogeneity is given by the following formula:

  • image(1)

where N is the total number of friends and ng is the number of people in group g. This measure ranges from 0, when all friends are of the same race/ethnic group, to 0.75 in the four-group case, when the proportions of friends from each race/ethnic group are roughly equal. Given that friends must be whole numbers and four groups cannot be divided equally by 10, in this individual-level implementation of heterogeneity the maximum possible value of this measure will be less than 0.75. The average heterogeneity of college friends is shown by race/ethnic group below the group composition figures I just discussed and summarizes in one number much of what we have already observed. Whites have the lowest average heterogeneity in their friendship group due to a strong in-group preference at 0.32. One way of interpreting this number is that it is the likelihood that two randomly chosen friends will belong to different race/ethnic groups. Blacks have only slightly more heterogeneous friendship networks at 0.37. On the other hand, the friendship network of Hispanics is considerably more heterogeneous with an average value of 0.44. Asians fall in-between with an average heterogeneity score of 0.41. This measure of friendship heterogeneity will serve as a dependent variable in analyses below.

School-Level Measures

These observed patterns of friendship do not tell a complete story about the propensity of individuals to form interracial relationships since these racial/ethnic groups are not represented in equal numbers in the population. If friendship choice was completely random, for instance, we would expect that each individual's friendship group would have the same racial/ethnic composition as the school as a whole. The extent to which this is not the case suggests differential patterns of preference. A convenient summary measure of the racial/ethnic composition of the school campus is school-level racial/ethnic heterogeneity, which applies school-level racial/ethnic proportions to Equation (1), resulting in a single measure for each school. The means for these variables are also shown in Table 1 but do not vary significantly across racial/ethnic groups because each group was sampled in roughly equal numbers at each school. Prior research has shown nonlinear effects of school diversity on friendship formation (Moody, 2001), so quadratic and cubic forms of heterogeneity will also be tested.

The school racial/ethnic composition can also be measured directly by the group percentages that compose the heterogeneity measure. The lower panels in Table 1 show the average percentage of each group on these college campuses.2 Whites are the majority on these campuses, averaging about 60 percent of the population, while Asians represent on average about 21 percent of students on these campuses. Black and Hispanic students are represented in much lower proportions, averaging around 10 percent each. Given the low level of representation of black students on these campuses, the fact that the average black student reports that slightly over 50 percent of his or her friendship network is black suggests a strong in-group bias.

We can gauge more specifically the extent of this in-group friendship bias by calculating a measure that captures the likelihood that members of one racial/ethnic group will befriend members of their own group beyond the level that would be expected based on their group size. Laumann and colleagues (1994) use the following formula to measure this relative in-group or homophily bias:

  • image(2)

where pi is the observed rate of in-group choices by group i, Ni is the size of group i, and N is the total size of the population. The measure ranges from −1 to 1, with −1 representing extreme out-group bias, 0 representing no bias, and 1 representing complete in-group bias. It is calculated in this article for each racial/ethnic group separately for each school in the sample. Although this measure is not used in the analyses, it is a useful descriptive measure to show how in-group bias varies across groups within schools, as well as across schools. The average school-level value for each racial/ethnic group is shown in Table 1 and the values for each group by school are illustrated in Figure 1.

Although all groups on average show some degree of in-group bias (i.e., average values are positive), it is clear from this measure that blacks show the greatest average in-group bias at 0.53. White students have a somewhat lower average in-group preference at 0.39. Asians and Hispanics display the lowest average in-group bias at 0.20 and 0.12, respectively. Figure 1 suggests that there are significant differences within and across schools in the degree of in-group friendship bias. Note that in three of the schools, white students display an out-group preference once the racial composition of the school is taken into account. The extreme across school differences suggests that there may be school environmental differences that help account for friendship selection beyond racial composition alone. We can test this more specifically using the multilevel modeling framework outlined below.

In addition to the racial composition of the school, several other school-level measures that could have an effect on inter-group interactions were considered for the analysis. These measures include whether the college or university is a private institution, whether the campus is all female, whether it is located in a major urban center, and the total size of the student population. Because the size of the student population turned out to be the only factor significantly related to interracial friendship formation in the preliminary models, it was the only one of these additional school-level variables that were retained in the models shown here.

Individual Prior Experiences and Orientations

Beyond the demographic constraints and structural characteristics of each institution, there are also individual-level factors that might explain the propensity of students to befriend members of other groups. In the survey administered as students were entering college, they were asked to rate on a scale from 0 (very close) to 10 (very distant) the degree of closeness they felt to each racial/ethnic group in general and in particular to young men and young women in that group. These three items were combined to form a scale for each racial/ethnic group, such that there are four measures for each individual (including a rating of social distance felt toward their own group). The means for measures are shown in Table 1, as well as a summary measure of the average distance to out-groups calculated for each individual.3

Not surprisingly, each group expresses the lowest average level of social distance toward those of the same race/ethnicity. The remaining expressions of social distance roughly reflect the degree of contact each group is likely to have previously had with one another. For instance, both blacks and Hispanics expressed the highest average social distance toward Asians—a group with whom they were unlikely to share residential space. Limited prior exposure is also a likely factor in explaining the high social distance ratings toward Hispanics by Asian and white students. Note, for instance, the higher percentages of Asian and white students who reported having at least one black friend in high school relative to having at least one Hispanic friend. The measure of average distance to out-groups is shown below the detailed means by race/ethnicity and is used in the models to control for prior attitudes toward other groups.

In addition to measuring attitudes toward other groups upon entering college, the first-wave survey also asks students about the racial/ethnic composition of their 10 closest friends during their senior year of high school. This information was used to construct a measure of high school friendship heterogeneity using Equation (1). For all groups, high school friendship heterogeneity is on average much lower than the degree of heterogeneity reported in friendships made during the first year of college. In some of the models, indicator variables of friendships with members of different groups (white, black, Hispanic, and Asian) are used instead of this measure of high school friendship heterogeneity. A value of 1 is assigned if the student indicates having had one or more friends from a racial/ethnic group in high school and 0 if no friends from that racial/ethnic group were reported. The models also control for gender and parent socioeconomic status, measured by whether family income was greater than $75,000. Although age is often an important factor in friendship selection, it is not considered in this analysis because this sample of first-time freshmen does not have sufficient variation in age.

Methods

This article employs hierarchical linear models to explore the relationship between school heterogeneity, individual characteristics, and friendship diversity in college. Hierarchical modeling strategies are ideal for this analysis for a number of reasons. First, the NLSF is a school-based sample, so any modeling strategy needs to account for possible unobserved heterogeneity among respondents from the same campus. However, for this analysis we potentially want to do more than simply correct for unobserved heterogeneity. Given that this article concerns the predictors of interracial friendships on campus, significant cross-campus variation could be theoretically important. Ideally, we would like to measure the extent to which this variation exists, how much it is reduced after accounting for school racial/ethnic composition and individual characteristics, and how much cross-campus variation remains to be explained. Significant remaining variation in friendship diversity between schools may suggest the presence of unmeasured school characteristics that make an institution more or less conducive to cross-group interaction.

Friendship heterogeneity was modeled in three stages: baseline, individual level, and school level. The final model is shown in Equation (3). Individual-level characteristics discussed above are represented as fixed effects in the equation by WHITE (white vs. minority indicator), FEMALE (female vs. male indicator), INCOME5 (parent's household income >$75,000), FBPAR (at least one parent is foreign born), OUTDIST (average social distance to out-groups), and HSHETER (heterogeneity of 10 closest friends in high school). School-level characteristics are represented in the equation by SCHHET (school racial/ethnic heterogeneity) and LNSIZE (log of the total student population). Nonlinearities in the effect of school heterogeneity are tested with squared SCHHET2 and cubed terms SCHHET3. All variables in the analysis are uncentered. The Level 1 intercept, B0, represents the average level of friendship heterogeneity when all the Level 1 variables are set to zero, while the error term r represents the model's random effects. In the Level 2 model, the intercept from Level 1 is modeled as a function of the Level 2 intercept y00, the Level 2 variables school size and school heterogeneity, and an error term. A cross-level interaction between WHITE and SCHHET is represented in the Level 2 equation for B1, testing whether there is a difference in the slope of school heterogeneity in predicting friendship heterogeneity for white students. The random intercept at Level 1 in the model allows for variation in the slopes across schools and was retained in the model because it was statistically significant. The models shown are estimated using reduced maximum likelihood (REML) in HLM 6.0, which is appropriate for data with few Level 2 predictors because they take into account degrees of freedom lost from estimating the regression parameters (Snijders and Bosker, 1999). However, deviance statistics based on full maximum likelihood estimates (FEML) were used to make comparisons across models.

Level 1 Model

  • image

Level 2 Model

  • image(3)

The first set of models is concerned with predictors of overall friendship diversity, which is a useful summary measure of friendship diversity, but we may also be interested in the propensity of students to form friendships with specific out-groups. Prior research that contained multiple out-groups has suggested that that there is significant variation in patterns of friendship formation between different groups (see, e.g., Joyner and Kao, 2000; Quillian and Campbell, 2003). There is also some evidence that contact may matter differently for different out-groups (Dixon and Rosenbaum, 2004), providing further justification for examining each out-group separately. The second part of this article explores this issue by estimating a series of hierarchical generalized linear models (HGLMs) predicting the likelihood of having at least one out-group friend who is black, Hispanic, or Asian using a Bernolli model in HLM 6.0 (Raudenbush et al., 2004). For each of these models, the group that is the dependent variable is left out of the sample so that the predictions are for out-group friendship only. The general form of the final model is shown in Equation (4). Group X represents the out-group friendship that is being estimated and the prior experiences and orientation variables in the model refer to this group (social distance and high school friends). For the models predicting minority out-group friendships, whites are the reference group and so the dummy variables in the model are for the other two groups in the sample for that model. For example, in the model predicting black out-group friendships, there are indicator variables for Hispanic and Asian, leaving whites as the reference group. Whites are also the reference group in the models predicting having any out-group friends, at least one Hispanic friend, and at least one black friend.

Level 1 Model

P(OUTFRND_X=1)=B0+B1(∼X_g1)+B2(∼X_g2)+B3(FEMALE)+B4(FAMSES)+B5(SOCDIST_X)+B6 (HSFRND_X)+r

Model 1_X=Any Minority Friend, ∼X=Black, ∼X=Asian, ∼X=Hispanic (ref. white)

Model 2_X=Black,∼X=Asian,∼X2=Hispanic (ref. white)

Model 3_X=Hispanic,∼X=Black,∼X2=Asian (ref. white)

Model 4_X=Asian,∼X=Black,∼X2=Hispanic (ref. white)

Level 2 Model

  • image(4)

Predicting Friendship Heterogeneity

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

I explore the determinants of friendship heterogeneity in a series of nested models starting with a null model that gives us an initial estimate of the degree to which friendship heterogeneity varies across schools. Model 1 in Table 2 shows that there are significant between-school differences in the level of friendship heterogeneity and that the estimated mean friendship heterogeneity across the sample is 0.392. The intra-class correlation (ICC) calculated from this model indicates that 8 percent of the observed variation in friendship heterogeneity is accounted for by differences between schools.

Table 2.    Hierarchical Linear Models with Random Intercepts Predicting Heterogeneity in Friendship Network at End of Freshman Year as a Function of Individual and School Characteristics
 Model 1Model 2Model 3Model 4Model 5Model 6
BSEBSEBSEBSEBSEBSE
  1. ***p<0.001;**p<0.01;*p<0.05. Source: National Longitudinal Survey of Freshmen.

Fixed Effects
 Intercept0.392***0.0130.434***0.0150.464***0.0820.379***0.0790.396***0.080−0.1410.277
Level 1 Slopes
Individual Characteristics
 White  −0.068***0.009−0.068***0.009−0.044***0.008−0.113**0.035−0.113**0.035
 Female  −0.030***0.007−0.028***0.007−0.026***0.007−0.026***0.007−0.026***0.007
 Foreign-born parent  0.029***0.0080.028**0.0080.017*0.0070.017*0.0070.017*0.007
 Parent income>$75,000  −0.033***0.007−0.035***0.007−0.032***0.007−0.032***0.007−0.032***0.007
Prior Experiences and Orientations
 Social distance to out-groups      −0.003***0.001−0.003***0.001−0.003***0.001
 HS friendship diversity      0.252***0.0140.251***0.0140.250***0.014
Level 2 Variables
 School heterogeneity    0.388***0.0570.345***0.0550.311***0.0574.135*1.842
 School heterogeneity2          −8.757*4.128
 School heterogeneity3          6.299*2.942
 School size (ln)    −0.027**0.008−0.021*0.008−0.021*0.008−0.018*0.008
Cross-Level Interactions
 White* School heterogeneity        0.130*0.0630.130*0.063
Random Effects
 Between schools (too)0.004*** 0.004*** 0.0009*** 0.0008*** 0.0008*** 0.0007*** 
 Within schools (s2)0.046 0.045 0.0447 0.0409 0.0409 0.0409 
 Proportion reduced variance between  0.03 0.78 0.80 0.80 0.823 
 Proportion reduced variance within  0.04 0.04 0.12 0.12 0.112 
 Deviance−782.6 −931.7 −964.6 −1274.0 −1278.2 −1211.6 

Model 2 adds individual-level predictors of friendship heterogeneity to this baseline model as fixed effects, all of which are statistically significant. The negative and significant coefficient for white indicates lowered predicted levels of friendship diversity for whites compared to minority students. Females report slightly lower levels of friendship diversity than do males (−0.030), as do students coming from a higher-income family (−0.033). Having a foreign-born parent is positively related to friendship heterogeneity (0.029), which is a likely proxy for previous exposure to diversity since children of the foreign born are more likely to interact with an ethnic community than children of native-born parents.

Model 3 adds school-level factors to explore the impact of college racial/ethnic diversity and student population size on friendship diversity.4 School size has a negative impact on individual friendship heterogeneity net of other factors, suggesting that smaller schools may be more conducive to cross-group interaction and friendship formation. School racial/ethnic heterogeneity has a strong linear effect on friendship diversity. For students in a school with a racial/ethnic heterogeneity score of 0.52, the sample mean, there is a predicted 0.202 increase in average friendship diversity after controlling for other factors in the model. The addition of these school-level factors to the model reduces the between-school variation in friendship diversity by 78 percent from the null model, which is a considerable improvement over the previous model that only took into account individual-level factors. The continued statistical significance of the random intercept suggests that there remains significant variation in friendship diversity across schools not accounted for in the model. Some of this unexplained variation may be to due to unmeasured organizational differences across schools that influence cross-group interaction and relationships on campus.

So far we have considered the impact of individual demographic characteristics and school racial composition on the diversity of friendship networks formed during the first year of college. However, it may be more than just the demography of campus life that plays a role in the formation of these cross-group relationships. Students' prior interracial experiences and predispositions toward other groups could also influence their willingness to form friendships across group lines. This idea is tested in Model 4 with the addition of two indicators of prior interracial experiences and orientations, both of which are significantly related to friendship heterogeneity in expected ways. Students who expressed greater social distance to out-groups had less diverse friendship networks in college, while students who had more diverse friendship networks in high school had more diverse friendship networks in college. Since it is difficult to visualize this relationship given that both of these measures are nonlinear, a hypothetical example may help. For a student who reported seven in-group and one of each out-group friends in high school (heterogeneity=0.490), there would be a predicted increase in college friendship heterogeneity of 0.17 (which, if you started with a baseline of seven in-group friends, would be roughly equivalent to gaining approximately two out-group friends). Campus heterogeneity remains a strong predictor of friendship diversity in college. Introducing prior experiences and orientations to the model reduces the unexplained variance across individuals within schools by 13 percent over the baseline model.

The next Model (Model 5) adds a cross-level interaction between white and school heterogeneity to test whether the effect of school heterogeneity on friendship heterogeneity differs for white students compared to minority students. We may expect this differential relationship based on the fact that whites are the majority group on all the campuses in this analysis and therefore will be relatively less likely to come into contact with minorities on less diverse campuses. This interaction is positive and statistically significant, which indicates that as school diversity increases it has an increasingly positive effect on whites' predicted friendship heterogeneity net of other factors in the model. At the highest level of school heterogeneity possible with four racial/ethnic groups (0.75), the predicted differences between whites and minorities in friendship diversity nearly disappear.

The final model in Table 2 adds squared and cubed terms for school heterogeneity to test for nonlinearities in the effects of school heterogeneity on friendship heterogeneity that have been found in previously studies. As shown in Model 6, these variables are all statistically significant with positive first- and third-order terms and a negative squared term. The estimates indicate that school heterogeneity is transmitted more quickly into individual friendship heterogeneity at lower and higher levels of school heterogeneity, but lags at moderate levels of school heterogeneity. Using the coefficients from Model 6, Figure 2 illustrates this nonlinear effect of campus racial heterogeneity on predicted levels of friendship heterogeneity for whites compared to minority students. At the lowest levels of school heterogeneity observed in the sample (0.20), minority students are predicted to have friendship networks that are on average 76 percent more diverse than those of whites. As school heterogeneity increases, the gap between minority and whites in predicted friendship diversity decreases because school diversity is translated into friendship diversity at a faster pace for whites compared to minorities (as signified by the cross-level interaction). However, for both groups there is a lull in the translation of school heterogeneity into friendship heterogeneity in schools that are moderately heterogeneous (school racial heterogeneity scores between 0.433 and 0.529). Once campus heterogeneity exceeds 0.529, the slope of the curve rises more sharply, indicating a stronger relationship between campus heterogeneity and individual friendship heterogeneity and the gap between whites and minorities in friendship heterogeneity nearly disappears. The translation of school racial/ethnic diversity into increased friendship diversity for individual students is consistent with the contact hypothesis.

image

Figure 2. Predicted Friendship Heterogeneity for Whites and Nonwhites by School Heterogeneity

Download figure to PowerPoint

Predicting Specific Out-Group Friendships

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

Thus far we have seen that there is a basic relationship between school racial composition and the extent of racial/ethnic heterogeneity in an individual's friendship networks on campus, even after accounting for prior interracial experiences and social distance felt toward other groups. Although individual friendship heterogeneity is a useful way to summarize racial/ethnic composition of friendships, we may also be interested in whether there are differences in the propensity to form certain types of cross-racial friendships across schools and/or by racial/ethnic group. I first estimate a model predicting whether the student reported any out-group friends and then estimate separate models predicting the likelihood of having formed at least one out-group friendship with black, Hispanic, or Asian students. For each of these models, the group that is the dependent variable is left out of the sample so that the predictions are for out-group friendship only. The results are shown in Table 3.

Table 3.  Hierarchical Generalized Linear Random Intercept Models Predicting the Probability of Having Specific Out-Group Friendships
 Out-Group FriendBlack FriendHispanic FriendAsian Friend
BSEBSEBSEBSE
  1. ***p<0.001;**p<0.01;*p<0.05.^ Percentage reduction in variance calculated using (too null–too current)/too null. Source: National Longitudinal Survey of Freshmen.

Fixed Effects
 Intercept3.886***0.9262.343*0.8840.4671.061−0.6220.771
Level 1 Slopes
 Black−0.0950.132 0.198*0.100−0.533***0.104
 Hispanic2.521***0.2960.448***0.100 −0.1070.104
 Asian1.089***0.168−0.1010.095−0.0470.099 
 Whiterefrefrefref
 Female−0.369**0.118−0.277**0.082−0.1350.083−0.304**0.084
 Parent income >$75,000−0.0340.116−0.369***0.0820.0180.083−0.201*0.086
 Social distance to out-groups−0.053***0.013      
 Social distance to blacks  −0.033***0.009    
 Social distance to Hispanics    −0.029***0.008  
 Social distance to Asians      −0.040***0.008
 Social distance to whites        
 HS friendship diversity2.285***0.245      
 HS any black friends  0.843***0.085    
 HS any Hispanic friends    0.672***0.086  
 HS any Asian friends      0.713***0.085
 HS any white friends
Level 2 Variables
 School racial heterogeneity1.494*0.573−0.0630.6051.994*0.7524.495***0.528
 School size (ln)−0.304**0.088−0.190*0.088−0.195+0.109−0.0860.076
Random Effects
 Between schools (too)0.0363 0.0906*** 0.1703*** 0.0527** 
 Reduction in variance from null^80.9% 11.3% 36.5% 86.7% 
 N3,620 2,844 2,921 2,865 

The first model in Table 3 predicts the probability of reporting having made at least one out-group friend out of the 10 closest friends students mentioned at the end of their first year. Students who had more diverse friendship networks in high school are significantly more likely to report having at least one out-group friend in college, while those who express greater social distance to out-groups at the start of college are somewhat less likely to report having an out-group friend. Both Hispanic and Asian students were significantly more likely to report an out-group friend than were white students. However, there is no difference between white and black students in reporting an out-group friend, a fact that is surprising given the overabundance of potential out-group friends for minority students on these predominantly white campuses.

The next three models predict the likelihood of having at least one out-group friend who is black, Hispanic, and Asian, respectively. Across all the models, there is a significant relationship between having a high school friend from the target out-group, feelings of social distance toward the target out-group, and the likelihood of subsequently forming a cross-group friendship with a member of that group in college. Students who expressed greater social distance toward the target group upon entering college were less likely to report having at least one friend from that group by the end of the first year. For instance, each one unit increase in social distance felt toward blacks resulted in a 0.968 decrease in the odds of having at least one black out-group friend for nonblack students. Prior friendships with members of the target group had the opposite effect. Nonblack students who reported having at least one black friend in high school were over two times more likely to report having made at least one black friend on campus by the end of the first year, net of other factors. These effects were similar in the models predicting out-group friendships with Hispanics and Asians.

The racial/ethnic differences in the likelihood of forming specific out-group friendships echo the descriptive statistics for cross-group friendships shown in Table 1. Hispanics are 1.7 times more likely to report having made at least one black friend on campus compared to white students. Similarly, black students are 1.2 times more likely to report having at least one Hispanic friend on campus compared to white students. Black students, however, are significantly less likely than white students to report having at least one Asian friend and significantly more likely than white students to report friendships with Hispanics. Even though social distance toward the target group is controlled for in the model, these friendship tendencies still roughly mirror the average social distance patterns in the descriptive statistics presented earlier.

The biggest differences across Models 2–4 in Table 3 come in the effects of school-level variables on the likelihood of forming specific out-group friendships. Recall the strong relationship found in Table 2 between school heterogeneity and individual friendship heterogeneity. This relationship might lead us to suspect that greater campus racial heterogeneity would result in an increased likelihood of forming out-group friendships with members of all minority groups given sufficient representation on campus.5 This is true for the models predicting the likelihood of having at least one Hispanic friend (for non-Hispanic students) and at least one Asian friend (for non-Asian students), but this relationship is not significant for predicting the likelihood of having at least one out-group black friend. The model predicting black out-group friends differs from those predicting Hispanic and Asian out-group friends in other ways, too. For instance, there is a great deal more variability across campuses in the baseline models for Hispanics and Asians compared to blacks. This suggests that whatever predicts forming friendships with black students does not differ greatly across schools and is perhaps more related to individual-level factors.

Predictors of Friendship Diversity: Concluding Remarks

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES

This article has explored the relationship between campus racial/ethnic diversity and the racial/ethnic diversity in the friendship networks of college students at the end of their first year, testing the premise that exposure to greater diversity will result in more cross-group friendships. The college setting is an ideal testing ground for examining the contact hypothesis because conditions are theoretically conducive to the formation of interracial relationships: campuses are more racially and ethnically diverse than the settings from which most students come, plus students arrive on campus as relative status equals and in their role as students share the same goal of successfully completing their coursework to get a degree. In addition, academic communities are among the most embracing of diversity in the United States, so the “support by authority” condition of the contact hypothesis is also generally met by college campuses (although there are likely variations in the degree of support for diversity across and even within communities on campus).

As controversies over the use of affirmative action in college admission continue, research on social aspects of campus life such as the present study can play an important role in informing the debate as to the benefits of diversity to both the targets of affirmative action and the wider college community. This study offers considerable evidence that racial and ethnic diversity on campus leads to the formation of cross-group friendships, which presumably entailed meaningful cross-group interaction along the way. The ability of this study to control for students' prior interracial experiences and attitudes toward target groups rules out the argument that it is purely selection that drives choice in interracial friendship formation. However, the fact that students coming to campus with more prior interracial experience more readily formed interracial friendships on campus speaks to the continued need to desegregate primary and secondary education—the subject of which has been the target of recent activity in the U.S. Supreme Court (see Parents v. Seattle and Meredith v. Jefferson).

The current findings support the relationship between school heterogeneity and friendship heterogeneity. As school diversity rises, predicted friendship diversity also increases, although whites still have lower predicted levels of friendship diversity than minorities. However, this relationship shifts as schools become more diverse because there are additional returns to school diversity for whites. Figure 2 illustrated this interactive effect of school diversity on friendship diversity, showing that whites and minorities have nearly equal predicted friendship diversity scores when school diversity is at its maximum. It also illustrated the curvilinear nature of the relationship between campus heterogeneity and individual friendship diversity wherein campus heterogeneity is translated more rapidly into friendship heterogeneity at lower and higher levels of campus heterogeneity. The findings suggest that campuses focusing on creating and maintaining a high level of racial and ethnic diversity in their student body will result in an increased diversity of friendship networks for students from all racial/ethnic backgrounds and will have a particularly strong impact on the friendship networks of white students.

The second part of the analysis predicted specific out-group friendships to test whether school heterogeneity had a similar effect across out-groups. The effects of prior interracial experiences and social distance had similar effects across groups as we had observed in the models predicting friendship heterogeneity. The likelihood of friendship with each out-group was increased for those who previously had a friend from that group and decreased for those who expressed greater social distance toward the out-group. However, the effects of college racial heterogeneity were not consistent across these models. Although college racial heterogeneity had a positive and significant impact on the likelihood of having an out-group friend in general and a Hispanic or Asian friend in particular, it had no impact on the likelihood of having an out-group friend who is black. The preliminary results from these models suggest that while college racial/ethnic diversity does translate into greater friendship diversity, this diversity is often achieved through the addition of out-group friends who are not black.

It is important to keep in mind a few drawbacks to this study. One potential weakness is the measure of friendship heterogeneity. All students are asked to list how many friends out of 10 that they have made since coming to college are from each racial/ethnic group. This is problematic for several reasons. First, it is possible that some students have fewer or more than 10 friends. Also, we have no sense of the closeness of these friends. Some students' listing may be dominated by mere acquaintances, while others may have more close relations. We do not have any additional information about these friends, including whether the friendship nomination is reciprocal, how the students met, or the shared activities in which the friends participate. The latter information would be especially helpful in understanding the specific conditions that promote (or dissuade) cross-group interaction and friendship. It is also possible that having this more detailed information about the content of friendships would help explain away some of the between-school variance that remained in all the models shown in this article. Although this friendship measure is not ideal, work by Smith comparing different measures of interracial friendship suggests that this type of measure is at least less prone to the overreporting of out-group friendships than simply asking whether the respondent has any friends from different target groups (Smith, 2002). Furthermore, this study is limited in showing campus-level effects by the fact that there are only 27 schools in the analysis. However, given the lack of availability of any information about race relations and friendship in large-scale surveys of college students, this study makes a significant contribution to our understanding of interracial friendships in college by providing evidence that the racial/ethnic composition of the campus does indeed have an impact on friendship formation above and beyond individual characteristics.

Footnotes
  1. 1Howard is excluded from the sample for this article because it does not have substantial numbers of students from different racial backgrounds and the only students from this school in the NLSF sample are black.

  2. 2The range in racial composition across campuses is from 4.2 percent to 16.9 percent for blacks, 2 percent to 16.1 percent for Hispanics, 2.8 percent to 55 percent for Asians, and 27.1 percent to 88 percent for whites.

  3. 3The alpha scores for these scales are as follows: social distance to whites 0.778, social distance to blacks 0.734, social distance to Hispanics 0.760, and social distance to Asians 0.784.

  4. 4Several combinations of Level 2 variables were tried, including urban campus location, private school, and all-female school, none of which were significant. In the final models, only school size and school racial/ethnic heterogeneity were retained as Level 2 predictors.

  5. 5Note that the relationship between heterogeneity and having white friends is likely to be nonexistent because, as the majority on these campuses, almost all students have at least one white friend in their network of 10 closest friends.

REFERENCES

  1. Top of page
  2. Abstract
  3. Explaining Friendship Patterns
  4. Contact Hypothesis
  5. Data and Measures
  6. Predicting Friendship Heterogeneity
  7. Predicting Specific Out-Group Friendships
  8. Predictors of Friendship Diversity: Concluding Remarks
  9. REFERENCES
  • Allport, Gordon W. 1958. The Nature of Prejudice. Garden City, NY: Doubleday.
  • Blau, Peter. 1977. Inequality and Heterogeneity: A Primitive Theory of Social Structure. New York: Free Press.
  • Dixon, Jeffrey C. 2006. “The Ties That Bind and Those That Don't: Toward Reconciling Group Threat and Contact Theories of Prejudice. Social Forces 84:21792204.
  • Dixon, Jeffrey C., and Michael S. Rosenbaum. 2004. “Nice to Know You? Testing Contact, Cultural, and Group Threat Theories of Anti-Black and Anti-Hispanic Stereotypes. Social Science Quarterly 85:25780.
  • Duncan, Greg J., Johanne Boisjoly, Dan M. Levy, Michael Kremer, and Jacque Eccles. 2003. Empathy or Antipathy? The Consequences of Racially and Socially Diverse Peers on Attitudes and Behaviors. Working Paper Series. Chicago, IL: Joint Center for Policy Research.
  • Emerson, Michael O., Rachel Tolbert Kimbro, and George Yancey. 2002. “Contact Theory Extended: The Effects of Prior Racial Contact on Current Social Ties. Social Science Quarterly 83:74561.
  • Forbes, H. D. 1997. Ethnic Conflict: Commerce, Culture, and the Contact Hypthesis. New Haven, CT: Yale University Press.
  • Goldsmith, Pat Antonio. 2004. “Schools' Role in Shaping Race Relations: Evidence on Friendliness and Conflict. Social Problems 51:587612.
  • Gurin, Patricia. 1999. The Compelling Need for Diversity in Education. Available at 〈http://www.vpcomm.umich.edu/admissions/research/〉.
  • Hallinan, Maureen T., and Ruy A. Teixeira. 1987a. “Opportunities and Constraints: Black-White Differences in the Formation of Interracial Friendships. Child Development 58:135871.
  • Hallinan, Maureen T., and Ruy A. Teixeira. 1987b. “Students' Interracial Friendships: Individual Characteristics, Structural Effects, and Racial Differences. American Journal of Education 95:56383.
  • Hallinan, Maureen T., and Richard A. Williams. 1989. “Interracial Friendship Choices in Secondary Schools. American Sociological Review 54:6778.
  • Jackman, Mary R., and Marie Crane. 1986. “‘Some of My Best Friends Are Black …’: Interracial Friendship and Whites' Racial Attitudes. Public Opinion Quarterly 50:459586.
  • Joyner, Kara, and Grace Kao. 2000. “School Racial Composition and Adolescent Racial Homophily. Social Science Quarterly 81:81025.
  • Massey, Douglas S., Camille Z. Charles, Garvey F. Lundy, and Mary J. Fischer. 2002. The Source of the River: The Social Origins of Freshmen at America's Selective Colleges and Universities. Princeton, NJ: Princeton University Press.
  • McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001. “Birds of a Feather: Homophily in Social Networks. Annual Review of Sociology 27:41544.
  • Moody, James. 2001. “Race, School Integration, and Friendship Segregation in America. American Journal of Sociology 107:679716.
  • Pettigrew, Thomas. 1998. “Intergroup Contact Theory. Annual Review of Psychology 49:6585.
  • Pettigrew, Thomas F., and Linda R. Tropp. 2006. “A Meta-Analytic Test of Intergroup Contact Theory. Journal of Personality and Social Psychology 90:75183.
  • Powers, Daniel A., and Christopher G. Ellison. 1995. “Interracial Contact and Black Racial Attitudes: The Contact Hypothesis and Selectivity Bias. Social Forces 74:20526.
  • Quillian, Lincoln, and Mary E. Campbell. 2003. “Beyond Black and White: The Present and Future of Multiracial Friendship Segregation. American Sociological Review 68:54066.
  • Raudenbush, Stephen W., Anthony S. Bryk, Yuk Fai Cheong, and Richard T. Congdon. 2004. HLM 6: Hierarchical Linear and Nonlinear Modeling. Lincolnwood, IL: Scientific Software International, Inc.
  • Sacerdote, Bruce, and David Marmaros. 2005. How Do Friendships Form? NBER Working Paper W11530. Available at 〈http://ssrn.com/abstract=776569〉.
  • Sigelman, Lee, Timothy Bledsoe, Susan Welch, and Michael W. Combs. 1996. “Making Contact? Black-White Social Interaction in an Urban Setting. American Journal of Sociology 101:130632.
  • Sigelman, Lee, and Susan Welch. 1993. “The Contact Hypothesis Revisited: Black-White Interaction and Positive Racial Attitudes. Social Forces 71:78195.
  • Smith, Tom W. 2002. “Measuring Inter-Racial Friendships. Social Science Research 31:576.
  • Snijders, Tom, and Roel Bosker. 1999. Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. Thousand Oaks, CA: Sage Publications.