Issuu on Google+

Beyond Black and White: The Present and Future of Multiracial Friendship Segregation Author(s): Lincoln Quillian and Mary E. Campbell Source: American Sociological Review, Vol. 68, No. 4 (Aug., 2003), pp. 540-566 Published by: American Sociological Association Stable URL: Accessed: 11/05/2010 18:37 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact

American Sociological Association is collaborating with JSTOR to digitize, preserve and extend access to American Sociological Review.



Universityof Wisconsin-Madison

Universityof Wisconsin-Madison

How will racial divisions in studentfriendshipnetworkschange as U.S. schools incorporatea growingAsian and Hispanicpopulation?Drawing on theories of race in assimilationprocesses and the effects of relativegroup size on intergrouprelations, several hypothesesare developedto address this question.These hypotheses are tested using data onfriendships among studentsin grades 7 to 12 from the National LongitudinalStudyof AdolescentHealth. Keyfindings are that (1) cross-race friendships includingAsian and Hispanic studentsare more commonthan those betweenwhite and black students,but race and Hispanic backgroundhave significant influenceson studentfriendships thatpersist over immigrantgenerations;(2) black or white racial identificationsare stronglyassociated with thefriendship choices of Hispanic students;(3) cross-racefriendships increase with school racial diversity;and (4) own-groupfriend selection intensifiesfor studentsin small racial minoritiesin a school. The resultssupporttheories of racially segmentedpatternsof assimilationin primarygroup relationsand suggest that studentsin small racial minoritiesseek to maintaina friendship networkincludingseveral own-racefriends. Implicationsare discussed.


LTHOUGHstatements to the court in Brown v. Board of Education proposed desegregation as a policy to increase the achievement and self-esteem of black children, contemporary discussions focus more on fostering improved racial relations and reducing minority exclusion as majorrationales for desegregation (Wells and Crain 1994). Central to these goals is the proposition that racially desegregated schooling will tend to generate cooperative, equal-status contact across racial lines and increase inDirect correspondenceto Lincoln Quillian, Departmentof Sociology,Universityof WisconDrive,Madison,WI53706 sin, 1180Observatory This researchwas sup( portedby the Institutefor Researchon Poverty, An earlyverUniversityof Wisconsin-Madison. sion of this paperwas presentedat the August 2001 meetingof the AmericanSociologicalAssociationin Anaheim,California.We benefited from the comments of Stephen Plank, Ted Mouw,andtheASRreviewers.Thisresearchuses data from the National LongitudinalStudy of 540

terracial friendship. Supporters of desegregation programs argue that the increase in cross-race interactionand friendships resulting from desegregation will improve confidence in interracial interaction, reduce stereotypes, and in the long run, contribute to a more integrated and equal society. The ability to manage social relationships across racial lines may be especially importantfor the future socioeconomic status of many minority studentsbecause their entranceinto many high-status occupations will require regular interactions with white teachers, coworkers, and bosses (Dawkins and Braddock 1994). Consistent with this line of argument,past social science research has generally supported Allport's (1954) "contact hypothAdolescentHealth(Add Health),a programdesigned by J. RichardUdry and Peter Bearman andfundedby grantP01-HD31921fromthe National Instituteof Child Healthand HumanDevelopmentto the CarolinaPopulationCenter, Universityof NorthCarolinaat ChapelHill, with cooperativefundingfrom 17 otheragencies.


VOL. 68 (AUGUST:540-566)


esis:" When cooperative, equal-status contact across racial lines is achieved, interracial attitudes tend to improve, and comfort in interracial encounters tends to increase (Powers and Ellison 1995; Sigelman and Welch 1993). This argumentalso rests, however, on the debatable claim that racially diverse schooling will substantially increase the extent of positive cross-race contact and friendship. Past studies of black-white schools found that the experience of many students is one of racial separation rather than integration because the social activities and friendships in many numerically integrated schools are highly segregated by race (Clotfelter 2002; Hallinan and Williams 1989). Amidst the current retreat from active efforts to desegregate schools, the increasing racial diversity of American society is altering the terms of the debate about racial segregation in schooling. Since 1960, the proportion of Hispanics in the nation's public school population has more than tripled, and an "almost invisible" minority of Asian students has grown to 4 percent (Orfield and Yun 1999). The result has been a gradualdecline in the percentage of whites in the average white student's school and the emergence of increasing numbers of schools with multiracial populations. Because most prior research has focused on black-white desegregation, little is known about racial relations in these emerging multiracial schools. Will race be as important a barrier to friendship in these increasingly Hispanic and Asian schools as it has been in black-white schools? How will friendship relations shift as the share of Asians and Hispanics in schools increases? These questions are importantboth for evaluating school racial diversity as a goal of educational policy and for understanding racial relations as American society becomes more diverse. After reviewing past studies of cross-race student friendship formation, we develop a set of hypotheses to address these questions based on two lines of sociological theory. First, we consider theories of adaptation among Asian and Hispanic' immigrants,ask1 In our discussionwe use racial and ethnic termsthat are consistentwith those used in the National Longitudinal Study of Adolescent



ing whether members of these "new" immigrant groups will follow a patternof assimilation in which racial divisions fade in importance over time, or whether a model of friendship assimilation that accounts for racial difference is required. Second, we draw on theories addressing how the relative sizes of racial groups in a setting influence the structure and content of cross-group relations. We then test our expectations using sociometric data on the friendships of more than 70,000 students in more than 130 schools from the National Longitudinal Study of Adolescent Health (Add Health). PAST RESEARCH AND THEORY STUDIES OF FRIENDSHIP IN AMERICAN SCHOOLS

Theories of friendship have identified many bases of interpersonal attraction, including homophily, propinquity,status, and reciprocity (HallinanandWilliams 1989; McPherson, Smith-Lovin, and Cook 2001). Particularly relevant for understandingfriendship in racially diverse schools are homophily-the tendency to form friendships with similar others-and propinquity-the tendency to form friendships with others who share the same social situation. Because the settings from which friends are chosen are usually relatively homogeneous, homophily and propinquity usually combine to produce high levels of racial homogeneity in social networks (Feld 1982). In racially integratedsettings, on the other hand, propinquity and homophily work to opposite purposes: Homophily in selecting friends promotes continued racial segregation of friendship relations, while propinquity promotes increased cross-race friendships. Studies of friendship relations among students in black-white schools reveal that both propinquity and homophily are important. Supportingpropinquity,most studies find an increase in black-white friendships as the shares of students who are black and white in schools or classrooms move toward parity (Hallinan 1982; Hallinan and Smith 1985; Patchen 1982; Schofield and Sagar Health(Add Health),which is the basis of our analysis.




1977; but see Gerardand Miller 1975). Supporting homophily, even in numerically integrated black-white schools interracial friendships remain far less frequent than same-race friendships. Hallinan and Williams (1989), in a comprehensive analysis of racial homophily among secondary school students, find that students are only onesixth as likely to realize a possible friendship with a schoolmate of a different race as with a schoolmate of the same race. Because court-ordered school desegregation focused on black-white segregation, most analyses of interracial friendship patterns in schools in the 1970s and 1980s examined schools in which most students were black or white. We know of only two largescale sociometric studies that have considered multiracial student populations. Joyner and Kao (2000) consider the relationshipbetween the same-race share of the school population and the probability of having an interracialfriend for members of several racial groups. Moody (2001) examines overall levels of friendship segregation, dividing respondents into same-race and other-racecategories and using a single index for school racial diversity. Consistent with the blackwhite studies, these two studies find increasing rates of cross-race friendship with increased racial diversity, although the increase is not linear. While these studies take the importantstep of incorporatingstudents who are not black or white, neither study reports results for pairings of particular racial groups (e.g., white-black, white-Asian, black-Asian, etc.). Because both studies employ the racial categories of same-race and other-race, their results cannot address our central question of changes in student relations as the shares of Asian and Hispanic students increase. Theory and empirical studies suggest that segregation may vary for different combinations of racial groups, and in particularthat Asian and Hispanic students may have patterns of friendship segregation that differ from white-black patterns. ASSIMILATION THEORIES FOR FRIENDSHIP FORMATION

Relations between black students and white students face the difficulties of black-white

relations in America generally: high levels of black-white socioeconomic inequality, a history of racial oppression, and the persistence of deeply internalized stereotypes and racial resentments. Given the persisting significance of the black-white division in other areas of life, it is not surprising that these divisions remain evident in the social relations among students. Schools, of course, are not isolated from racial relations in the wider society. By contrast, the level of social segregation for Hispanic students and Asian students from white and black students is less clear, reflecting the ambiguous location of Hispanics and Asians within the traditionally dual American racial system. The history of Hispanics and Asians as voluntary immigrant minorities suggests that their friendship relations with whites may follow a substantially different pattern than that between whites and blacks (Lieberson 1961). Because many Asian Americans and Hispanic Americans are recent immigrants, the dominant frameworks for understandingtheir social positions have been derived from theories of immigrant adaptation,which has led to two schools of thought on the likely future position of these groups in American society.2 One view is that Asian Americans and Hispanic Americans are on a path leading to assimilation with white patterns. In this view, Hispanic immigrants and Asian immigrants are following the predictions of traditional assimilation theory, in which immigrant groups progress from having the characteristics of immigrants on arrival in the United States toward the characteristics of the "core group"in the host society (Gordon 1964; Warnerand Srole 1945).3 Traditional 2 We do not investigatevariationsin nationaloriginamongHispanicandAsianimmigrantsbecause includingnationalorigin in our analysis would demandanotherresearchproject. Evidence suggests that the panethniccategoriesof

Hispanic American and Asian American are of greater significance than is national origin for many later generation Asian and Hispanic Americans (Tuan 1998). 3 Few empirical studies of assimilation examine friendship, probably because data are limited. Friendship, however, has been included in assimilation theory. In Gordon's (1964) influential


assimilation theory usually associates this "core group" with the white middle class. Scholars who defend this argumentview discrimination against Asians and Hispanics as fundamentallysimilar to that encounteredby earlier generations of white immigrants,suggesting that the non-European origins of these groups should not have a lasting influence that fundamentally alters assimilation processes (Glazer 1993; Sowell 1981). Applying these ideas to friendship, traditional assimilation theory suggests the following proposition: Hypothesis 1, Traditional Assimilation: For Asian and Hispanic Americans, owngroup preference in friendship will weaken across immigrant generations, with their cross-race friendship patterns in later generations becoming similar to the friendship patterns among whites. Other scholars disagree with these predictions, contending that traditional assimilation theory, based on a sophisticated form of popular melting-pot ideas thought to hold among early twentieth-century white immigrants, is limited in its application to contemporary immigration (for a review, see Rumbaut 1999).4 Several differences between past and recent generations of immigrants are noted, and many accounts particularly emphasize the role of racial discrimination arising from the non-European origins of much contemporary immigration (see esp. Chan and Hune 1995; Johnson, Farrell, and Guinn 1999). Discrimination directly influences immigrant adaptation by excluding the target group from certain avenues of assimilation, and in addition, it tends to correspondingly increase solidarity among the target group. The definition of a new immigrant group as racially different is especially likely to have lasting consequences if the definition leads to differential treatmentthat persists across generations. An influential and systematic statement that serves as an example of this line of stagesof assimilation,assimilationin friendship is a formof "structural assimilation." 4 Some scholarshave arguedthat traditional assimilationtheory applies imperfectlyeven to early generations of white immigrantgroups (e.g., GlazerandMoynihan1963).



thought appears in the discussion of "segmented assimilation" by Portes and Zhou (1993) and Portes and Rumbaut (1996, 2001). Their model primarily considers assimilation in socioeconomic status, but several ideas from the theory are applicable to friendship as well. A central point of segmented assimilation is the view that, in contrast to traditional assimilation theory, there is no single "core group" with which immigrants merge. Instead, they propose that assimilation theory must consider the question "assimilation to what?"Their description of multiple paths of assimilation, influenced partly by race, is a view we adapt in proposing "alternative"paths of immigrant adaptation in friendship relations. In addition to the "traditional"assimilation path-toward the characteristics of the white middle class-segmented assimilation theory suggests two alternatives. First, an immigrant minority group that receives a sufficiently hostile reception may become permanently segregated from the white middle class, similar to the continuing segregation between most whites and blacks. We propose this as a distinct path that may characterize the evolving friendship relations of Hispanic Americans and Asian Americans: Hypothesis 2a, Strong Friendship Segmentation: In later generations, the friendship patterns of Asian Americans and Hispanic Americans will remain highly segregated from whites and other groups, similar to the continuing segregation between whites and blacks. A second alternative is that the social relations of immigrant minorities will follow an intermediate path, one that leads to moderately high levels of contact with persons of other races, but remains not fully integrated with either blacks or whites. This pattern may result from less extreme forms of discrimination by the host society and/or from successful efforts of the immigrant group to maintain solidarity. Portes and Rumbaut (1996) discuss this option as one that might lead to rapid economic assimilation while the group simultaneously and deliberately preserves the immigrant culture and ethnic social ties-a choice they call "selective acculturation."We propose this as




a final possible pattern of friendship relations for later generations: Hypothesis 2b, Selective Friendship Assimilation: In later generations, Asian Americans and Hispanic Americans will retain own-group preferences in their friendship selections, but the two groups will show a greater tendency toward friendships with whites or blacks than whites or blacks have for each other. We believe that these two alternative paths (Hypotheses 2a and 2b) are more likely to characterize those of Asian immigrants and Hispanic immigrants than the "traditional" path of assimilation to white patterns. Our expectations, however, are further differentiated by the specific circumstances of Asian and Hispanic Americans. Asian Americans make up the racial group most often described as being on a path to assimilation with white patterns. Supporters of this view (e.g., Glazer 1993; Hacker 1993) note that Asian Americans often live in neighborhoods with many whites (Massey and Denton 1987), intermarrywith whites at relatively high rates (Qian 1997), and are close to or above white levels on many measures of socioeconomic status (Hirschman and Snipp 1999). Critics of this view emphasize that later generations of Asian Americans remain subject to anti-immigrantsentiment and racial discrimination. They also point out persisting Asian-white gaps on some measures of socioeconomic status (Chan and Hune 1995; Tuan 1998), and strong ethnic social capital ties that help Asians to maintain separate and distinct ethnic cultures (Bankston and Zhou 1995; Zhou and Bankston 1994). Our overall expectation is that later generations of Asians are likely to be relatively close to white patterns but that they will retain some in-group favoritism, most closely following the course of selective assimilation. Hispanic Americans are less often described as taking a path toward full assimilation with whites than are Asians because persistent gaps remain in socioeconomic status between Hispanics and non-Hispanic whites (Hirschman and Snipp 1999). The importance of race in assimilation theory, however, raises important additional ambiguities in the case of Hispanics. "Hispanic"

is an ethnic category that indicates a common ancestral connection to Latin America and often to the Spanish language as well. Behind the uniformity implied by this term is the reality of racial diversity among Hispanic immigrants,who might phenotypically appear to be white, black, or different than either whites or blacks.5 Studies that have distinguished Hispanics by skin tone (or between predominantly "white" and predominantly "black" Hispanic national origin groups) often find sharp differences by race in residential environments and socioeconomic outcomes (Denton and Massey 1989; Murguia and Telles 1996). The influence of race on the experiences of immigrant minorities suggests that while the panethnic category Hispanic may have an importanteffect on friendship formation, common racial identification may also be significant to the racial friendshippatternsof Hispanic respondents. Consistent with our expectation that racial phenotype has consequences for immigrant adaptationin friendship, we propose a third hypothesis for the friendship relations of Hispanic students: Hypothesis 3, Racial Differences among Hispanics: In later generations, friendship selection among "white Hispanics" will resemble white patterns,and friendship selection among "black Hispanics" will resemble black patterns. THE EFFECTS OF RACIAL CONTEXT

To this point, we have focused on the consequences of the shift toward increasingly Asian and Hispanic student populations, arguing that Asian and Hispanic students are likely to occupy a structurallydifferent position in friendship relations than do white or black students.A second factor that is systematically changed by the increasing Asian and Hispanic student populations is school 5 Phenotypical variations that can be viewed as racial also exist among Asian American groups from different national origins (e.g. Filipino, Japanese, Thai). We expect the black-white dimension among Hispanic immigrants will be especially salient in the U.S. context, however, because it corresponds closely to black-white divisions among non-Hispanics.


racial composition. Several established theories in racial and ethnic relations suggest that the relative size of racial groups in a setting has important consequences for intergroup contact and relations. In the case of cross-group social ties, Blau's (1977) macrostructural theory provides a basic proposition. We have previously alluded to this proposition as "propinquity": Hypothesis 4, Propinquity: The extent of cross-race friendship between members of racial groupA and racial group B will be in direct proportion to the relative size of A to B within the school. As B increases in relative size, the share of friendship nominations of group A to members of group B will correspondingly increase. Propinquity reflects the fact that the frequency of day-to-day contacts among strangers strongly predicts the likelihood of friendship formation, as demonstrated most convincingly by the college dorm experiment of Festinger, Schachter, and Back (1950). Most studies of black-white relations in classrooms and schools support this proposition in the case of school friendship patterns,and thus we expect that this will hold for Asian and Hispanic students as well.6 Propinquity implies that the total number of cross-group friendships in a setting will be maximized when the groups are equal in size. A different perspective comes from the literature on the psychology of friendship and ethnic solidarity. Social psychological discussions view friendship as providing a means of sharing, coping with stress, and finding acceptance. These functions are especially importantduring adolescence when friendships become more intimate and play a large role in the development of self-concept. Applying these friendship functions to racial friendship relations, Tatum (1987, 1999) argues that a network including sev6 Others have pointed out that cross-race friendshipmight decline in integratedsettings becauseof the increasingthreatperceivedby the dominantgroupfromthe subordinategroup(St. John and Lewis 1975). Because past studiesof black-whiteschools have consistentlysupported

propinquity effects, we doubt this prediction will hold in the case of friendship relations.



eral racially similar peers is importantto fulfill needs for social support and advice in dealing with racial issues, especially when students are faced with experiences of racial discrimination or prejudice. Bankston and Zhou (1995) make a compatible point in their study of second-generationVietnamese youth. They find that strong ties with coethnics provide a resource that supports an orientation toward academic achievement and describe efforts of immigrant parents to maintain these ties. Although not noted by these authors, efforts to maintain a network of co-racial and co-ethnic friends have implications for composition effects on friendship formation. If students act to maintain a network of samerace peers when they are in a small minority, then the strength of same-race selection for members of racial groups that are small minorities within their school must increase, suggesting the following sociometric pattern: Hypothesis 5, Social Support: As minority groups become small in size, they will increasingly favor the available samerace peers in friend selection to achieve the goal of maintaining a friendship network with several same-race friends. This proposition reminds us of the positive functions of co-racial friendship in a society in which race is highly consequential, especially for students who are in small numerical minorities. When a group is heavily outnumberedon the basis of a highly salient social categorization, we expect that solidarity within the outnumbered group is likely to increase. DATA We examine friendship matching using data from the in-school survey of the National Longitudinal Study of Adolescent Health (Add Health).7The survey is a self-administered questionnaire given to students in 144 schools in 80 U.S. communities between September 1994 andApril 1995. The schools 7

Persons interested in obtaining data files

from Add Health should contact Add Health, Carolina Population Center, 123 West Franklin Street, Chapel Hill, NC 27516-2524 (http://




were selected to representthe universe of all schools in the United States with students in grades 7 to 12. All students present in the schools on the days the surveys were administered were included in the sample, resulting in a total of 90,118 completed questionnaires. Because the student questionnaireincluded all students in each school, many students named as friends are also members of the sample. This allows us to determine the characteristics of friends based on their responses to the survey. For more on the Add Health Study, see Bearman,Jones, and Udry (1997).8 We discarded data from seven schools for which almost no friendships with other students in the school were reported (N = 5,043),9 and we eliminated data from students who went to schools in which no school administratorsurvey was completed or who did not answer survey questions about Hispanic origin or gender. These deletions left us with a base sample of 72,957 respondents. In addition, there are a large number of cases that have missing data (N = 21,276) on at least one of five variables used in our analysis: education of mother, occupation of mother, grade-point average (GPA), mother's place of birth (native or foreign), and student's place of birth (native or foreign). Following a suggestion by Little (1992), we imputed values for these variables based on regressions with other covariates used as explanatory variables. We also estimated all results excluding cases missing on any of these variables. This change does not alter the substantiveconclusions of our analysis. 8 FollowingWinshipand Radbill(1994), we do not use the Add Healthprobabilitysampling weightsin modelsbecausethe samplingweights arefunctionsof the independentvariablesin our analysis. We also estimatedthe basic models with weights, and this changedid not alterour substantivefindings. 9 In seven schools,studentsnamedon average less than one other studentin the school as a friend. Studentscould indicatezero friendsby not writingany down,but the very low average numberof friendsfor these schoolsprobablyreof the survey flects an errorin the administration or in the coding of the data,and thus we eliminatedthese schools.


The Add Health in-school questionnaireasks students to list their five best male friends and, in a separate question, their five best female friends, including girlfriends and boyfriends.10Students wrote in the names of their five best friends as placeholders, and then they were asked to look up their friends on a roster of students in their school and a sister school and write in the identifying code matching that friend. A separate code was written in to indicate that the friend did not go to either the sister or sample school. The friendship question does not make it possible to distinguish between other-sex friendships and other-sex romantic relationships. Because we believe that the dynamics of romantic relationships are probably different than friendships, we analyze only same-sex friendships. As Hallinan (1974) discusses, a disadvantage of using fixed-choice questions to measure friendship is that respondents who want to name more than five friends cannot. Nonetheless, in a comparison of sociometric methods of friendship selection, Schofield and Whitley (1983) find this to be a valid method of friendship identification. They make the importantpoint, however, that this method tends to identify close friends only, leaving out "friendly acquaintances."Segregation among close friendships is likely to be higher than that among acquaintances. Although we agree that weak friendship ties (like acquaintances) are important to study, we also think it is importantto study strong friendship ties because those ties are more likely to have lasting effects on attitudes and behaviors (Powers and Ellison 1995). MEASURES OF RACE

Like the 2000 Census, the Add Health inschool survey presents a series of racial categories and allows respondents to choose more than one. The possible racial catego10The exactwordingof the questionsis: "List yourclosest (male/female)friends.Listyourbest (male/female)friend first, then your next best friend, and so on. (Girls/Boys) may include (boys/girls)who arefriendsand(boyfriends/girlfriends)."




ries are: white, black, Asian or Pacific Islander, Native American, and other. In a separate question, respondents are asked if they are of Hispanic descent.11 This data raises the issue of defining racial categories for students who check more than one racial category. We initially considered a conventional coding system that considered some racial categorizations as "dominant," for instance counting everyone who checked "black" and another race as black. We were concerned, however, thatmultiracialstudents may differ systematically in their friendship patterns from monoracial students. Instead, we constructed our primaryracial categories of adolescents who chose only one race. In all, we use eight categories of race by Hispanic origin: white, black, Asian, white Hispanic, black Hispanic, other Hispanic, other race, and multiracial. The white, black, and Asian categories all include white, black, and Asian single-race students. Likewise, white Hispanic and black Hispanic are Hispanic studentswho indicate they are white or black, and no other race. The "other Hispanic" category includes two patterns of response to the race questions. First, we include respondents who chose Hispanic and "other" as their race, who make up about 36 percent of students who are of Hispanic descent. Second, we include students who indicated Hispanic but did not check any racial category, who make up 28 percent of persons who indicated Hispanic background.12Our guess is that students in the "other Hispanic" category think of Hispanic as their race rather than thinking of Hispanic as an ethnic category distinct from race. Finally, we include students who indicate they are Native American or "other"race in an "other race" category; students who choose more than one race are included in a multiracial category. These last two categories constitute less than 10 percent of the Add Health sample and are not analyzed at length below. In anotherstudy we plan to in-

All student characteristics are based on selfreports from the school survey. Grade-point average (GPA) of the student is based on the student's grades in the most recent period for English/language arts, mathematics, history/ social studies, and science. Grades were averaged to create a single 0 to 4 GPA scale. Mother's years of education completed was recoded from questions asking about highest grade and degree obtained.13 Mother's occupation is likewise recoded from a 20category typology into a dummy variable for whether or not the mother works in a professional occupation. Immigrant generation is coded based on place of birthof the student and the student's mother, as reportedby the student on the inschool survey. First-generation students are those born outside the United States (immigrants). Second-generation students are born in the United States to a foreign-born mother. The third-plus generation consists of native-born students whose mothers were also born in the United States. School characteristicsare based on school administratorreports, except for school racial composition, which is calculated from student self-reports of their race from the school survey.

'1The racequestionis: "Whatis yourrace?If you are of morethanone race, you may choose morethanone."The Hispanicquestionis: "Are you of Hispanicor Spanishorigin?" 12Only about1 percentof non-Latinorespondentsdid not checkanyracialcategory.

13We comparedstudentand parents'reports for the subsampleof studentswhoseparentswere surveyed by Add Health. Child's report and mother'sreportof mother'seducationcorrelate .75; child's report of father's education and

vestigate the friendship patterns of multiracial students. Descriptive statistics on friend nominations and the racial composition of students and schools in the survey are shown in Appendix A. STUDENT AND SCHOOL CHARACTERISTICS


The simplest method of examining interracial friendship is to use a dependent variable to indicate either whether a student has any cross-race friendships (Joyner and Kao




2000) or the number or share of friends who are of anotherrace. A majorlimitation of this approach,however, is that it allows for consideration of only a single characteristic of the friend chosen in friendship formation. This modeling strategy is suited to considering how characteristics of the individual choosing a friend influence the choice of a same-race or other-race friend, but it provides no way to examine the simultaneous matching on the multiple dimensions that actually characterizefriendship formation. To account for the multiple bases of friendshipmatching, we employ an approach previously used by Hallinan and her colleagues to study school friendship segregation (Hallinan 1982; Hallinan and Williams 1989). The approach was recently extended by Wassermanand colleagues, whose models for social networks, called P* models, are based on graph-theoreticmodels of network dependence (Anderson, Wasserman, and Crouch 1999; Wasserman and Pattison 1996). Although based on a more developed formal model than Hallinan's models, in practice the approaches are similar. Our models generally follow the P* model approach, with small modifications as appropriate for the specific goals of our analysis. In P* models, ratherthan taking individual students as the cases, the cases in the analysis are all possible dyadic pairs of respondents who are in the same school and thus who could become friends. By coding dyads in which one student named the other as a friend as 1 and other dyads as 0, friendship can be modeled as an outcome using logistic regression. Characteristics of the two students, including relative characteristics(e.g., whether both students are of the same race, and the difference between studentsin socioeconomic status) serve as independent variables. In the text that follows, senders are the respondents in our study who are naming their five best friends; receivers are persons who could be chosen as friends by senders-students in the same school. In practice, the first step to estimating the P* model is to create dyads that representall

possible friendship pairs within a school. If there are N students within a particular school, there are a total of [Nx(N-1)]/2 within-school dyads for that school. Each dyad is representedtwice in the data because each individual appearsboth as a sender and a receiver. Note that the numberof dyads increases geometrically with school size, becoming extremely large for large schools: A school of 100 studentswill be representedby (100 x 99) = 9,900 dyads. This data structure implicitly controls for opportunitiesfor contact among persons of different race in the school, as interracial dyads are represented in proportionto the racial diversity of the student body. When all within-school dyads are combined across the approximately 130 schools in our sample, the total number of possible friendship dyads is more than 5 million. To analyze the data efficiently, we, like Hallinan and Williams (1989), choose a sample of dyads. We select all dyads in which one individual selected the other as a friend and a share of nonfriendship dyads. We select nonfriend dyads in proportions stratifiedby the race of the sender: 5 percent for non-Hispanic whites, 15 percent for nonHispanic blacks, 20 percent for Asians, 50 percent for white Hispanics, 50 percent for black Hispanics, and 20 percent for other Hispanics. We then employ weights in the analysis to represent the sampling proportions, weighting by the inverse probability that a dyad is included in our sample. More formally, let pij be the probability that the ith student will name the jth student in the same school as one of her or his five best same-sex friends. We model the logit of the probability that the ith student will select the jth student as a friend as a function of characteristicsof the dyad:

father'sreportcorrelate.8. Some disagreement likely resultsfrom errorin parentalreports.We concludethat studentreportsare reasonablyaccurate.

This specification follows the P* model in most respects. Main effects of sender and receiver characteristics-such as mother'seducation or respondent's gender-and the main




ij = aCaaFriendlinessai +-Pij a +Y]bPopularitybj + b

+ YgControlsijg.






effect of school characteristicsdo not appear in this model.14 In a dyadic friendshipmodel, coefficients of main effects of sender and receiver characteristics would represent the association between these variables and the total number of friends (or correlation with size of school), regardlessof the race or other characteristics of the friends. Instead of including main effects, we control for the number of friends selected-friendliness-and the number of persons who select an individual as a friend-popularity-with a series of dummy variables representing levels of friendliness and popularity (see Anderson, Wasserman,and Crouch 1999).15The log of the number of students in the school is also included as a control variable because the odds that any given dyad will be a friendship dyad decline with the size of the school. We estimate these models separately for senders of each race by Hispanic origin group-one model for non-Hispanic whites, one for non-Hispanic blacks, and so on. The race dummy variables shown in equation 1 representthe race and Hispanic origin of the receiver, with the sender's race and Hispanic origin group serving as the reference category. The coefficients of the race variables (the Ss) can then be interpretedin terms of the log odds of nominating a receiver of another race and Hispanic origin relative to nominating a receiver of the sender's own race and Hispanic origin. We control several other bases of friendship matching using variables representing differences between the members of the dyadic pair. Anderson et al. (1999) find that popularpersons are more likely to be friends with other popular persons, so we include a variable for absolute sender-receiver difference in popularity. A dummy variable indicates that the sender and receiver are in the same grade in school. Measures of relative 14We also estimated models separatelyfor each gender.We foundalmostno differencesin friendshipmatchingby gender. 15The P* model includes dummy variables

(fixed effects) for each senderandreceiver.Like Moody (2001), we control for friendliness of senderandpopularityof receiverratherthanincludingdummyvariablesfor each senderandreceiverbecausetherearetoo manyrespondentsin the sampleto includeso manydummyvariable terms.



socioeconomic status and student scholastic achievement are also included in some models, as we will discuss in the results section. CONTROLS NOT INCLUDED

The goal of our analysis is to estimate the extent of friendship segregation that is likely to result in schools with particularracial and socioeconomic status compositions. In view of this objective, we omitted two sets of variables that sometimes are included in models of friendship selection: certain network effects that may act to intensify homophily, and measures of similarity in values or activity participation. First, we did not include variables to represent transitivity and reciprocity, which are often included in P* models. Transitivityreflects the principle that friendship between two individuals is more likely if the individuals have friends in common. Transitivity intensifies racial homophily because new friends are likely to be racially similar to existing friends, who are likely to be of the same race as the sender. Because our goal is to estimate the extent of friendship segregation that is likely to result in a school with a given racial mix, it is importantnot to "control out" the intensification of racial homophily that occurs through the influence of transitivity on friendship choices. Likewise, we exclude controls for choice of the receiver by the sender-reciprocityrepresenting the principle that individuals are more likely to select a friend if the friend reciprocates the choice. Controlling for reciprocity would "control out" part of the race effect because the reciprocal choices made by receivers are partly race-guided. Second, we do not include measures of similarity of attitudes or similarity of club membership that might be a basis for homophily. In estimating the degree of racial friendship segregation that is likely to occur in a school with a particular racial and socioeconomic composition, we do not want to impose counterfactualconditions, such as no racial differences in attitudes or integration of social clubs, because these conditions will not hold in most schools in the future. Another problem with these controls is that past research suggests that similarity between friends in attitudes (Kandel 1978) and club




membership (Fink and Wild 1995) are as much a result of friendship as a cause of friendship, which results in upward bias to the estimates of the effects of these controls. STATISTICAL INFERENCE

The Add Health school survey is a schoolbased sample-students are clustered within schools. Accordingly, we calculate standard errors adjusted using the Huber/White estimator with schools as the clusters. In the matching models, the dyadic pairs can be seen as a second level of clustering within schools, with receivers clustered within the corresponding sender. Because the adjustment used to deal with school clustering does not assume independent observations within the primary clustering units (in this case, schools), this second level of clustering should not bias the standarderrors (see StataCorp2001, sect. 30.2.1). MULTIRACIAL FRIENDSHIP MATCHING We now turn to the models of friendship selection. We begin by considering the patterns of racial homophily for later-generation Asian and Hispanic students to test the contrasting predictions of traditional immigration theory and our adaptationof segmented assimilation theory for the cross-race friendship patternsof the third-plusgeneration.We then consider change across generations, and finally we addressracial composition effects. DO ASIAN AND HISPANIC STUDENTS DIFFER?

Table 1 shows the estimated coefficients of the friendship selection models. Separate models are shown for senders from each race and Hispanic origin group. The race of receiver coefficients in the first panel of the table indicate the relative likelihood a sender of the indicated group will nominate a receiver of the indicated race and Hispanic origin as a friend, relative to a receiver of the sender's own race and Hispanic origin. Not surprisingly,these coefficients are negative, indicating that students are less likely to select as a friend a student of anotherrace and/ or Hispanic origin status than a student of

their own race and Hispanic origin status. The models also include interactionvariables between immigrant generation and the race of receiver coefficients to allow for variation across generations in the strength of racial homophily. Because the interactions are included, the race of receiver coefficients can be interpretedas indicating racial homophily in friendship formation for the third-plus generation, which is the reference group in assessing generational change.16 Several measures are also included as controls. The model includes two measures of parental socioeconomic status: the absolute differences in mothers' education between sender and receiver (educational distance), and a dummy variable indicating whether both sender and receiver have mothers in professional occupations. The model also includes controls for friendship matching on GPA, popularity, friendliness, and a control for school size, as discussed in the methods section. Comparingthe coefficients for race selection effects to the other predictors, common race and Hispanic origin are among the strongest predictorsof friendship among students. The common race influence on friendship is far greater than similarity in parental socioeconomic status: A 20-year difference in mother's education is not as great a barrier to friendship as race between black and white students.17 We also found that the race-matching effects are almost unchanged in models without socioeconomic statusmatching and GPA-matching controls (not shown), indicating that race-matching does not reflect the effect of socioeconomic status or academic-based cliques.18 16

interactionswere Some second-generation droppedfor whiteHispanicsbecausesamplesize constraintsled to difficultiesin modelestimation. Thus, for some receiver groups, the reference group for white Hispanics is the second and third-plusgenerationcombined. 17We also estimatedmodelsaddingsocioeconomic status controls for father as well as mother,using only cases for which information on bothparentsis available.The matchingcoefficients for father'ssocioeconomicstatus were similarto those for the motherand do not sigeffects. nificantlyalterthe race-matching 18 Theseresultsare availablefromthe authors on request.




Table 1. Coefficients from the Logistic Regression of Friendship on Dyadic Characteristics, Including Interactions with Immigrant Generation: Add Health In-School Questionnaire, 1994 to 1995

IndependentVariable Race of Receiver White Black White Hispanic




-2.16*** (.15) [Reference]

-1.56*** (.11) -.07*

-1.57*** (.21) -.09 (.06) -1.56*** (.19) -1.84** (.21)

Black Hispanic

(.03) -1.55*** (.19)









(.04) -.20***

(.10) -.42***



(.09) (.07)

Race of Receiverby First Generation .06 White x First generation (.05) .38* Black x First generation (.19) .47** White Hispanicx First (.17) generation 1.13** Black Hispanicx First (.35) generation x First -.22 OtherHispanic (.30) generation x .21 Asian First generation (.23) Race of Receiverby Second Generation White x Second generation .06 (.04) Black x Second generation .48** (.16) White Hispanicx Second -.12 (.22) generation -.62 Black Hispanicx Second (.62) generation .20 OtherHispanicx Second (.20) generation Asian x Second generation -.15 (.15)

.26 (.16) .09

(.10) .19 (.49) .04 (.26) .80**

Race of Sender Black White Hispanic Hispanic




(.16) -.58 (.55) .27 (.18)

(.32) .39 (.24) 1.08 (.90) .29 (.54) 1.37** (.51) .22 (.92)


(.12) -.19 (.34)


(.43) .40***

(.12) -.07


.02 (.08) -.40 (.28)


-.91'** (.10) -.92*** (.15) -44*** (.11) -.82*** (.21)


(.29) 1.10** (.28)

(.15) .07 (.15) -.10 (.34) .40 (.30) .56* (.23) .80**


-.88*** (.11) -1.71*** (.37) -.71 [Reference] (.32) -1.60 -.51 [Reference] (1.07) (.28) -1.43*** [Reference] -1.91*** -.35* (.50) (.31) (.15) -1.42*** -.99*** [Reference] -.54** (.24) (.20) (.17) -.89*** -.52** -.64*** -.61*** (.13) (.14) (.18) (.16) -1.49*** -.50*** -.87*** -.35*** (.10) (.36) (.10) (.08) -.36*** (.10) -1.23*** (.17)

-2.45*** (.24) -.36** (.13) -2.08*** (.37)

Other Hispanic


[Dropped] [Dropped] .41***

(.11) -.53* (.26)

1.02*** (.31) .09 (.19) 1.47* (.59) -.52 (.61) .53 (.45) -.36 (.55)


(.13) .35** (.12) -.69 (.35) .32* (.13) -.46*



(.36) -1.34 (1.21) .17 (.45) .31 (.16)



(.10) -.39** (.13) .19 (.13) -.01 (.30) .19** (.06) -.15 (.19)

(.08) -.30 (.35) -.54 (.29) .15 (1.03) .42 (.43) .19 (.13)

(Continuedon nextpage)




(Table 1 continuedfrom previous page)

Race of Sender IndependentVariable



White Hispanic

Black Hispanic

Other Hispanic


Socioeconomic Status and Achievement

Mothers'educational distance Both mothersare professionals Students'GPA distance

-.05*** (.01) .10*** (.03) -.37*** (.02)

-.04*** (.01) .13* (.06) -.18*** (.04)

-.04*** (.01) .16* (.08) -.35*** (.03)

-.02 (.01) .16 (.12) -.22*** (.06)

-.07*** (.01) .22*** (.05) -.28*** (.03)

-.04*** (.01) -.04 (.06) -.42*** (.04)

2.61*** (.06) -.85*** (.05) -.09*** (.00)

2.24*** (.11) -.63*** (.10) -.09*** (.01)

2.31*** (.07) -.87*** (.04) -.10*** (.01)

2.24*** (.09) -.65*** (.10) -.11*** (.02)

2.16*** (.09) -.82*** (.08) -.11 ** (.01)

2.86*** (.18) -.71*** (.13) -.10*** (.01)

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes

Yes Yes








Same grade Size of school (logged) Difference in popularity Network Dummy Variables

Friendliness Popularity Numberof dyads

Notes: Standarderrorsare in parenthesesand are calculatedusing the Huber/Whiteformula,adjustedfor school clustering. Some coefficients were droppeddue to small sample sizes of sender-receiverpairs for some generations.All models were estimatedwith a constantterm, but the constantis not shown. Models were estimatedusing pseudo-maximum-likelihoodmethods and thus standardlikelihood-ratiotests are invalid. *p < .05

**p < .01

***p< .001 (two-tailed tests)

To clarify the cross-group patterns,Figure 1 shows the exponentiatedrace and Hispanic origin selection coefficients from Table 1, which are more easily interpreted than the raw coefficients. The bars on the graph indicate the odds that a within-school dyad with the indicated sender race and receiver race will be a friendship dyad, relative to the odds of selecting a friend of the same race and Hispanic origin as the sender. The graph shows the estimates for third-plusgeneration senders only. The value of 21 for "W chooses B," for instance, indicates that the odds are only about 21 percent as great that a white sender-black receiver dyad will be a friendship dyad as a white sender-white receiver dyad. By contrast,white studentshave relatively high odds of nominatingother Hispanic, Asian, and (especially) white Hispanic students as friends. Do later-generation Hispanic and Asian students have much greater odds of friendship across racial and ethnic lines than do

whites or blacks, controlling for opportunities for contact? We find some evidence that race is a lower barrierto friendship between third-plus generation Hispanic and Asian students and certain other-race groups than it is between non-Hispanic white and nonHispanic black students (this is discussed in detail below). We also find, however, that third-plus generation white Hispanics, black Hispanics, other Hispanics, and Asians show clear own-race preferences in their friendship nominations that are nearly as strong as (or in some cases stronger than) the ownrace preferences of non-Hispanic whites and non-Hispanic blacks. Figure 1 also indicates some friendshipmatching on the basis of Hispanic ethnicity. Cross-race friendships are especially likely among students who share Hispanic ethnicity, and most likely between white Hispanic and other Hispanic students. Asian students are often viewed as the nonwhite group that is closest to whites, and





W chooses B W chooses WH





m w

W chooses BH W chooses OH -E c---.l


W chooses A


___- --_


- _ .....i


B chooses W B chooses WH B chooses a) ._





B chooses 0 B chooses A

'c WH chooses W L. (0 CU

WH chooses B



---qt r

7- a. -;......

WH chooses BH


~ __







WH chooses OH WH chooses A


.v-F- r...W--;-


0 'O





BH chooses W BH chooses B -


BH chooses WH




BH chooses OH BH chooses A

a() (0 O C

OH chooses W OH chooses


-ri ~-~--_Race-

OH chooses WH OH chooses

OH chooses A

A chooses W

, "^



"--'--' .....


- ' .-- -

Categories: -




gB = Black WH = White Hispanic

1% - ,, -_ _-___:--.

A chooses B

BH = Black Hispanic


OH = Other Hispanic

A chooses WH A chooses


= White

' A chooses OH



:-I -----I


A = Asian














Odds of Friendship, Relative to an Own-GroupDyad (Own-Group= 1) Figure 1. Odds That a Cross-Group Dyad Will Be a Friendship Dyad, Relative to an Own-Group Dyad, by Race of Sender and Receiver: Third-Plus Generation, Add Health In-School Questionnaire, 1994 to 1995 Note: Figure is based on the exponentiated race selection coefficients shown in Table 1. Odds of selection are shown as a proportion of the odds that an own-group dyad in a school will be a friendship dyad (odds of own-group dyad friendship = 1).

thus we expected white-Asian friendships to be among the most common cross-race friendships in the third-plus generation. We find some evidence consistent with this idea, but this evidence is mixed. In supportof low social distance between Asian and white students, when an Asian student nominates an other-race friend, whites (including white

Hispanics) are the most preferred choice, and white students are relatively likely to nominate Asian receivers as friends. On the other hand, Asian students display a strong within-race selection preference, with low odds of selecting friends of other races. The findings of persisting own-group preference for members of most racial groups




contradictthe prediction of traditionalimmigration theory-that third-plus generation Asian and Hispanic students will assimilate to white friendship patterns (Hypothesis 1). Because we find that white Hispanic, other Hispanic, and Asian students are more likely to have cross-race friendships than are whites with blacks, we also reject strongly segmented patterns of friendship relations for these groups (Hypothesis 2a) in favor of the pattern we previously referred to as selective assimilation in friendship relations (Hypothesis 2b). We also find that the white-black divide is clearly evident in Hispanic friendship selections, consistent with Hypothesis 3. NonHispanic whites and white Hispanics are likely to nominate each other as friends, as are non-Hispanic blacks and black Hispanics.19Generally,black-nonblackis an important division in student friendships, a division also found among students of Hispanic descent. These results suggest that white Hispanics and black Hispanics are being incorporated into white and black friendship circles, respectively, with Hispanic origin as an importantsecondary influence.20 GENERATIONAL CHANGE IN FRIENDSHIP RELATIONS

Although race and Hispanic origin clearly matter for friendship patterns of the thirdplus generation, this result does not preclude the possibility that own-group preferences become weaker over immigrantgenerations. We now consider that possibility. To allow the racial homophily effects to vary by generation, the models in Table 1 include interaction terms between race of receiver and 19Thecoefficientfor whiteHispanicselection of blackHispanicsis marginallystatisticallysignificant(p = .07) becausein oursamplethereare few later-generation whiteHispanicsin the same schools as black Hispanics.In modelsthatpool immigrantgenerations,the coefficientfor white Hispanicselectionsof black Hispanicreceivers is -.83 (significantatp < .01). 20This may occurpartlybecausethe race of friendsinfluencesracialidentification.Hispanic studentswith white friendsmay be morelikely to identifyas white,whileHispanicstudentswith black friendsmay be morelikely to identifyas black.

dummy variables indicating first or second immigrant generation. The third-plus generation is the reference category. If there is weakening own-race preference across generations, we should find positive own-group and negative other-groupinteraction terms between the first and second generation and the race of receiver variables. Most of the interaction terms between generation and race of receiver in Table 1, however, are not statistically significant, indicating no statistically discernible differences between the first and second generation and the third-plus generation. Some significant coefficients are also in the direction opposite that predicted by traditional assimilation. Only one group-other Hispanics-follows a patternthat is clearly consistent with the weakening of own-race preference and increasing other-race preference predicted by traditional assimilation theory (Hypothesis 1). Indeed, much more notable than the differences across generations are the similarities. To clarify the overall patterns of racial homophily across generations, we calculated the odds that a cross-race dyad will be a friendship dyad for each immigrant generation, based on the coefficients in Table 1. We then computed correlationcoefficients of the odds that a cross-race dyad will be a friendship dyad across the three immigrantgeneration groups. The resulting correlations, using the 36 sender-receiverrace and Hispanic origin pairs as the cases (6 sender race/Hispanic origin groups by 6 receiver race/Hispanic origin groups), are shown in top panel of Table 2. The correlation coefficients for the odds of selection over generations are all above .8, indicating highly similar patterns of racial and Hispanic origin friendship homophily over the three immigrantgeneration groups. The bottom panel of Table 2 shows a similar calculation including only pairs with Hispanic or Asian senders, and provides very similar results. Correlations calculated for specific sender race and Hispanic origin groups (not shown) yield similar conclusions, with correlations above .7. The weak reductions of within-race friendship selection across generations in some instances, most clearly other Hispanics, are consistent with traditional assimilation theory. But because the dominant pat-

FRIENDSHIP Table 2. Pearson Correlation Coefficients for the Odds of Friend Selection, by Race and Hispanic Origin, between Immigrant Generations: Add Health In-School Questionnaire, 1994 to 1995 SenderGeneration




Second Third-Plus

Correlationsover Generationsfor All SenderReceiverPairs (36 pairs)

First Second Third-Plus








Correlationsover Generationfor Sender-Receiver Pairs with a Hispanic or Asian Sender(24 pairs)

First Second Third-Plus








of odds Notes:Resultsarebasedon correlations of friend selection calculatedover the pairs of sender/receiver raceandHispanicorigin.Oddsof friendselectionarebasedon calculationsfromthe modelcoefficientsin Table1. exponentiated tern is of similarity in racial homophily across generations rather than change, we find more supportfor persistentracial effects on friendship networks than for generational assimilation in friendship relations. RACIAL COMPOSITION AND FRIENDSHIP SELECTION To this point, the models have been structured so that the odds that an interracialdyad will be a friendship dyad depend only on characteristicsof the studentsinvolved, most notably on the race and Hispanic origin of the two students who form the dyadic pair. Because there are more cross-race dyads in racially diverse schools than in racially homogeneous schools, these dyadic models imply that the number of cross-race friendships increases in diverse schools, thus implicitly incorporatingpropinquity effects. Although we have used models that assume these effects to be present, we now respecify the models to test this assumption. Our respecified models also allow us to examine the possibility that friend-making behavior changes depending on the racial composition of the school population, in particular our hypothesized social support effects.



Finally, because our analysis in Table 1 found few significant differences across immigrant generations, we drop the interactions with immigrant generation from the models.21 To allow for variations in the strength of racial selection with racial composition, we introduce interactions between the race of receiver variables and dummy variables representing the racial composition of the school population.We interactedall race and Hispanic origin receiver variables with dummy variables representingracial composition in three categories: 10-30 percent, 3060 percent, and 60-100 percent own race. The reference category is 0-10 percent ownrace. In defining the own-race percent of school for Hispanic students, we use school percent white for white Hispanic senders, school percent black for black Hispanic senders, and school percent other-race Hispanics for other-race Hispanic senders. Table 3 displays selected coefficients from this new model, focusing on the strength of selection of friends from one's own racial group.22 Figure 2 graphs the estimated strength of own-group selection by racial composition of the school, from the exponentiated coefficients in Table 3, with own-group selection in a 0 to 10 percent school normalized to equal 1. The results show that the odds that an own-group dyad will be a friendship dyad are highest for students in a small racial minority in their school (less than 10 percent and/or 10-30 percent), are lower for students in 30-60 percent own-race schools, and do not change much further in 60-100 percent own-race schools. These composition effects are strong. For instance, the odds are about 10 times greater that a white-white dyad will be a friendship dyad in a 0-10 percent white school than in a 60-100 percent white school (e-238 =.09

1/10). The num-

ber of white-white dyads increases more 21 We calculatedthe CriBayesianInformation teria (Raftery 1995) for the models with and withoutthe interactions,which indicated"very strong"support(a BIC differenceof about40) for the simplifiedmodels.Thesestatisticsdo not adjustfor school-levelclustering,however,and thuswe regardthe resultsas advisory. 22 All racial coefficients in the model are shownin AppendixB.




Table 3. Selected Coefficients from the Logistic Regression of Friendship on Dyad Characteristics, Including School Racial Composition Interactions: Add Health In-School Questionnaire, 1994 to 1995




Race of Sender Black White Hispanic Hispanic

Other Hispanic


School CompositionEffects School 0-10% own-race [Reference] [Reference] [Reference] [Reference] [Reference] [Reference] x Receiver own-race .13 -.97*** .10 -.15 -.87"** School 10-30% own-race -1.46*** x Receiver own-race (.69) (.27) (.19) (.31) (.12) (.21) -.52* -1.16 -1.47*** .32 School 30-60% own-race -1.92*** [Dropped] x Receiverown-race (.68) (.25) (.13) (.25) (.15) -1.15 -2.04*** -.58 own-race -2.38*** 60-100% School [Dropped] [Dropped] x Receiver own-race (.67) (.12) (.30) (.14) Race of Receiver White Black White Hispanic Black Hispanic OtherHispanic Asian Otherrace Multiracial Numberof dyads

-.15 [Reference] -4.27*** (.55) (.43) -3.42*** [Reference] -1.62** (.35) (.17) -1.88*** -3.32*** [Reference] (.39) (.54) -.21 -.68 -3.10'** (.35) (.20) (.19) -2.75*** -2.45*** -.62*** (.18) (.20) (.16) -.50*** -2.75*** -2.53*** (.15) (.18) (.15) -2.34*** -.70*** -2.49*** (.12) (.15) (.14) -2.00*** -.54*** -2.5 1** (.11) (.11) (.14) 708,510



-.63* -3.65*** -2.97** (.30) (.26) (1.09) -1.47*** -2.33*** .36 (.29) (.32) (.70) -2.27*** -.70** -2.46*** (.69) (.24) (.61) -2.81**" [Reference] -1.32** (.61) (.44) -1.60* [Reference] -1.48*** (.26) (.80) -1.21*** [Reference] -1.84** (.27) (.65) -2.01*** -2.34*** -1.13*** (.22) (.25) (.69) -1.04** -1.99*** -1.53* (.19) (.24) (.66)




Notes: Standarderrors are in parenthesesand are calculated using Huber/Whiteformula, adjustedfor school clustering.Own race is white for whites, black for blacks, Asian for Asians, white for white Hispanics, black for black Hispanics,and other Hispanicfor other Hispanics.Models are estimatedusing pseudomaximum-likelihoodmethods and thus standardlikelihood-ratiotests are invalid. Also included but not shown: racial compositioninteractionsfor other-racegroup and all controls in Table 1. AppendixB shows coefficients for all racial compositioninteractions. **p< .01 p < .05 ***p< .001 (two-tailedtests)

quickly than this-the

number of same-race

dyads increases geometrically with the number of same-race peers-so the model results still imply that white students in a 60-100 percent white school will have more white friends than those in a 0-10 percent white school. Thus, as we consider furtherin Table 4, these results remain consistent with the basic prediction of propinquity. Because of the intensification of own-race homophily for students who are a small ra-

cial minority in their school, the increase in own-race friends present in 60-100 percent own-race schools compared with 0-10 percent own-race schools is not nearly as large as would be predictedbased purely on opportunities for contact. Students in schools with few same-race peers are especially likely to befriend the (few) other same-race peers available, consistent with Hypothesis 5. These composition effects also explain some differences in overall degree of own-




1.40 -7.. ::

School Racial Composition 0-10% own-race

l0 (a

30-60% own-race:: / 60-100% own-race

CI ?? ?? ?? ?? ??

"---- ----------....~~~~~~~~~~iil~:.\ .-- --...-......-..--------; ;-:- . -- . j ^~~~~~~~~~~~~~~~~----iiiiR'i~ ^ ll^~~~i!i~~::::






.80 -

?? ??

C ':*






?? ?? ??


::::!.. ::::

?? ?:? ?:?







~~~~~~~~~~~~~~~~~~~~~~~~:..':'~ -..-- ....--<;^ ?.~~~~~~~~~~~~~~~~~~:: ,, ;; -- - -;,?~~~~~~~~ii ? -; -------------------N??? ?~~~~~~~~~~~~~~~~~::,:: ???? i:^ : ? lJ.:"<!~ Ji ~'~:i.' J!i',, ii!~ii~ . ?,,~ii ??


- \ ....-------:' 7ii^

- -

....: : .:.: ?:~: ...................i~..... ... -----------::. . ::::.:.:N...----------:------

?;? ??






? . . .... .................-..-...... ......-.. ....-.............................. ...-..-. ... -::......-

10 -30%own-race


f 1.20co


:::. .


. '^::

'11:1 II Ill ii~~~~~~~~~~~~~~~~~~~~::

?1 ?? ??



?? ?_?


0 0




* |-

. I-













WhiteHispanic BlackHispanic OtherHispanic





Race and Hispanic Origin of Sender

Figure 2. Odds That an Own-Group Dyad Will Be a Friendship Dyad, by School Racial Composition and Race and Hispanic Origin of Sender: Add Health In-School Questionnaire, 1994 to 1995 Note: Figure is based on the exponentiatedcoefficients in Table 3. Bars indicate the odds that an owngroup dyad will be a friendshipdyad in a school of the indicated racial composition, relative to a 0-10 percentown-raceschool. a Relative to a 0-10 percentown-raceschool (0-10% own-race= 1). *p < .05 (significantlydifferentfrom 0-10% own-raceschool)

race bias. Examining the third-generation coefficients in Table 1, for instance, nonHispanic whites are more likely to choose Asian friends thanAsians are to choose nonHispanic white friends. In Table 3, we see that the disparity results in part because Asians are more often part of a small racial minority within their schools than are whites. How much of the racial segregation in friendship results from school racial segregation? Table 4 draws implications of these models for the overall share of friends of another race. The top panel of Table 4 presents the percentage distribution of friends nominated in the raw Add Health data, broken down by sender's race, as a basis for comparison. The bottom panel of Table 4 gives the predicted share of friends from each race for

senders of each race for a hypothetical school in which the racial mix of the school is equal to the overall racial distribution of students in the Add Health data (except we assume no students are in the multiracial or other race categories). The models shown in Table 3 (and in Appendix B) are used for the predictions. In this hypothetical school, all students have the same socioeconomic background and the same GPA, and all students are equally popular. In the bottom panel, Table 4 results show a clear school propinquity effect that holds for all racial groups, consistent with our basic Hypothesis 4. All groups except non-Hispanic whites and non-Hispanic blacks have a majorityof friends in other groups, reflecting the small size of these groups in this hypothetical school and the greater odds of interracial friendships for Asians and Hispan-




Table 4. Percentage Distributions of Friends, by Sender Race, in Actual and Hypothetical School Populations

School Racial Composition

Race of Receiver



Race of Sender Black White Hispanic Hispanic

Race of NamedSchool Friends in the Add Health In-SchoolSurveya 7.6 49.6 White 90.0 2.5 81.5 1.9 Black -

Other Hispanic


8.3 72.3





White Hispanic







Black Hispanic



















Other Hispanic


EstimatedShare of Friends of Each Race in a HypotheticalSchool with CompositionReflectingthe Add Health StudentRace Distribution 40.2 41.3 15.7 66.9 White 79.7 15.2 59% 5.7 11.1 51.4 5.8 3.2 59.5 Black 18% 4% 2% 9% 8%

White Hispanic Black Hispanic OtherHispanic Asian


4.5 .4






















Notes: Estimatesin the bottom panel are based on models in Table 3 and AppendixB. Estimatesin the bottompanel assume that both studentsin a dyad have the same GPA and motherswith the same socioeconomic background,and that the studentsare equally popular. aExcludes multiracialand other-racecategories.

ics than for black students.At the same time, all groups select members of their own group in substantially greater share than their proportionate presence in the school, reflecting the continuing influence of racial homophily in friendship selection. OTHER CONSIDERATIONS: TRACKING, BUSING, AND NETWORK SIZE We also considered several other influences on social networks that might be important in understandingthe social relations of students in increasingly diverse schools. We briefly discuss the main results of these additional analyses, none of which substantially altered the strength of racial homophily or altered our main findings. (Tables showing these results are available from the authors on request.) First, we thought it possible that much friendship segregation might result from the segregation of academic tracks. The Add Health survey does not provide the informa-

tion necessary to identify the track of individual students, but it is possible to identify from the school administratorquestionnaire whether the school includes separate instructional tracks. We estimated our basic models of friend selection including interactions allowing for variation in the strength of homophily between schools with and without tracking. Consistent with results reported by Hallinan and Williams (1989) and Kubitschek and Hallinan (1998), we found slightly stronger racial homophily in schools with tracking. The difference, however, was small and accounted for little of the overall racial homophily in friendship. Second, we considered that racial integration in the schools in our sample might be achieved as a part of race-conscious policies of student school assignment, often involving busing. If these policies have negative influences on interracialrelations in schools, the influence of busing might be confounded with racial composition effects. Our analysis, however, found no statistically signifi-


cant negative effects of busing on the formation of interracialfriendships.23 Third, we considered whether students in small racial minorities in a school might tend to have fewer friends overall. Changes in total network size are not captured in our matching models, but could contribute to fewer cross-race friendships for students whose racial group is the numerical minority in their school. We estimated models of the number of friends named on the Add Health survey as a function of racial composition to investigate this possibility. We found a statistically significant reduction in the size of social networks for white, white Hispanic, and Asian students who are members of small minorities (less than 10 percent) in their school population. These changes were substantively small, however, and not sufficient to alter our conclusions. CONCLUSIONS Our results provide only weak support for the prediction of traditional assimilation theory, as applied to race and Hispanic background, that across generations friendship networks tend to converge toward a single American "core" standard.We do find generational changes for some racial and Hispanic-origin groups in the direction predicted by traditionalassimilation theory, but the main friendship pattern is of considerable similarity in racial and Hispanic-origin homophily across immigrant generations. Race and Hispanic backgroundare powerful factors influencing adolescent friendship, with racial similarity a substantially stronger predictorof friendship formation than parental similarity in socioeconomic status. Some have suggested that Asian students are on a path of assimilation toward white patterns, a view Tuan (1998) calls becoming "honorarywhites."Asian students do tend to have many white friends, but this occurs primarily because of propinquity effects, as 23 This conclusion differs from Moody's (2001).Thedifferencein resultscouldbe because Moodyexaminestotalracialsegregation,without separatelyconsideringraces involved. Schools with busingin the Add Healthdatahave higher sharesof blacksandwhitesthanschoolswithout busing,whichmayresultin higherapparentoverall friendshipsegregationin the school.



Asian students are usually a small minority in their schools. Although there is some evidence that white students are relatively likely to select Asian friends, we find that Asian students are considerably more likely to nominate an Asian student than a white student as a friend, relative to the number of Asian and white schoolmates. We conclude from these results that Asian students appear to be following a path of group solidarity in their primary group relations, displaying some affinity for white friends, but without assimilation to white or black patterns. This is consistent with the suggestion of Portes and Rumbaut (1996) that Asian immigrants may be following a path of "selective acculturation"in which partial assimilation is accompanied by the simultaneous maintenance of ethnic ties. Hispanic students are in the unique position in our study of having their friendship selections divided by crosscutting racial and ethnic (Hispanic origin) lines. We find that Hispanic origin is an important basis for friendship matching, most notably between white Hispanic students and Hispanic students who indicate their race as "other."Yet we also find that race remains highly salient to the friendship choices of Hispanic students-the friends of white Hispanics are mostly whites, white Hispanics, and other Hispanics; the friends of black Hispanics are mostly blacks and black Hispanics. These results suggest that, while ethnic solidarity exists, white and black Hispanics are assimilating in different ways, with racially white Hispanics joining white peer groups and racially black Hispanics joining black peer groups. We take these results as contributingto the evidence that racial difference is important in understandingprocesses of immigrantadaptation and should be included in theories of immigration adaptation.The influence of black and white race on friendship patterns among Hispanics provides an especially clear example of how the racial categorization of an immigrant group may have a lasting influence on the incorporation of the group members into the host society. Although we view the stability of racial homophily across generations as mostly inconsistent with traditional assimilation theory as applied to racial groups, we note




that traditional assimilation theory may still describe other dimensions of immigrant adaptation.An analysis of generational change in friendship homophily based on nationalorigin ethnicity, for instance, may well be more consistent with traditionalassimilation theory. That is, the persistent racial and Hispanic-origin homophily we find here could representhomophily on the basis of nationalorigin groups for early generations and racial or panethnic selection in later generations. Consistent with past research, we find clear evidence of propinquityeffects for students of all races: The share of cross-race friends depends strongly on the share of potential friends in the school context who are other-race. We also find, however, that students in schools in which they are members of a small racial minority substantially increase their odds of own-race friend selection. We believe this to be because students desire several friends of their own racial group for reasons of social support and they alter their friend-makingbehavior to achieve this goal when there are only a few samerace friends in their school. Future research incorporating direct measures of the attitudes, beliefs, or feelings that are relevant to racial differences in friendship selection would be useful to provide more definitive evidence regarding this interpretationof the "intensification"effect. We note three final caveats. First, our results reflect the organizationaland racial environments present in our sample of junior and senior high schools, and racial relations in American society more generally. If school officials were to manage integrated environments differently, or racial relations in American society were to significantly change, our conclusions would need to be correspondingly adjusted. Second, we have said little about multiracial students because the complexity of multiracial identity requires a separate analysis. The inclusion of multiracial students would somewhat increase the extent of cross-race friendship in our results. Finally, adolescents are at a stage in the life cycle in which racial identity and racial awareness crystallize. A study of adults in social contexts with racial compositions similar to our sample of schools might well find lower levels of racial homophily in friend selection.

DISCUSSION If currenttrends continue, Hispanic students will soon become the largest minority group in schools, and Asians will make up 10 percent of all students by mid-century (Orfield and Yun 1999). Will the increasing prevalence of these new immigrant minorities break down high levels of social segregation among students of different races? How will the growing presence of these new immigrant minorities alter the structureof crossrace networks among students? We find evidence consistent with reduction in the extent of racial segregation in friendship, but we also find continuity in several respects. Our results indicate that interracial and interethnic friendships in schools will increase as the share of students who are of Asian and Hispanic origins increases, because of propinquityeffects and because race is less of a barrierto friendship among some combinations of non-Hispanic white, white Hispanic, other Hispanic, and Asian students than between white students and black students. The clear own-race preference of all groups in their friendship nominations also suggests, however, that own-race friends will continue to be overrepresentedin the social networks of most students relative to their school's racial composition. We also find that black-white divisions will continue to be importantin the increasingly Hispanic schools of the future, because white Hispanic and black Hispanic students are joining, respectively, white and black peer groups. Disturbingly,we find especially high levels of segregation of blacks, including black Hispanics, from all other racial groups. Our results suggest that black-nonblack is an importantdividing line in multiracial schools, a division that will likely become more evident as the share of school populations who are neither white nor black increases. The increased own-race favoritism in friendship among students who are members of small racial minorities suggests that dispersing minority students in very small concentrations would not necessarily maximize the extent of cross-race ties in a school district (see Feld and Carter 1998). Instead, maintaining more balanced racial proportions in some schools might actually lead to


more reciprocatedcross-race friendships because it will not result in the intensification of homophily that we find among students in especially small minorities. More detailed conclusions sorting out own-race intensification and propinquity effects will require a detailed formal model that is beyond the scope of this paper.Our tentative conclusion is that a low degree of clustering of a racial minority in schools may be desirable to increase reciprocatedcross-race friendships in situations in which the minority represents a small share of the overall population of a school district. Although race will remain an important influence on student friendship choices in multiracial schools, our results do not mean that the current levels of segregation in friendshipare inevitable. On the contrary,the school propinquity effects indicate that increasing school racial diversity is likely to increase friendships across racial lines, especially when combined with the greaterlikelihood of cross-race friendship for Asian and Hispanic students and the possible weakening of "intensification"effects as numbersof Asian and Hispanic students grow beyond a token few in more schools. Our results suggest that a substantial increase in racial diversity in schools will lead to a notable increase in cross-race friendships, relative to currentlevels. Yet this potential will only be realized if the increasing diversity of student populations translates into increasing racial diversity within schools. We believe that demographic change makes at least small increases in within-school racial diversity very APPENDIX



likely in the future, but we are uncertain of how closely within-school diversity will mirrorincreasing diversity in the population at large. One recent study finds that a large share of Hispanic students in the Southwest are educated in almost entirely Hispanic schools, raising the possibility that growing diversity in the school-aged population might be accompanied by increased schoollevel segregation (Frankenberg, Lee, and Orfield 2003). All else equal, the increasing diversity of American society should translate into greater within-school diversitybut we cannot assume that all else will necessarily be equal. Continued attention to racial proportions in schools remains important if schools are to fulfill their promise as institutions in which Americans learn to manage racial diversity and interact across lines of racial and cultural difference. Lincoln Quillian is Associate Professor of Sociology at the University of Wisconsin-Madison. His research focuses on inequality, racial attitudes, and urban population distribution. His current research projects investigate the causes and consequences of urban spatial segregation on the basis of income and the friendship networks of multiracial adolescents. Mary E. Campbell is a Ph.D. candidate in Sociology at the University of Wisconsin-Madison. Her research focuses on racial identity and racial stratification in education and employment. She is currently completing her dissertation, which investigates the patterns of racial identification for multiracial Americans who are forced to choose a single race and explores stratification outcomes for multiracial adolescents and adults.


Mean Number of Same-Sex Friends and Student/School Counts, by Race and School Racial Composition: Add Health In-School Questionnaire, 1994 to 1995 Mean Number of: Race of Sender/ School Racial Composition White Respondents (N = 38,512) School is <10% white

Own-Race Friends in School

Other-Race Friends in School

Number of Students

(Number of Schools)a





School is 10-30% white





School is 30-60% white





School is >60% white





(Continued on next page)





(AppendixA continued from previous page) Mean Number of: Number of Students

(Number of Schools)a

Own-Race Friends in School

Other-Race Friends in School



School is 10-30% black



School is 30-60% black



School is >60% black











Race of Sender/ School Racial Composition Black Respondents (N = 11,243) School is <10% black

White Hispanic Respondents (N = 2,606) School is <10% white School is 10-30% white









School is 30-60% white





School is >60% white












(29) (14)

Black Hispanic Respondents (N = 1,027) School is <10% black School is 10-30% black




















Asian Respondents (N = 3,479) School is <10% Asian





School is 10-30% Asian













School is 30-60% black School is >60% black Other Hispanic Respondents (N = 8,428) School is <10% Other Hispanic School is 10-30% Other Hispanic School is 30-60% Other Hispanic

Other respondents (N = 1,564) Multiracial respondents (N = 6,113)

Notes: Own-race friends are white for whites, black for blacks, white or white Hispanic for white Hispanics, black or black Hispanic for black Hispanics, other Hispanic for other Hispanics, Asian for Asians, other race for persons of other race, and multiracial for multiracial persons. aIncludes all schools with at least one student of the indicated race/Hispanic origin group.

APPENDIX B Coefficients from the Regression of All Race and School Racial Composition Interactions in Models with Race Composition Effects: Add Health In-School Questionnaire, 1994 to 1995 Independent Variable Race of Receiver White Black White Hispanic



White Hispanic

Black Hispanic

Other Hispanic



-4.27*** (.43)

-.15 (.55)

-2.97** (1.09)

-.63* (.26)

-3.65*** (.30)

-3.42*** (.17) -1.88*** (.54)


.36 (.70) -2.27*** (.61)

-1.47*** (.32) -.70** (.24)

-2.33*** (.29) -2.46*** (.69)

-3.32*** (.39)

-1.62*** (.35) [Reference]

(Continued on next page)




(Appendix B continued from previous page) White


White Hispanic

Black Hispanic

Other Hispanic


-3.10*** (.19)

-.21 (.20)

-.68 (.35)


-1.32** (.44)

-2.81*** (.61)

Other Hispanic

-2.75*** (.16)

-2.45*** (.20)

-.62*** (.18)

-1.60* (.80)


-1.48*** (.26)


-2.53*** (.15)

-2.75*** (.18)

-.50*** (.15)

-1.84** (.65)

-1.21*** (.27)


Other race

-2.49*** (.15)

-2.34*** (.14)

-.70*** (.12)

-2.34*** (.69)

-1.13*** (.25)

-2.01 *** (.22)


-2.51*** (.14)

-2.00*** (.11)

-.54*** (.11)

-1.53* (.66)

-1.04*** (.24)

-1.99*** (.19)



10-30% white x White receiver

.83 (.44)

-.07 (.59)

-.13 (.86)

-.81** (.18)

1.26*** (.33)

30-60% white x White receiver

.51 (.44)

-.27 (.58)

-.24 (.82)

-1.10*** (.25)

1.23*** (.27)

60-100% white x White receiver

.61 (.44)

-.47 (.58)

-.07 (.80)

-.93*** (.13)

1.45*** (.25)

Independent Variable Race of Receiver (Continued) Black Hispanic

School Racial Composition x 0-10% white x White receiver

[Reference] [Reference][Reference]


0-10% black x Black receiver



10-30% black x Black receiver

-.66** (.20)

-.10 (.42)

-.88*** (.17)

-.35 (.30)

-.84* (.35)

30-60% black x Black receiver

-.65** (.24)

.23 (.42)

-1.45*** (.19)

-.53 (.33)

-1.17** (.36)

60-100% black x Black receiver

-.76* (.38)

.59 (.57)

-2.04*** (.16)

.55 (.30)

.23 (.25)




0-10% white x White Hispanic receiver

[Reference] [Refer[Reference]

[Reference] [Reference] [Refer[Reference]


10-30% white x White Hispanic receiver

-.05 (.50)

.69 (.45)

-.09 (.77)

-.13 (.16)

.12 (.70)

30-60% white x White Hispanic receiver

-.29 (.50)

.10 (.42)

-.29 (.86)

.00 (.16)

-.24 (.70)

60-100% white x White Hispanic receiver

-.59 (.49)

.22 (.43)

-.41 (.81)

-.46* (.18)

.33 (.71)

0-10% black x Black Hispanic receiver


[R [Reference]


[Reference] [Refer[Reference]

10-30% black x Black Hispanic receiver

-1.61*** (.34)

-.78** (.26)

-.78 (.62)

-.13 (.39)

-.65 (.97)

30-60% black x Black Hispanic receiver

-1.38*** (.41)

-1.26*** (.24)

-.48 (.82)

-.57 (.43)

-.73 (1.25)

60-100% black x Black Hispanic receiver

-1.71** (.61)

-1.94*** (.24)

1.33 (.71)

-.30 (.77)

1.54 (1.15)

(Continued on next page)




(Appendix B continued from previous page) Independent Variable



School Racial Composition x Race (Continued) 0-10% other Hispanic x [Reference] [Reference] Other Hispanic receiver

White Hispanic

Black Hispanic



Other Hispanic




10-30% other Hispanic x Other Hispanic receiver

-.24 (.15)

-.31 (.27)

.47 (.31)

-.06 (.53)


-1.26*** (.26)

30-60% other Hispanic x Other Hispanic receiver

-.44* (.22)

-1.14*** (.21)

.40* (.19)

-.89 (.51)


-1.56*** (.34)






-.34* (.14)

-.81*** (.24)

-.68** (.24)

-1.34*** (.36)

-.79** (.30)

0-10% Asian x Asian receiver 10-30% Asian x Asian receiver Number of dyads








7:* (i19)


Notes: Standard errors are adjusted for school clustering. Models are estimated using pseudo-maximum-likelihood methods and thus standard likelihood-ratio tests are invalid. Also included but not shown: all variables in Table 3. Also see notes to Table 3. iShaded cells indicate composition effects on own-race friend selection. Race effects and diagonal (ownrace) effects are also shown in Table 3. **<.01 *p <.05 ***p< .001 (two-tailed tests)

REFERENCES Allport, Gordon. 1954. The Nature of Prejudice. Cambridge, MA: Addison-Wesley. Anderson, Carolyn J., Stanley Wasserman, and Bradley Crouch. 1999. "A P* Primer: Logit Models for Social Networks." Social Networks 21:37-66. Bankston, Carl L., III and Min Zhou. 1995. "Effects of Minority-Language Literacy on the Academic Achievement of Vietnamese Youths in New Orleans." Sociology of Education 68:1-17. Bearman, Peter S., Jo Jones, and J. Richard Udry. 1997. "The National Longitudinal Study of Adolescent Health: Research Design." Retrieved July 29, 2003 ( addhealth). Blau, Peter. 1977. Inequality and Heterogeneity: A Primitive Theory of Social Structure. New York: Free Press. Chan, Kenyon and Shirley Hune. 1995. "Racialization and Panethnicity: From Asians in America to Asian Americans." Pp. 205-56 in Toward a Common Destiny: Improving Race and Ethnic Relations in America, edited by W. Hawley and A. Jackson. San Francisco, CA: Jossey-Bass. Clotfelter, Charles. 2002. "InterracialContact in High School Extracurricular Activities." Urban Review 34:25-46. Dawkins, Marvin P. and Jomills Henry Braddock

II. 1994. "The Continuing Significance of Desegregation: School Racial Composition and African American Inclusion in American Society." Journal of Negro Education 63:394405. Denton, Nancy and Douglas Massey. 1989. "Racial Identity among Caribbean Hispanics: The Effect of Double Minority Status on Residential Segregation." American Sociological Review 54:790-808. Feld, Scott L. 1982. "Social StructuralDeterminants of Similarity among Associates." American Sociological Review 47:797-801. Feld, Scott L., and William C. Carter. 1998. "When Desegregation Reduces Interracial Contact: A Class Size Paradox for Weak Ties." American Journal of Sociology 103:1165-86. Festinger, Leon, Stanley Schachter, and Kurt Back. 1950. Social Pressures in Informal Groups: A Study of Human Factors in Housing. Stanford, CA: Stanford University Press.

Fink, Benedykt and Klaus-Peter Wild. 1995. "Similarities in Leisure Interests: Effects of Selection and Socialization in Friendships." Journal of Personality and Social Psychology 135:471-82.

Frankenberg, Erica, Chungmei Lee, and Gary Orfield. 2003. "A Multiracial Society with Segregated Schools: Are We Losing the Dream?" A report of the Civil Rights Project

FRIENDSHIP of Harvard University. Harvard University, Cambridge, MA. Retrieved July 29, 2003 ( research/resegO3/reseg-full.php). Gerard,Harold and Norman Miller. 1975. School Desegregation: A Long-Term Study. New York: Plenum. Glazer, Nathan. 1993. "Is Assimilation Dead?" Annals of the American Academy of Political and Social Science 530:122-36. Glazer, Nathan and Daniel Moynihan. 1963. Beyond the Melting Pot. Cambridge, MA: MIT Press. Gordon, Milton. 1964. Assimilation in American Life: The Role of Race, Religion and National Origins. New York: Oxford University Press. Hacker, Andrew. 1993. Two Nations: Black and White, Separate, Hostile, Unequal. New York: Ballantine Books. Hallinan, Maureen. 1974. The Structure of Positive Sentiment. Amsterdam, The Netherlands: Elsevier Scientific. . 1982. "Classroom Racial Composition and Children's Friendships." Social Forces 61:56-72. Hallinan, Maureen T. and Stevens S. Smith. 1985. "The Effects of Classroom Racial Composition on Students' InterracialFriendliness." Social Psychology Quarterly 48:3-16. Hallinan, Maureen T., and Richard A. Williams. 1989. "InterracialFriendship Choices in Secondary Schools." American Sociological Review 54:67-78. Hirschman, Charles and Matthew Snipp. 1999. "The State of the American Dream: Race and Ethnic Socioeconomic Inequality in the United States, 1970-1990." Pp. 89-107 in A Nation Divided: Diversity, Inequality, and Community in American Society. New York: Columbia University Press. Johnson, James H., Jr. Walter C. Farrell Jr., and Chandra Guinn. 1999. "Immigration Reform and the Browning of America: Tensions, Conflicts and Community Instability in Metropolitan Los Angeles." Pp. 390-411 in The Handbook of International Migration. New York: Russell Sage Foundation. Joyner, Kara and Grace Kao. 2000. "School Racial Composition and Adolescent Racial Homophily." Social Science Quarterly 81: 810-25. Kandel, Denise. 1978. "Homophily, Selection, and Socialization in Adolescent Friendships." American Journal of Sociology 84:427-36. Kubitschek, Warren and Maureen Hallinan. 1998. "Tracking and Students' Friendships." Social Psychology Quarterly 61:1-15. Lieberson, Stanley. 1961. "A Societal Theory of Race and Ethnic Relations." American Sociological Review 26:902-10.



Little, Roderick. 1992. "Regression with Missing X's: A Review." Journal of the American Statistical Association 87 (420): 1227-37. Massey, Douglas S. and Nancy Denton. 1987. "Trends in the Racial Segregation of Blacks, Hispanics, and Asians: 1970-1980." American Sociological Review 52:802-25. McPherson, Miller, Lynn Smith-Lovin, and James M. Cook. 2001. "Birds of a Feather: Homophily and Social Networks." Annual Review of Sociology 27:415-44. Moody, James. 2001. "Race, School Integration, and Friendship Segregation in America." American Journal of Sociology 107:679-716. Murguia, Edward and Edward E. Telles. 1996. "Phenotype and Schooling among Mexican Americans." Sociology of Education 69:27689. Orfield, Gary and John T. Yun. 1999. "Resegregation in American Schools." A report of the Civil Rights Project of Harvard University. Harvard University, Cambridge, MA. Retrieved July 29, 2003 (http://www. resegO3/reseg-full.php). Patchen, Martin. 1982. Black-White Contact in Schools: Its Social and Academic Effects. West Lafayette, IN: Purdue University Press. Portes, Alejandro and Ruben C. Rumbaut. 1996. ImmigrantAmerica: A Portrait. Berkeley, CA: University of California Press. . 2001. Legacies: The Story of the Immigrant Second Generation. Berkeley, CA: University of California Press. Portes, Alejandro and Min Zhou. 1993. "The New Second Generation: Segmented Assimilation and Its Variants among Post-1965 ImmigrantYouth."Annals of the American Academy of Political and Social Science 530:74-96. Powers, Daniel A., and Christopher G. Ellison. 1995. "Interracial Contact and Black Racial Attitudes: The Contact Hypothesis and Selectivity Bias." Social Forces 74:205-26. Qian, Zhenchao. 1997. "Breakingthe Racial Barriers: Variation in Interracial Marriage Between 1980 and 1990." Demography 34:26376. Raftery, Adrian. 1995. "Bayesian Model Selection in Social Research."Sociological Methodology 25:111-63. Rumbaut, Ruben G. 1999. "Assimilation and Its Discontents: Ironies and Paradoxes."Pp. 17295 in The Handbook of International Migration. New York: Russell Sage Foundation. St. John, Nancy H. and Ralph G. Lewis. 1975. "Race and the Social Structureof the Elementary Classroom." Sociology of Education 48: 346-68. Schofield, Janet and H. A. Sagar. 1977. "Peer Interaction Patterns in an Integrated Middle




School." Sociometry 490:130-38. Schofield, Janet Ward and Bernard E. Whitley. 1983. "PeerNomination vs. Rating Scale Measurement in Children's Peer Preferences." Social Psychology Quarterly 46:242-51. Sigelman, Lee and Susan Welch. 1993. "The Contact Hypothesis Revisited: Black-White Interaction and Positive Racial Attitudes." Social Forces 71:781-95. Sowell, Thomas. 1981. Ethnic America. New York: Basic Books. StataCorp. 2001. Stata Statistical Software 7.0 User's Guide. College Station, TX: Stata Corporation. Tatum, Beverly Daniel. 1987. Assimilation Blues: Black Families in a White Community. New York: Basic Books. . 1999. WhyAre All the Black Kids Sitting Together in the Cafeteria? New York: Basic Books. Tuan, Mia. 1998. Forever Foreigners or Honorary Whites? The Asian Ethnic Experience To-

day. New Brunswick, NJ: Rutgers University Press. Warner, Lloyd and Leo Srole. 1945. The Social Systems of American Ethnic Groups. New Haven, CT: Yale University Press. Wasserman,Stanley, and Philippa Pattison. 1996. "Logit Models and Logistic Regression for Social Networks: I. An Introduction to Markov Graphs and P*." Psychometrica 61:401-25.

Wells, Amy Stuartand Robert Crain. 1994. "Perpetuation Theory and the Long-Term Effects of School Desegregation." Review of Educational Research 64:531-53. Winship, Christopher, and Larry Radbill. 1994. "Sampling Weights and Regression Analysis." Sociological Methods and Research 23:23057. Zhou, Min and Carl L. Bankston III. 1994. "Social Capital and the Adaptation of the Second Generation: The Case of Vietnamese Youth in New Orleans."InternationalMigration Review 28:821-45.

friendship segregation