beyond black and white: the present and future of multiracial ... - kenneth … · 2019. 11....

BEYOND BLACK AND WHITE:

THE PRESENT AND FUTURE OF

MULTIRACIAL FRIENDSHIP SEGREGATION

LINCOLN QUILLIANUniversity of Wisconsin-Madison

MARY E. CAMPBELLUniversity of Wisconsin-Madison

How will racial divisions in student friendship networks change as U.S. schoolsincorporate a growing Asian and Hispanic population? Drawing on theories of racein assimilation processes and the effects of relative group size on intergroup rela-tions, several hypotheses are developed to address this question. These hypothesesare tested using data on friendships among students in grades 7 to 12 from the Na-tional Longitudinal Study of Adolescent Health. Key findings are that (1) cross-racefriendships including Asian and Hispanic students are more common than thosebetween white and black students, but race and Hispanic background have signifi-cant influences on student friendships that persist over immigrant generations; (2)black or white racial identifications are strongly associated with the friendshipchoices of Hispanic students: {3) cross-race friendships increase with school racialdiversity: and (4) own-group friend selection intensifies for students in small racialminorities in a school. The results support theories of racially segmented patterns ofassimilation in primary group relations and suggest that .students in small racialminorities seek to tnaintain a friendship network including several own-race friends.Implications are discussed.

ALTHOUGH statements to the court inBrown v. Board of Education proposed

desegregation as a policy to increase theachievement and self-esteem of hlack chil-dren, contemporary discussions focus moreon fostering improved racial relations andreducing minority exclusion as major ration-ales for desegregation (Wells and Crain1994). Central to these goals is the proposi-tion that racially desegregated schooling willtend to generate cooperative, equal-statuscontact across racial lines and increase in-

Direct correspondence to Lincoln Quillian.Department of Sociology. University of Wiscon-.sin. 1180 Observatory Drive. Madison, WI 53706(quillian(s'ssc.wisc.edu). This research was sup-ported by the Institute for Research on Poverty,University of Wisconsin-Madison. An early ver-sion of this paper was presented at the August2001 meeting of the American Sociological As-sociation in Anaheim, California. We benefitedfrom the comments of Stephen Plank. TedMouw. and the ASR reviewers. This research usesdata from the National Longitudinal Study of

terracial friendship. Supporters of desegre-gation programs argue that the increase incross-race interaction and friendships result-ing from desegregation will improve confi-dence in interracial interaction, reduce ste-reotypes, and in the long run, contribute to amore integrated and equal society. The abil-ity to manage social relationships across ra-cial lines may be especially important for thefuture socioeconomic status of many minor-ity students because their entrance into manyhigh-status occupations will require regularinteractions with white teachers, coworkers,and bosses (Dawkins and Braddock 1994).

Consistent with this line of argument, pastsocial science research has generally sup-ported Allport's (1954) "contact hypoth-

Adolescent Health (Add Health), a program de-signed by J. Richard Udry and Peter Bearmanand funded by grant P01-HD3192I from the Na-tional Institute of Child Health and Human De-velopment to the Carolina Population Center.University of North Carolina at Chapel Hill, withcooperative funding from 17 other agencies.

540 At^ERtCAN SOCIOLOGICAL REVIEW, 2003. VOL. 68 (AUGUST:54O-566)

FRIENDSHIP SEGREGATION 541

esis:" When cooperative, equal-status con-tact across racial lines is achieved, interra-cial attitudes tend to improve, and comfortin interracial encounters tends to increase(Powers and Ellison 1995; Sigelman andWelch 1993). This argument also rests, how-ever, on the debatable claim that racially di-verse schooling will substantially increasethe extent of positive cross-race contact andfriendship. Past studies of black-whiteschools found that the experience of manystudents is one of racial separation ratherthan integration because the .social activitiesand friendships in many numerically inte-grated schools are highly segregated by race(Clotfelter 2002; Hallinan and Williams1989).

Amidst the current retreat from active ef-forts to desegregate schools, the increasingracial diversity of American society is alter-ing the terms of the debate about racial seg-regation in schooling. Since I960, the pro-portion of Hispanics in the nation's publicschool population has more than tripled, andan "almost invisible" minority of Asian stu-dents has grown to 4 percent (Orfield andYun 1999). The result has been a gradual de-cline in the percentage of whites in the aver-age white student's school and the emer-gence of increasing numbers of schools withmultiracial populations.

Because most prior research has focusedon black-white desegregation, little is knownabout racial relations in these emerging mul-tiracial schools. Will race be as important abarrier to friendship in these increasinglyHispanic and Asian schools as it has been inblack-white schools? How will friendshiprelations shift as the share of Asians and His-panics in schools increases? These questionsare important both for evaluating school ra-cial diversity as a goal of educational policyand for understanding racial relations asAmerican society becomes more diverse.

After reviewing past studies of cross-racestudent friendship formation, we develop aset of hypotheses to address these questionsbased on two lines of sociological theory.First, we consider theories of adaptationamong Asian and Hispanic' immigrants, ask-

' In our discussion we use racial and ethnicterms that are consistent with those used in theNational Longitudinal Study of Adolescent

ing whether members of these "new" immi-grant groups will follow a pattern of assimi-lation in which racial divisions fade in im-portance over time, or whether a model offriendship assimilation that accounts for ra-cial difference is required. Second, we drawon theories addressing how the relative sizesof racial groups in a setting influence thestructure and content of cross-group rela-tions. We then test our expectations usingsociometric data on the friendships of morethan 70,000 students in more than 130schools from the National LongitudinalStudy of Adolescent Health (Add Health).

PAST RESEARCH AND THEORY

STUDIES OF FRIENDSHIP INAMERICAN SCHOOLS

Theories of friendship have identified manybases of interpersonal attraction, includinghomophily, propinquity, status, and reciproc-ity (Hallinan and Williams 1989; McPherson,Smith-Lovin, and Cook 2001). Particularlyrelevant for understanding friendship in ra-cially diverse schools are homophily—thetendency to form friendships with similarothers—and propinquity—the tendency toform friendships with others who share thesame social situation. Because the settingsfrom which friends are chosen are usuallyrelatively homogeneous, homophily and pro-pinquity usually combine to produce highlevels of racial homogeneity in social net-works (Feld 1982). In racially integrated set-tings, on the other hand, propinquity andhomophily work to opposite purposes:Homophily in selecting friends promotescontinued racial segregation of friendshiprelations, while propinquity promotes in-creased cross-race friendships.

Studies of friendship relations among stu-dents in black-white schools reveal that bothpropinquity and homophily are important.Supporting propinquity, mo.st studies find anincrease in black-white friendships as theshares of students who are black and whitein schools or classrooms move toward par-ity (Hallinan 1982; Hallinan and Smith1985; Patchen 1982; Schofield and Sagar

Health (Add Health), which is the basis of ouranalysis.

542 AMERICAN SOCIOLOGICAL REVIEW

1977; but see Gerard and Miller 1975). Sup-porting homophily, even in numerically in-tegrated black-white schools interracialfriendships remain far less frequent thansame-race friendships. HalHnan and Will-iams (1989), in a comprehensive analysis ofracial homophily among secondary schoolstudents, find that students are only one-sixth as likely to realize a possible friend-ship with a schoolmate of a different race aswith a schoolmate of the same race.

Because court-ordered school desegrega-tion focused on black-white segregation,most analyses of interracial friendship pat-terns in schools in the i97Os and 1980s ex-amined schools in which most students wereblack or white. We know of oniy two large-scale sociometric studies that have consid-ered multiracial student populations. Joynerand Kao (2000) consider the relationship be-tween the same-race share of the schoolpopulation and the probability of having aninterracial friend for members of several ra-cial groups. Moody (2001) examines overalllevels of friendsbip segregation, dividing re-spondents into same-race and other-race cat-egories and using a single index for schoolracial diversity. Consistent with the black-white studies, these two studies find increas-ing rates of cross-race friendship with in-creased racial diversity, although the in-crease is not linear.

While these studies take the important stepof incorporating students who are not blackor white, neither study reports results forpairings of particular racial groups (e.g.,white-black, white-Asian, black-Asian, etc.).Because both studies employ the racial cat-egories of same-race and other-race, theirresults cannot address our central questionof changes in student relations as the sharesof Asian and Hispanic students increase.Theory and empirical studies suggest thatsegregation may vary for different combina-tions of racial groups, and in particular thatAsian and Hispanic students may have pat-terns of friendship segregation that differfrom white-black patterns.

ASSIMILATION THEORIES FOR

FRIENDSHIP FORMATION

Relations between black students and whitestudents face the difficulties of black-wbite

relations in America generally: high levelsof black-white socioeconomic inequality, ahistory of racial oppression, and the persis-tence of deeply internalized stereotypes andracial resentments. Given the persisting sig-nificance of the biack-white division in otherareas of life, it is not surprising that thesedivisions remain evident in the social rela-tions among students. Schools, of course, arenot isolated from racial relations in the widersociety.

By contrast, the level of social segregationfor Hispanic students and Asian studentsfrom white and black students is less clear,reflecting the ambiguous location of Hispan-ics and Asians within the traditionally dualAmerican racial system. Tbe history of His-panics and Asians as voluntary immigrantminorities suggests that their friendship re-lations with whites may follow a substan-tially different pattern than that betweenwhites and blacks (Lieberson 1961). Be-cause many Asian Americans and HispanicAmericans are recent immigrants, the domi-nant frameworks for understanding their so-cial positions have been derived from theo-ries of immigrant adaptation, which has ledto two schools of thought on the likely fu-ture position of these groups in Americansociety.^

One view is that Asian Americans andHispanic Americans are on a path leading toassimilation with white patterns. In thisview, Hispanic immigrants and Asian immi-grants are following the predictions of tradi-tional assimilation theory, in which immi-grant groups progress from having the char-acteristics of immigrants on arrival in theUnited States toward the characteristics ofthe "core group" in the host society (Gordon1964; Warner and Srole 1945).^ Traditional

- We do not investigate variations in national-origin among Hispanic and Asian immigrants be-cause including national origin in our analysiswould demand another research project. Evi-dence suggests that the panethnic categories ofHispanic American and Asian American arc ofgreater significance than is national origin formany later generation Asian and Hispanic Ameri-cans (Tuan 1998).

^ Few empirical studies of assimilation exam-ine friendship, probably because data are limited.Friendship, however, has been included in as-similation theory. In Gordon's (1964) influential


assimilation theory usually associates this"core group" with the white middle class.Scholars who defend this argument view dis-crimination against Asians and Hispanics asfundamentally similar to that encountered byearlier generations of white immigrants, sug-gesting that the non-European origins ofthese groups should not have a lasting influ-ence that fundamentally alters assimilationprocesses (Glazer 1993; Sowell 1981). Ap-plying these Ideas to friendship, traditionalassimilation theory suggests the followingproposition:

Hypothesis I. Traditional Assimilation: ForAsian and Hispanic Americans, own-group preference in friendship willweaken across immigrant generations,with their cross-race friendship patternsin later generations becoming similar tothe friendship patterns among whites.

Other scholars disagree with these predic-tions, contending that traditional assimila-tion theory, based on a sophisticated form ofpopular melting-pot ideas thought to holdamong early twentieth-century white immi-grants, is limited in its application to con-temporary immigration (for a review, seeRumbaut 1999).̂ * Several differences be-tween past and recent generations of immi-grants are noted, and many accounts particu-larly emphasize the role of racial discrimi-nation arising from the non-European ori-gins of much contemporary immigration(see esp. Chan and Hune 1995; Johnson,Farrell, and Guinn 1999). Discrimination di-rectly influences immigrant adaptation byexcluding the target group from certain av-enues of assimilation, and in addition, ittends to correspondingly increase solidarityamong the target group. The definition of anew immigrant group as racially different isespecially likely to have lasting conse-quences if the definition leads to differentia]treatment that persists across generations.

An influential and systematic statementthat serves as an example of this line of

stages of assimilation, assimilation in friendshipis a form of "structural assimilation."

'' Some scholars have argued that traditionalassimilation theory applies imperfectly even toearly generations of white immigrant groups(e.g.. Glazer and Moynihan 1963).

thought appears in the discussion of "seg-mented assimilation" by Portes and Zhou(1993) and Portes and Rumbaut (1996,2001). Their model primarily considers as-similation in socioeconomic status, but sev-eral ideas from the theory are applicable tofriendship as well. A central point of seg-mented assimilation is the view that, in con-trast to traditional assimilation theory, thereis no single "core group" with which immi-grants merge. Instead, they propose that as-similation theory must consider the question"assimilation to what?" Their description ofmultiple paths of assimilation, influencedpartly by race, is a view we adapt in propos-ing "alternative" paths of immigrant adapta-tion in friendship relations.

In addition to the "traditional" assimila-tion path—toward the characteristics of thewhite middle class—segmented assimilationtheory suggests two alternatives. First, animmigrant minority group that receives asufficiently hostile reception may becomepermanently segregated from the whitemiddle class, similar to the continuing seg-regation between most whites and blacks.We propose tbis as a distinct path that maycharacterize the evolving friendship rela-tions of Hispanic Americans and AsianAmericans:

Hypothesis 2a, Strong Friendship Segmenta-tion: In later generations, the friendshippatterns of Asian Americans and His-panic Americans will remain highly seg-regated from whites and other groups,similar to the continuing segregation be-tween whites and blacks.

A second alternative is that the social re-lations of immigrant minorities will followan intermediate path, one that leads to mod-erately high levels of contact with personsof other races, but remains not fully inte-grated with either blacks or whites. This pat-tern may result from less extreme forms ofdiscrimination by the host society and/orfrom successful efforts of the immigrantgroup to maintain solidarity. Portes andRumbaut (1996) discuss this option as onethat might lead to rapid economic assimila-tion while the group simultaneously and de-liberately preserves the immigrant cultureand etbnic social ties—a choice they call"selective acculturation." We propose this as

5 4 4 AMERICAN SOCIOLOGICAL REVIEW

a final possible pattern of friendship rela-tions for later generations:

Hypothesis 2b, Selective Friendship Assimi-lation: In later generations, AsianAmericans and Hispanic Americans willretain own-group preferences in theirfriendship selections, but the two groupswill show a greater tendency towardfriendships with whites or blacks thanwhites or blacks have for each other.

We believe that these two alternative paths(Hypotheses 2a and 2b) are more likely tocharacterize those of Asian immigrants andHispanic immigrants than the "traditional"path of assimilation to white patterns. Ourexpectations, however, are further differen-tiated by the specific circumstances of Asianand Hispanic Americans.

Asian Americans make up the racial groupmost often described as being on a path toassimilation with white patterns. Supportersof this view (e.g., Glazer 1993; Hacker1993) note that Asian Americans often livein neighborhoods with many whites (Masseyand Denton 1987), intermarry with whites atrelatively high rates (Qian 1997), and areclose to or above white levels on many mea-sures of socioeconomic status (Hirschmanand Snipp 1999). Critics of this view empha-size that later generations of Asian Ameri-cans remain subject to anti-immigrant senti-ment and racial discrimination. They alsopoint out persisting Asian-white gaps onsome measures of socioeconomic status(Chan and Hune 1995; Tuan 1998), andstrong ethnic social capital ties that helpAsians to maintain separate and distinct eth-nic cultures (Bankston and Zhou 1995; Zhouand Bankston 1994). Our overall expectationis that later generations of Asians are likelyto be relatively close to white patterns butthat they will retain some in-group favorit-ism, most closely following the course of se-lective assimilation.

Hispanic Americans are less often de-scribed as taking a path toward full assimi-lation with whites than are Asians becausepersistent gaps remain in socioeconomic sta-tus between Hispanics and non-Hispanicwhites (Hirschman and Snipp 1999). Theimportance of race in assimilation theory,however, raises important additional ambi-guities in the case of Hispanics. "Hispanic"

is an ethnic category that indicates a com-mon ancestral connection to Latin Americaand often to the Spanish language as well.Behind the uniformity implied by this termis the reality of racial diversity among His-panic immigrants, who might phenotypicallyappear to be white, black, or different thaneither whites or blacks.'' Studies that havedistinguished Hispanics by skin tone (or be-tween predominantly "white" and predomi-nantly "black" Hispanic national origingroups) often find sharp differences by racein residential environments and socioeco-nomic outcomes (Denton and Massey 1989;Murguia and Telles 1996).

The influence of race on the experiencesof immigrant minorities suggests that whilethe panethnic category Hispanic may havean important effect on friendship formation,common racial identification may also besignificant to the racial friendship patterns ofHispanic respondents. Consistent with ourexpectation that racial phenotype has conse-quences for immigrant adaptation in friend-ship, we propose a third hypothesis for thefriendship relations of Hispanic students:

Hypothesis 3, Racial Differences amongHispanics: In later generations, friend-ship selection among "white Hispanics"will resemble white patterns, and friend-ship selection among "black Hispanics"will resemble black patterns.

THE EFFECTS OF RACIAL CONTEXT

To this point, we have focused on the conse-quences of the shift toward increasinglyAsian and Hispanic student populations, ar-guing that Asian and Hispanic students arelikely to occupy a structurally different po-sition in friendship relations than do whiteor black students. A second factor that is sys-tematically changed by the increasing Asianand Hispanic student populations is school

^ Phenotypical variations that can be viewedas racial also exist among Asian Americangroup.s from different national origins (e.g. Fili-pino, Japanese, Thai). We expeci the black-whitedimension among Hispanic immigrants will beespecially salient in the U.S. context, however,because it corresponds closely to black-white di-visions among non-Hispanics.


racia! composition. Several establisbed theo-ries in racial and ethnic relations suggest thatthe relative size of racial groups in a settinghas important consequences for intergroupcontact and relations.

In the case of cross-group social ties,Blau's (1977) macrostructural theory pro-vides a basic proposition. We have previ-ously alluded to this proposition as "propin-quity": -

Hypothesis 4. Propinquity: The extent ofcross-race friendsbip between membersof racial group A and racial group B willbe in direct proportion to the relativesize of A to B within the school. As Bincreases in relative size, the share offriendship nominations of group A tomembers of group B will correspond-ingly increase.

Propinquity reflects tbe fact tbat tbe fre-quency of day-to-day contacts among strang-ers strongly predicts the likelihood of friend-ship formation, as demonstrated most con-vincingly by tbe college dorm experiment ofFestinger, Schachter, and Back (1950). Moststudies of black-white relations in class-rooms and schools support this propositionin the case of school friendship patterns, andthus we expect that this will bold for Asianand Hispanic students as well.^ Propinquityimplies that the total number of cross-groupfriendships in a setting will be maximizedwben the groups are equal in size.

A different perspective comes from the lit-erature on the psychology of friendship andethnic solidarity. Social psychological dis-cussions view friendship as providing ameans of sharing, coping with stress, andfinding acceptance. These functions are es-pecially important during adolescence whenfriendships become more intimate and playa large role in tbe development of self-con-cept. Applying tbese friendship functions toracial friendship relations, Tatum (1987,1999) argues that a network including sev-

^' Others have pointed out that cross-racefriendship might decline in integrated settingsbecause of the increasing threat perceived by thedominant group from the subordinate group (St.John and Lewis 1975). Because past studies ofblack-white schools have consistently supportedpropinquity effects, we doubt this prediction willhold in the case of friendsbip relations.

eral racially similar peers is important to ful-fill needs for social support and advice indealing with racial issues, especially whenstudents are faced with experiences of racialdiscrimination or prejudice. Bankston andZhou (1995) make a compatible point intheir study of second-generation Vietnameseyouth. They find that strong ties with co-ethnics provide a resource that supports anorientation toward academic achievementand describe efforts of immigrant parents tomaintain these ties.

Although not noted by these authors, ef-forts to maintain a network of co-racial andco-etbnic friends have implications for com-position effects on friendship formation. Ifstudents act to maintain a network of same-race peers when they are in a small minority,then the strength of same-race selection formembers of racial groups that are small mi-norities witbin their scbool must increase,suggesting the following sociometric pattern:

Hypothesis 5, Social Support: As minoritygroups become small in size, they willincreasingly favor the available same-race peers in friend selection to achievethe goal of maintaining a friendship net-work with several same-race friends.

This proposition reminds us of the positivefunctions of co-racial friendship in a societyin which race is highly consequential, espe-cially for students who are in small numeri-cal minorities. When a group is heavily out-numbered on the basis of a highly salient so-cial categorization, we expect that solidaritywithin the outnumbered group is likely toincrease.

DATA

We examine friendship matching using datafrom the in-school survey of the NationalLongitudinal Study of Adolescent Health(Add Health).'' The survey is a self-adminis-tered questionnaire given to students in 144schools in 80 U.S. communities betweenSeptember 1994 and April 1995. The schools

' Persons interested in obtaining data filesfrom Add Health should contact Add Health.Carolina Population Center, 123 West FranklinStreet, Chape! Hill. NC 27516-2524 (http://www.cpc.unc.edu/addhealth).


were selected to represent the universe of allschools in the United States with students ingrades 7 to 12. All students present in theschools on the days the surveys were admin-istered were included in the sample, result-ing in a total of 90,118 completed question-naires. Because the student questionnaire in-cluded all students in each school, many stu-dents named as friends are also members ofthe sample. This allows us to determine thecharacteristics of friends based on their re-sponses to the survey. For more on the AddHealth Study, see Bearman, Jones, and Udry(1997).'*

We discarded data from seven schools forwhich almost no friendships with other stu-dents in the school were reported (N =5,043),^ and we eliminated data from stu-dents who went to schools in which noschool administrator survey was completedor who did not answer survey questionsabout Hispanic origin or gender. These dele-tions left us with a base sample of 72,957respondents.

In addition, there are a large number ofcases that have missing data (N = 21.276)on at least one of five variables used in ouranalysis: education of motber, occupation ofmother, grade-point average (GPA),mother's place of birtb (native or foreign),and student's place of birth (native or for-eign). Following a suggestion by Little(1992), we imputed values for these vari-ables based on regressions with other co-variates used as explanatory variables. Wealso estimated all results excluding casesmissing on any of these variables. Thischange does not alter the substantive conclu-sions of our analysis.

>* Following Winship and Radbill (1994), wedo not use the Add Health probability samplingweights in models because the sampling weightsare functions of the independent variables in ouranalysis. We also estimated the basic modelswith weights, and this change did not alter oursubstantive findings.

' In seven schools, students named on averageless than one other student in the school as afriend. Students could indicate zero friends bynot writing any down, but the very low averagenumber of friends for these schools probably re-flects an error in the administration of the surveyor in the coding of the data, and thus we elimi-nated these schools.

MEASURING FRIENDSHIP >

The Add Health in-school questionnaire asksstudents to list their five best male friendsand, in a separate question, their five bestfemale friends, including girlfriends andboyfriends.'° Students wrote in the names oftheir five best friends as placeholders, andthen they were asked to look up their friendson a roster of students in their school and asister school and write in the identifyingcode matching that friend. A separate codewas written in to indicate that the friend didnot go to either the sister or sample school.The friendship question does not make itpossible to distinguish between other-sexfriendships and other-sex romantic relation-ships. Because we believe that the dynamicsof romantic relationships are probably dif-ferent than friendships, we analyze onlysame-sex friendships.

As Hallinan (1974) discusses, a disadvan-tage of using fixed-choice questions to mea-sure friendship is that respondents who wantto name more than five friends cannot.Nonetheless, in a comparison of sociometricmethods of friendship selection, Schofieldand Whitley (1983) find this to be a validmethod of friendship identification. Theymake the important point, however, that thismethod tends to identify close friends only,leaving out "friendly acquaintances." Segre-gation among close friendships is likely tobe higher than that among acquaintances.Although we agree that weak friendship ties(like acquaintances) are important to study,we also thitik it is important to study strongfriendship ties because those ties are morelikely to have lasting effects on attitudes andbehaviors (Powers and Ellison 1995).

MEASURES OF RACE

Like the 2000 Census, the Add Health in-school survey presents a series of racial cat-egories and allows respondents to choosemore than one. The possible racial catego-

"'The exact wording of the questions is: "Listyour closest (male/female) friends. List your best(male/female) friend first, then your next bestfriend, and so on. (Girls/Boys) may include(boys/girls) who are friends and (boyfriends/girl-friends)." • • '


ries are; white, black, Asian or Pacific Is-lander, Native American, and other. In aseparate question, respondents are asked ifthey are of Hispanic descent."

This data raises the issue of defining racialcategories for students who check more thanone racial category. We initially considered aconventional coding system that consideredsome racial categorizations as "dominant,"for instance counting everyone who checked"black" and another race as black. We wereconcerned, however, that multiracial studentsmay differ systematically in their friendshippatterns from monoracial students. Instead,we constructed our primary racial categoriesof adolescents who chose only one race. Inall, we use eight categories of race by His-panic origin: white, black, Asian, white His-panic, black Hispanic, other Hispanic, otherrace, and multiracial. The white, black, andAsian categories all include white, black, andAsian single-race students. Likewise, whiteHispanic and black Hispanic are Hispanicstudents who indicate they are white or black,and no other race.

The "other Hispanic" category includestwo patterns of response to the race ques-tions. First, we include respondents whochose Hispanic and "other" as their race,who make up about 36 percent of studentswho are of Hispanic descent. Second, we in-clude students who indicated Hispanic butdid not check any racial category, who makeup 28 percent of persons who indicated His-panic background.'- Our guess is that stu-dents in the "other Hispanic" category thinkof Hispanic as their race rather than think-ing of Hispanic as an ethnic category distinctfrom race.

Finally, we include students who indicatethey are Native American or "other" race inan "other race" category; students whochoose more than one race are included in amultiracial category. These last two catego-ries constitute less than 10 percent of theAdd Health sample and are not analyzed atlength below. In another study we plan to in-

' ' The race question is: "What is your race? Ifyou are of more than one race, you may choosemore than one." The Hispanic question is: "Areyou of Hi.spanic or Spanish origin?"

'̂ Only about I percent of non-Latino respon-dents did not check any racial category.

vestigate the friendship patterns of multira-cial students. Descriptive statistics on friendnominations and the racial composition ofstudents and schools in the survey are shownin Appendix A.

STUDENT AND SCHOOL

CHARACTERISTICS

All student characteristics are based on self-reports from the school survey. Grade-pointaverage (GPA) of the student is based on thestudent's grades in the most recent period forEnglish/language arts, mathematics, history/social studies, and science. Grades were av-eraged to create a single 0 to 4 GPA scale.Mother's years of education completed wasrecoded from questions asking about high-est grade and degree obtained.'^ Mother'soccupation is likewise recoded from a 20-category typology into a dummy variable forwhether or not the mother works in a profes-sional occupation.

Immigrant generation is coded based onplace of birth of the student and the student'smother, as reported by the student on the in-school survey. First-generation students arethose born outside the United States (immi-grants). Second-generation students are bomin the United States to a foreign-bornmother. The third-plus generation consistsof native-bom students whose mothers werealso bom in the United States.

School characteristics are based on schooladministrator reports, except for school ra-cial composition, which is calculated fromstudent self-reports of their race from theschool survey.

METHODS FOR FRIENDSHIPMATCHING

STATISTICAL MODELS

The simplest method of examining interra-cial friendship is to use a dependent variableto indicate either whether a student has anycross-race friendships (Joyner and Kao

'•' We compared student and parents' reportsfor the subsample of students whose parents weresurveyed by Add Health. Child's report andmother's report of mother's education correlate.75; child's report of father's education and


2000) or the number or sbare of friends whoare of another race. A major limitation of thisapproach, however, is that it allows for con-sideration of only a single characteristic ofthe friend chosen in friendship formation.This modeling strategy is suited to consider-ing how characteristics of the individualchoosing a friend influence the choice of asame-race or other-race friend, but it pro-vides no way to examine the simultaneousmatching on the multiple dimensions thatactually characterize friendship formation.

To account for the multiple bases offriendship matching, we employ an approachpreviously used by Hallinan and her col-leagues to study school friendship segrega-tion (Hallinan 1982; Hallinan and Williams1989). The approach was recently extendedby Wasserman and colleagues, whose mod-els for social networks, called P* models,are based on graph-theoretic models of net-work dependence (Anderson, Wasserman,and Crouch 1999; Wasserman and Pattison1996). Although based on a more developedformal model than Hallinan's models, inpractice the approaches are similar. Ourmodels generally follow the P* model ap-proach, with small modifications as appro-priate for the specific goals of our analysis.

In P* models, rather than taking individualstudents as the cases, the cases in the analy-sis are all possible dyadic pairs of respon-dents wbo are in the same school and thuswho could become friends. By coding dyadsin wbich one student named tbe other as afriend as I and other dyads as 0, friendshipcan be modeled as an outcome using logisticregression. Characteristics of the two stu-dents, including relative characteristics (e.g.,whetber both students are of tbe same race,and the difference between students in socio-economic status) serve as independent vari-ables. In the text that follows, senders arethe respondents in our study who are nam-ing their five best friends; receivers are per-sons who could be chosen as friends bysenders—students in the same school.

In practice, the first step to estimating theP* model is to create dyads that represent all

father's report correlate .8. Some disagreementlikely results from error in parental reports. Weconclude tbat student reports are reasonably ac-curate.

possible friendship pairs within a school. Ifthere are N students within a particularschool, there are a total of [A'x(A'-])]/2within-school dyads for that school. Eachdyad is represented twice in the data becauseeach individual appears botb as a sender anda receiver. Note tbat tbe number of dyads in-creases geometrically with school size, be-coming extremely large for large schools: Aschool of 100 students will be represented by(100 X 99) = 9,900 dyads. This data struc-ture implicitly controls for opportunities forcontact among persons of different race intbe school, as interracial dyads are repre-sented in proportion to the racial diversity ofthe student body.

When all within-school dyads are com-bined across the approximately 130 schoolsin our sample, the total number of possiblefriendship dyads is more than 5 million. Toanalyze the data efficiently, we, likeHallinan and Williams (1989), choose asample of dyads. We select all dyads inwhich one individual selected the other as afriend and a share of nonfriendship dyads.We select nonfriend dyads in proportionsstratified by the race of the sender: 5 percentfor non-Hispanic whites, 15 percent for non-Hispanic blacks, 20 percent for Asians, 50percent for wbite Hispanics, 50 percent forblack Hispanics, and 20 percent for otherHispanics. We then employ weights in theanalysis to represent the sampling propor-tions, weighting by the inverse probabilitythat a dyad is included in our sample.

More formally, let /j,y be the probabilitythat the /tb student will name tbe 7th studentin the same scbool as one of ber or his fivebest same-sex friends. We model the logit ofthe probability that the ith student will se-lect the /th student as a friend as a functionof characteristics of the dyad:

' ' ' Z

(I)

This specification follows the P* model inmost respects. Main effects of sender and re-ceiver characteristics—such as mother's edu-cation or respondent's gender—and the main


effect of school characteristics do not appearin this model.''' In a dyadic friendship model,coefficients of maiti effects of sender and re-ceiver characteristics would represent theassociation between these variables and thetotal number of friends (or correlation withsize of school), regardless of the race or othercharacteristics of the friends. Instead of in-cluding main effects, we control for the num-ber of friends selected—friendliness—andthe number of persons who select an indi-vidual as a friend—popularity—with a seriesof dummy variables representing levels offriendliness and popularity (see Anderson,Wasserman. and Crouch 1999).'-'' Tbe log ofthe number of students in the school is alsoincluded as a control variable because theodds that any given dyad will be a friendsbipdyad decline with tbe size of the school.

We estimate tbese models separately forsenders of each race by Hispanic origingroup—one model for non-Hispanic whites,one for non-Hispanic blacks, and so on. Tberace dummy variables shown in equation 1represent the race and Hispanic origin of thereceiver, witb the sender's race and Hispanicorigin group serving as the reference cat-egory. The coefficients of the race variables(tbe 5s) can tben be interpreted in terms oftbe log odds of nominating a receiver of an-olber race and Hispanic origin relative tonominating a receiver of the sender's ownrace and Hispanic origin.

We control several other bases of friend-ship matching using variables representingdifferences between tbe members of tbe dy-adic pair. Anderson et al. (1999) find thatpopular persons are more likely to be friendswith other popular persons, so we include avariable for absolute sender-receiver differ-ence in popularity. A dummy variable indi-cates tbat tbe sender and receiver are in thesame grade in school. Measures of relative

'^ We also estimated models separately foreach gender. We found almost no differences infriendship matching by gender.

'̂ The P* model includes dummy variables(fixed effects) for each sender and receiver. LikeMoody (2001), we control for friendliness ofsender and popularity of receiver rather tban in-cluding dummy variables for each sender and re-ceiver because there are too many respondents inthe sample to include so many dummy variableterms.

socioeconomic status and student scbolasticacbievement are also included in some mod-els, as we will discuss in the results section.

CONTROLS NOT INCLUDED

The goal of our analysis is to estimate tbeextent of friendsbip segregation tbat is likelyto result in scbools witb particular racial andsocioeconomic status compositions. In viewof this objective, we omitted two sets ofvariables that sometimes are included inmodels of friendsbip selection: certain net-work effects that may act to intensify bomo-phily, and measures of similarity in valuesor activity participation.

First, we did not include variables to rep-resent transitivity and reciprocity, which areoften included in P* models. Transitivity re-flects the principle that friendsbip betweentwo individuals is more likely if the indi-viduals have friends in common. Transitiv-ity intensifies racial homophily because newfriends are likely to be racially similar to ex-isting friends, who are likely to be of thesame race as the sender. Because our goal isto estimate tbe extent of friendsbip segrega-tion that is likely to result in a scbool with agiven racial mix, it is important not to "con-trol out" tbe intensification of racial bomo-pbily that occurs througb tbe influence oftransitivity on friendship choices.

Likewise, we exclude controls for choiceof the receiver by tbe sender—reciprocity—representing tbe principle that individualsare more likely to select a friend if the friendreciprocates the choice. Controlling for reci-procity would "control out" part of the raceeffect because the reciprocal choices madeby receivers are partly race-guided.

Second, we do not include measures ofsimilarity of attitudes or similarity of clubmembership that might be a basis for bomo-pbily. In estimating the degree of racialfriendship segregation that is likely to occurin a school witb a particular racial and so-cioeconomic composition, we do not want toimpose counterfactual conditions, sucb as noracial differences in attitudes or integrationof social clubs, because these conditions willnot bold in most schools in tbe future. An-other problem witb these controls is that pastresearch suggests that similarity betweenfriends in attitudes (Kandel 1978) and club


membership (Fink and Wild 1995) are asmuch a result of friendship as a cause offriendship, which results in upward bias tothe estimates of the effects of these controls.

STATISTICAL INFERENCE

The Add Health school survey is a school-based sample—students are clustered withinschools. Accordingly, we calculate standarderrors adjusted using the Huber/White esti-mator with schools as the clusters. In thematching models, the dyadic pairs can beseen as a second level of clustering withinschools, with receivers clustered within thecorresponding sender. Because the adjust-ment used to deal with school clusteringdoes not assume independent observationswithin the primary clustering units (in thiscase, schools), this second level of cluster-ing should not bias the standard errors (seeStataCorp 2001, sect. 30.2.1).

MULTIRACIAL FRIENDSHIPMATCHING .̂

We now turn to the models of friendship se-lection. We begin by considering the patternsof racial homophily for later-generationAsian and Hispanic students to test the con-trasting predictions of traditional immigra-tion theory and our adaptation of segmentedassimilation theory for the cross-race friend-ship pattemsof the third-plus generation. Wethen consider change across generations, andfinally we address racial composition effects.

Do ASIAN AND HtspANicSTUDENTS DIFFER?

Table I shows the estimated coefficients ofthe friendship selection models. Separatemodels are shown for senders from each raceand Hispanic origin group. The race of re-ceiver coefficients in the first panel of thetable indicate the relative likelihood a senderof the indicated group will nominate a re-ceiver of the indicated race and Hispanic ori-gin as a friend, relative to a receiver of thesender's own race and Hispanic origin. Notsurprisingly, these coefficients are negative,indicating that students are less likely to se-lect as a friend a student of another race and/or Hispanic origin status than a student of

their own race and Hispanic origin status.The models also include interaction variablesbetween immigrant generation and the raceof receiver coefficients to allow for variationacross generations in the strength of racialhomophily. Because the interactions are in-cluded, the race of receiver coefficients canbe interpreted as indicating racial homophilyin friendship formation for the third-plusgeneration, which is the reference group inassessing generational change.'^

Several measures are also included as con-trols. The model includes two measures ofparental socioeconomic status: the absolutedifferences in mothers' education betweensender and receiver (educational distance),and a dummy variable indicating whetherboth sender and receiver have mothers inprofessional occupations. The model also in-cludes controls for friendship matching onGPA, popularity, friendliness, and a controlfor school size, as discussed in the methodssection.

Comparing the coefficients for race selec-tion effects to the other predictors, commonrace and Hispanic origin are among thestrongest predictors of friendship among stu-dents. The common race influence on friend-ship is far greater than similarity in parentalsoeioeconomic status: A 20-year differencein mother's education is not as great a bar-rier to friendship as race between black andwhite students.'^ We also found that therace-matching effects are almost unchangedin models without socioeconomic status-matching and GPA-matching controls (notshown), indicating that race-matching doesnot reflect the effect of socioeconomic sta-tus or academic-based cliques.'^

'* Some second-generation interactions weredropped for white Hispanics because sample sizeconstraints led to difficulties in model estimation.Thus, for some receiver groups, the referencegroup for white Hispanics is the second andthird-plus generation combined.

" We also estimated models adding socioeco-nomic status controls for father as well asmother, using only cases for which informationon both parent.s is available. The matching coef-ficients for father's socioeconomic status weresimilar to those for the mother and do not sig-nificantly alter the race-matching effects.

'̂ These results are available from the authorson request.

FRIENOSHIP SEGREGATION SSI

Table L CoefTicients from the Logistic Regression of Friendship on Dyadic Characteristics,Including Interactions with Immigrant Generation: Add Health in-School Questionnaire,1994 to 1995

Independent Variable

Race of ReceiverWhite I

Black

White Hispanic

Black Hispanic

Other Hispanic

Asian

Other race

Multiracial

White

Reference]

-1.56*"(.11)

- .07 '(.03)

-1.55"*(.19)

- .65"*(.09)

- . 3 6 " '(.07)

- .18*"(.04)

- .20*"(.03)

Race of Receiver by First GenerationWhite X First generation

Black X First generation

White Hispanic x Firstgeneration

Black Hispanic x Firstgeneration

Other Hispanic x Firstgeneration

Asian x First generation

.06(.05)

.38'(.19)

.47"(.17)

1.13"(.35)

-.22(.30)

.21(.23)

Race of Receiver by Second GenerationWhite X Second generation

Black X Second generation

White Hispanic x Secondgeneration

Black Hispanic x Secondgeneration

Other Hispanic x Secondgeneration

Asian x Second generation

.06(.04)

.48**(-16)

-.12(-22)

-.62(.62)

.20(.20)

-.15(.15)

Black

-2.16*"(.15)

[Reference]

- 1 . 5 7 " '(.21)

-.09(.06)

-1 .56" '(.19)

-1.84*"(.21)

-.72*"(.10)

- .42"*(.09)

.26(.16)

.09(.10)

.19(.49)

.04(.26)

.80**(.29)

l.IO"'(.28)

.65" '(.15)

.07(.15)

-.10(-34)

.40(.30)

.56*(.23)

.80"(.28)

Race of Sender

WhiteHispanic

-.36***(.10)

-1.23"*(.17)

IReference]

-.51(.28)

-.35*(.15)

-.54**(.17)

- . 5 2 "(.16)

- .35*"(.08)

-.34*(.16)

-.58(.55)

.27(.18)

-.92*(.43)

.40 ' "(.12)

-.07(.38)

.02(.08)

-.40(.28)

t Dropped]

[Dropped]

. 4 1 " '(.11)

-.53*(.26)

BlackHispanic

-2.45***(.24)

- . 3 6 "(.13)

-2.08*"(.37)

[Reference]

-1.43"*(.31)

-1.42*"(.24)

-.64*"(.18)

-1.49*"(.36)

.94"(.32)

.39(.24)

1.08(.90)

.29(.54)

1.37"(.51)

.22(.92)

1.02***(.31)

.09(-19)

1.47'(.59)

-.52(.61)

.53(.45)

-.36(.55)

OtherHispanic

-.91"*(.10)- . 9 2 ' "(.15)

- . 4 4 ' "(.11)

-.82***(.21)

[Reference]

-.99***(.20)

- . 6 1 * "(.14)

-.50***(.10)

-.33**(.11)

-.70***(.13)

.35"(.12)

-.69(.35)

.32*(.13)

-.46*(.18)

-.19*(.10)

- . 3 9 "(.13)

.19(.13)

-.01(.30)

.19*'(.06)

-.15(.19)

Asian

- . 8 8 ' "(.11)

-1 .71"*(.37)

-.71*(.32)

-1.60(1.07)

- 1 . 9 1 ' "(.50)

[Reference]

- .89*"(.13)

- .87"*(.10)

- . 5 1 * "(.12)

-.19(.34)

- .85 '(.36)

-1.34(1.21)

.17(.45)

.31(.16)

-.20*(.08)

-.30(.35)

-.54(.29)

.15(1.03)

.42(.43)

.19(.13)

(Continued ort next page)


(Table I continued from previous page)

Independent Variable White

Socioeconomic Status and AchievementMothers' educational

distance

Both mothers areprofessionals

Students' GPA distance

1

ControlsSame grade '

Size of school (logged)

Difference in popularity

Network Dummy VariablesFriendliness

Popularity

- . 0 5 " '(-01).10" '

(.03)

- . 3 7 " '

urn2.61***(.06)-.85'"(.05)

-.09'"(.00)

Yes

Yes

Number of dyads 708,510

Black

-.04*"(.01).13'

(.06)

-.18'"(.04)

2.24'"(.11)- .63 '"(.10)

-.09"'(.01)

Yes

Yes

300,595

Race

WhileHispanic

-.04'"(.01).16*

(.08)

-.35"'(.03)

2.31'"(.07)-.87'"(.04)

-.10'"(.01)

Yes

Yes

459.200

of Sender

BlackHispanic

-.02(.01).16

(.12)

-.22"'(.06)

2.24'"(.09)

-.65'**(.10)

- .11 ' "(.02)

Yes

Yes

88,865

OtherHispanic

-.07'"(.01)

.22'"(.05)

- .28'"(.03)

2.16"'(.09)-.82"'(.08)- . 11" '(.01)

Yes

Yes

268,320

Asian

-.04"'(.0!)-.04(.06)

-.42'"(.04)

2.86'"(.18)- . 71" '(.13)-.10'"(.01)

Yes

Yes

221,020

Notes: Standard errors are in parentheses and are calculated using the HuberAVhite formula, adjusted forschool clustering. Some coefficients were dropped due to .small sample sizes of sender-receiver pairs torsome generations. All models were estimated with a constant term, but the constant is not shown. Modelswere estimated using pseudo-maximum-likelihood methods and thus standard likelihood-ratio tests are in-valid.

*p < .05 "p < .01 "'p < .001 (two-tailed tests)

To clarify the cross-group patterns. FigureI shows the exponentiated race and Hispanicorigin selection coefficients from Table 1,which are more easily interpreted than theraw coefficients. The bars on the graph indi-cate the odds that a within-school dyad withthe indicated sender race and receiver racewill be a friendship dyad, relative to the oddsof selecting a friend of the same race andHispanic origin as the sender. The graphshows the estimates for third-plus generationsenders only. The value of .21 for "Wchooses B," for instance, indicates that theodds are only about 21 percent as great that awhite sender-black receiver dyad will be afriendship dyad as a white sender-white re-ceiver dyad. By contrast, white students haverelatively high odds of nominating other His-panic, Asian, and (especially) white Hispanicstudents as friends.

Do later-generation Hispanic and Asianstudents have much greater odds of friend-ship across racial and ethnic lines than do

whites or blacks, controlling for opportuni-ties for contact? We find some evidence thatrace is a lower barrier to friendship betweenthird-plus generation Hispanic and Asianstudents and certain other-race groups thanit is between non-Hispanic white and non-Hispanic black students (this is discussed indetail below). We also fmd, however, thatthird-plus generation white Hispanics, blackHispanics, other Hispanics, and Asians showclear own-race preferences in their friend-ship nominations that are nearly as strong as(or in some cases stronger than) the own-race preferences of non-Hispanic whites andnon-Hispanic blacks.

Figure 1 also indicates some friendship-matching on the basis of Hispanic ethnicity.Cross-race friendships are especially likelyamong students who share Hispanic ethnic-ity, and most likely between white Hispanicand other Hispanic students.

Asian students are often viewed as thenonwhite group that is closest to whites, and

FRIENDSHIP SEGREGATION 5S3

o£OJ

O

w chooses B

W chooses WH

W chooses BH

WctraosesOH

W chooses A

BchoosetW

BchooHsWH

Bchooses BH

B chooses O

B chooses A

WH chooses W

WH chooses B

WH chooses BH

WH ctiooses OH

WH chooses A

BH choosse W

BH chooses B

BH chcoBBB WH

BH ctTOOSeS OH

BH chooses A

OH chooses W

OH chooses B

OH chooses WH

OH chooses BH

OH chooses A

A chooses W

A chooses B

A chooses WH

A chcoses BH

A chooses OH

Race Categories:W = White

B = Black

WH = White Hispanic

BH = Black Hispanic

OH = Other Hispanic

A = Asian

0 .1 .2 .3 ,4 .5 ,6 .7 ,8 .9 1.0

Odds of Friendship, Relative to an Own-Group Dyad (Own-Group = 1)

Figure 1. Odds That a Cross-Group Dyad Will Be a Friendship Dyad, Relative to an Own-GroupDyad, by Race of Sender and Receiver: Third-Plus Generation, Add Health In-SchoolQuestionnaire, 1994 to 1995

Note: Figure is based on the exponentiated race selection coefficients shown in Table 1. Odds of selectionare shown as a proportion of the odds that an own-group dyad in a school will be a friendship dyad (odds ofown-group dyad friendship = 1).

thus we expected white-Asian friendships tobe among the most common cross-racefriendships in the third-plus generation. Wefind some evidence consistent with this idea,but tbis evidence is mixed. In support of lowsocial distance between Asian and white stu-dents, when an Asian student nominates another-race friend, whites {including white

Hispanics) are the most preferred choice,and white students are relatively likely tonominate Asian receivers as friends. On theother hand, Asian students display a strongwithin-race selection preference, with lowodds of selecting friends of other races.

The findings of persisting own-group pref-erence for members of most racial groups


contradict the prediction of traditional immi-gration theory—that third-plus generationAsian and Hispanic students will assimilateto white friendship patterns (Hypothesis I).Because we find that white Hispanic, otherHispanic, and Asian students are more likelyto have cross-race friendships than arewhites with blacks, we also reject stronglysegmented patterns of friendship relationsfor these groups (Hypothesis 2a) in favor ofthe pattern we previously referred to as se-lective assimilation in friendship relations(Hypothesis 2b).

We also find that the white-black divide isclearly evident in Hispanic friendship selec-tions, consistent with Hypothesis 3. Non-Hispanic whites and white Hispanics arelikely to nominate each other as friends, asare non-Hispanic blacks and black Hispan-ics.'^ Generally, biack-nonblack is an impor-tant division in student friendships, a divi-sion also found among students of Hispanicdescent. These results suggest that white His-panics and black Hispanics are being incor-porated into white and black friendshipcircles, respectively, with Hispanic origin asan important secondary influence.'^"

GENERATIONAL CHANGE IN

FRIENDSHIP RELATIONS

Although race and Hispanic origin clearlymatter for friendship patterns of the third-plus generation, this result does not precludethe possibility that own-group preferencesbecome weaker over immigrant generations.We now consider that possibility. To allowthe racial homophily effects to vary by gen-eration, the models in Table 1 include inter-action terms between race of receiver and

" The coefficient for white Hispanic selectionof black Hispanics is marginally statistically sig-nificant {p = .07) because in our sample there arefew later-generation white Hispanics in the sameschools as black Hispanics. In models that poolimmigrant generations, the coefficient for whiteHispanic selections of black Hispanic receiversis -.83 (significant at p < .01).

^̂ This may occur partly because the race offriends influences racial identification. Hispanicstudents with white friends may be more likelyto identify as white, while Hispanic students withblack friends may be more likely to identify asblack.

dummy variables indicating first or secondimmigrant generation. The third-plus gen-eration is the reference category.

If there is weakening own-race preferenceacross generations, we should find positiveown-group and negative other-group interac-tion terms between the first and second gen-eration and the race of receiver variables.Most of the interaction terms between gen-eration and race of receiver in Table 1, bow-ever, are not statistically significant, indicat-ing no statistically discernible differencesbetween the first and second generation andthe third-plus generation. Some significantcoefficients are also in the direction oppo-site that predicted by traditional assimila-tion. Only one group—other Hispanics—fol-lows a pattern that is clearly consistent withthe weakening of own-race preference andincreasing other-race preference predictedby traditional assimilation theory (Hypoth-esis 1).

Indeed, much more notable than the dif-ferences across generations are the similari-ties. To clarify the overall patterns of racialhomophily across generations, we calculatedthe odds that a cross-race dyad will be afriendship dyad for each immigrant genera-tion, based on the coefficients in Table 1. Wethen computed correlation coefficients of theodds that a cross-race dyad will be a friend-ship dyad across the three immigrant genera-tion groups. The resulting correlations, us-ing the 36 sender-receiver race and Hispanicorigin pairs as the cases (6 sender race/His-panic origin groups by 6 receiver race/His-panic origin groups), are shown in top panelof Table 2. The correlation coefficients forthe odds of selection over generations are allabove .8, indicating highly similar patternsof racial and Hispanic origin friendshiphomophiiy over the three immigrant genera-tion groups. The bottom panel of Table 2sbows a similar calculation including onlypairs with Hispanic or Asian senders, andprovides very similar results. Correlationscalculated for specific sender race and His-panic origin groups (not shown) yield simi-lar conclusions, with correlations above .7.

The weak reductions of within-racefriendship selection across generations insome instances, most clearly other Hispan-ics, are consistent with traditional assimila-tion theory. But because the dominant pat-

FRIENDSHrP SEGREGATION 555

Table 2. Pearson Correlation Coefflcients forthe Odds of Friend Selection, by Raceand Hispanic Origin, betweenImmigrant Generations: Add HealthIn-School Questionnaire, 1994 to 1995

ReceiverGeneration

Sender Generation

First Second Third-Plus

Correlations over Generations for All Sender-Receiver Pairs (36 pairs}

First 1.00 — —

Second .83 1.00 —>

Third-Plus .82 .85 1.00

Correlations over Generation for Sender-ReceiverPairs with a Hispanic or Asian Sender (24 pairs)

First IXX> — —

.S8

.80

Second

Third-Plus

1.00

.81 1.00

Notes: Results are based on correlations of oddsof friend selection calculated over the pairs ofsender/receiver race and Hispanic origin. Odds offriend selection are based on calculations from theexponentiated model coefficients in Table I.

tern is of similarity in racial homophilyacross generations rather than change, wefind more support for persistent racial effectson friendship networks than for generationalassimilation in friendship relations.

RACIAL COMPOSITION ANDFRIENDSHIP SELECTION

To this point, the models have been struc-tured so that the odds that an interracial dyadwill be a friendship dyad depend only oncharacteristics of the students involved, mostnotably on the race and Hispanic origin ofthe two students who form the dyadic pair.Because there are more cross-race dyads inracially diverse schools than in racially ho-mogeneous schools, these dyadic modelsimply that the number of cross-race friend-ships increases in diverse schools, thus im-plicitly incorporating propinquity effects.

Although we have used models that as-sume these effects to be present, we nowrespecify the models to test this assumption.Our respecified models also allow us to ex-amine the possibility that friend-making be-havior changes depending on the racial com-position of the school population, in particu-lar our hypothesized social support effects.

Finally, because our analysis in Table 1found few significant differences across im-migrant generations, we drop the interac-tions with immigrant generation from themodels.^'

To allow for variations in the strength ofracial selection with racial composition, weintroduce interactions between the race ofreceiver variables and dummy variables rep-resenting the racial composition of theschool population. We interacted all race andHispanic origin receiver variables withdummy variables representing racial compo-sition in three categories: 10-30 percent, 30-60 percent, and 60-100 percent own race.The reference category is 0-10 percent own-race. In defining the own-race percent ofschool for Hispanic students, we use schoolpercent white for white Hispanic senders,school percent black for black Hispanicsenders, and school percent other-race His-panics for other-race Hispanic senders.

Table 3 displays selected coefficients fromthis new model, focusing on the strength ofselection of friends from one's own racialgroup.^^ Figure 2 graphs the estimatedstrength of own-group selection by racialcomposition of the school, from the ex-ponentiated coefficients in Table 3, withown-group selection in a 0 to 10 percentschool normalized to equal 1.

The results show that the odds that anown-group dyad will be a friendship dyadare highest for students in a small racial mi-nority in their school (less than 10 percentand/or 10-30 percent), are lower for studentsin 30-60 percent own-race schools, and donot change much further in 60-100 percentown-race schools. These composition ef-fects are strong. For instance, the odds areabout 10 times greater that a white-whitedyad will be a friendship dyad in a 0-10 per-cent white school than in a 60-100 percentwhite school (^""^ =.09 = 1/10). The num-ber of white-white dyads increases more

^' We calculated the Bayesian Information Cri-teria (Raftery 1995) for the models with andwithout the interactions, which indicated "verystrong" support {a BIC difference of about 40)for the simplified models. These statistics do notadjust for school-level clustering, however, andthus we regard the results as advisory.

^̂ Ail racial coefficients in the model areshown in Appendix B.

S S 6 AMERICAN SOCIOLOGICAL REVIEW

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


School Composition EffectsSchool 0-10% own-race

X Receiver own-race

School 10-30% own-raceX Receiver own-race

School 30-60% own-raceX Receiver own-race

White

[Reference]

-1.46" '(.21)

-1 .92 '"(-15)

School 60-100% own-race -2 .38" 'X Receiver own-race (.14)

Race of ReceiverWhite

Black ' •

White Hispanic

Black Hi.spanic

Other Hispanic

Asian ] 1

Other race

Multiracial

Number of dyads

[Reference]

-3 .42" '(.17)

.-1.88"*(.54)

-3.10"*(.19)

-2 .75" '(.16)

-2.53"*(.15)

-2 .49 '"(.15)

-2.51"*(.14)

708.510

Black

[Reference]

-.87*"(.12)

-1.47"*(.13)

-2.04" '(.12)

-4 .27 '"(.43)

[Reference]

-3.32"*(.39)

-.21(.20)

-2.45'**(.20)

-2.75'**(.18)

-2.34'**(.14)

-2.00"*(.11)

300.595

Race of Sender

WhiteHispanic

[Reference]

.10(.31)

.32(.25)

-.58(.30)

-.15(-55)

-1 .62 '"(.35)

[Reference]

-.68(.35)

- . 6 2 " '(.18)

-.50"*(.15)

- . 7 0 ' "(.12)

-.54"*(.1!)

459,200

BlackHispanic

[Reference]

-.15(.69)

-I . I6(.68)

-1.15(.67)

-2.97"(1.09)

.36(.70)

-2.27"*(.61)

[Reference]

-1.60'(.80)

-1.84"(.65)

-2.34"*(.69)

-1.53*(.66)

88,865

OtherHispanic

[Reference]

.13(.27)

-.52'(.25)

[Dropped]

-.63*(.26)

-1.47*"(.32)

-.70**(.24)

- t . 3 2 "(.44)

[Reference]

- 1 . 2 1 ' "(.27)

-1.13*"(-25)

-1.04*"(.24)

268,316

Asian

[Reference]

- .97" '(.19)

[Dropped!

[Dropped]

-3 .65" '(.30)

-2 .33" '(.29)

-2.46"*(.69)

- 2 . 8 1 ' "(.61)

-1.48***(.26)

[Reference]

-2.01"*(.22)

-1.99"*(.19)

221.020

Notes: Standard errors are in parentheses and are calculated using Huber/White formula, adjusted forschool clustering. Own race is white for whites, black for blacks, Asian for Asians, white for white Hispan-ics, black for black Hispanics, and other Hispanic for other Hispanics. Models are estimated using pseudo-maximum-likelihood methods and thus standard likelihood-ratio tests are invalid. Also included but notshown: racial composition interactions for other-race group and all controls in Table 1. Appendix B showscoefficients for all racial composition interactions.

'p < .05 "p < .01 '"p < .001 (two-tailed tests)

quickly than tbis—the number of satne-racedyads increases geometrically with tbe num-ber of same-race peers^—^so the model resultsstill imply tbat white students in a 60-100percent wbite scbool will have more wbitefriends than those in a 0-10 percent wbiteschool. Thus, as we consider further in Table4, tbese results remain consistent witb thebasic prediction of propinquity.

Because of the intensification of own-racehomophily for students who are a small ra-

cial minority in their scbooi, tbe increase inown-race friends present in 60-100 percentown-race schools compared with 0-10 per-cent own-race schools is not nearly as largeas would be predicted based purely on oppor-tunities for contact. Students in schools witbfew same-race peers are especially likely tobefriend the (few) other same-race peersavailable, consistent with Hypothesis 5.

Tbese composition effects also explainsome differences in overall degree of own-


1.40

>Qa

a3o

5OcCO

20 -

Do£ 1.00

•C .80 -

.60 -

.40 -

.20 -

School Racial CompositionIJHi] 0-10% own-race0 10-30% own-race

|x|j 30-60% own-race

^ 60-100% own-race

t̂

\

White Black White Hispanic Biack Hispanic

Race and Hispanic Origin of Sender

Other Hispanic Asian

Figure 2. Odds That an Own-Group Dyad Will Be a Friendship Dyad, hy School Raeial Compositionand Raee and Hispanic Origin of Sender: Add Health In-Sehool Questionnaire, 1994 to1995

Note: Figure is based on the exponentiated coefficients in Table 3. Bars indicate the odds that an own-group dyad will be a friendship dyad in a school of the indicated racial composition, relative to a 0-10percent own-race school.

•" Relative to a 0-10 percent own-race school (0-10% own-race = 1),' p < .05 (significantly different from 0-10% own-race school)

race bias. Examining the third-generationcoefficients in Table 1, for instance, non-Hispanic whites are more likely to chooseA.sian friends than Asians are to choose non-Hispanic white friends. In Table 3, we seethat the disparity results in part becauseAsians are more often part of a small racialminority within their schools than arewhites.

How much of the racial segregation infriendship results from school racial segre-gation? Table 4 draws implications of thesemodels for the overall share of friends of an-other race. The top panel of Table 4 presentsthe percentage distribution of friends nomi-nated in the raw Add Health data, brokendown by sender's race, as a basis for com-parison.

The bottom panel of Table 4 gives the pre-dicted share of friends from each race for

senders of each race for a hypotheticalschool in which the racial mix of the schoolis equal to the overall raeial distribution ofstudents in the Add Health data (except weassume no students are in the multiracial orother race categories). Tbe models shown inTable 3 (and in Appendix B) are used for thepredictions. In tbis hypothetical scboo!, allstudents have the same socioeconomic back-ground and the same GPA, and all studentsare equally popular.

In tbe bottom panel. Table 4 results showa clear school propinquity effect that holdsfor all racial groups, consistent witb our ba-sic Hypothesis 4. All groups except non-His-panic u-hites and non-Hispanic blacks havea majority of friends in other groups, reflect-ing the small size of these groups in this hy-pothetical school and the greater odds of in-terracial friendships for Asians and Hispan-


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

School Racial Race ofComposition Receiver White Black

Race of Sender

WhiteHispanic

Race of Named School Friends in the Add Health In-School Survey ^— White

' '— Black

— White Hispanic

— Black Hispanic

— Other Hispanic

— Asian

90.0

1.9

2.6

.23.22.1

Estimated Share of Friends of Each Race inHealth Student Race Distribution

59% White

f#6 Black4% White Hispanic

2% Black Hispanic

9% Other Hispanic

8% Asian

79.7

3.24.5

.4

5.6

6.6

7.6

81.5

.6

5.8

3.3

1.3

49.6

2.5

18.1

.6

26.5

2.7

a Hypothetical School with

15.2

59.5

1.8

10.2

7.8

5.6

66.9

5.8

8.8

1.5

8.2

8.8

BlackHispanic

8.3

72.3

1.3

10.1

6.4

1.6

Composition

15.7

51.4

2.0

12.4

10.4

8.1

OtherHispanic

20.4

4.9

7.6

.7

62.8

3.7

Asian

27.9

3.8

1.5

.3

7.0

59.4

Reflecting the Add

41.3

11.1

7.4

3.1

27.8

9.4

40.2

5.7

2.2

1.0

13.3

37.6

Notes: Estimates in the bottom panel are based on models in Table 3 and Appendix B. Estimates in thebottom panel assume that both students in a dyad have the same GPA and mothers with the same socioeco-nomic background, and that the students are equally popular.

''Excludes multiracial and other-race categories.

ics than for black studetits. At the same time,all groups select members of their owngroup in substantially greater share thantheir proportionate presence in the school,reflecting the continuing influence of racialhomophily in friendship selection.

OTHER CONSIDERATIONS:TRACKING, BUSING, ANDNETWORK SIZE

We also considered several other influenceson social networks that might be importantin understanding the social relations of stu-dents in increasingly diverse schools. Webriefly discuss the main results of these ad-ditional analyses, none of which substan-tially altered the strength of racial homo-phily or altered our main findings. (Tablesshowing these results are available from theauthors on request.)

First, we thought it possible that muchfrietidship segregation might result from thesegregation of academic tracks. The AddHealth survey does not provide the informa-

tion necessary to identify the track of indi-vidual students, but it is possible to identifyfrom the school administrator questionnairewhether the school includes separate in-structional tracks. We estimated our basicmodels of friend selection including inter-actions allowing for variation in thestrength of homophily between schoolswith and without tracking. Consistent withresults reported by Hallinan and Williams(1989) and Kubitschek and Hallinan (1998),we found slightly stronger racial homophilyin schools with tracking. The difference,however, was small and accounted for littleof the overall racial homophiiy in friend-ship.

Second, we considered that racial integra-tion in the schools in our sample might beachieved as a part of race-conscious policiesof student school assignment, often involv-ing busing. If these policies have negativeinfluences on interracial relations in schools,the influence of busing might be confoundedwith racial composition effects. Our analy-sis, however, found no statistically signifi-


cant negative effects of busing on the forma-tion of interracial friendships.--*

Third, we considered whether students insmall racial minorities in a school mighttend to have fewer friends overall. Changesin total network size are not captured in ourmatching models, but could contribute tofewer cross-race friendships for studentswhose racial group is the numerical minor-ity in their school. We estimated models ofthe number of friends named on the AddHealth survey as a function of racial compo-sition to investigate this possibility. Wefound a statistically significant reduction inthe size of social networks for white, whiteHispanic, and Asian students who are mem-bers of small minorities (less than 10 per-cent) in their school population. Thesechanges were substantively small, however,and not sufficient to alter our conclusions.

CONCLUSIONS

Our results provide only weak support forthe prediction of traditional assimilationtheory, as applied to race and Hispanic back-ground, that across generations friendshipnetworks tend to converge toward a singleAmerican "core" standard. We do find gen-erational changes for some racial and His-panic-origin groups in the direction pre-dicted by traditional assimilation theory, butthe main friendship pattern is of consider-able similarity in racial and Hispanic-originhomophily across immigrant generafions.Race and Hispanic background are powerfulfactors influencing adolescent friendship,with racial similarity a substantially strongerpredictor of friendship formation than paren-tal similarity in socioeconomic status.

Some have suggested that Asian studentsare on a path of assimilation toward whitepatterns, a view Tuan (1998) calls becoming"honorary whites." Asian students do tend tohave many white friends, but this occurs pri-marily because of propinquity effects, as

^̂ This conclusion differs from Moody's(2001). The difference in results could be becauseMoody examines total racial segregation, withoutseparately considering races involved. Schoolswith busing in the Add Health data have highershares of blacks and whites than schools withoutbusing, which may result in higher apparent over-all friendship segregation in the school.

Asian students are usually a small minorityin their schools. Although there is some evi-dence that white students are relativelylikely to select Asian friends, we find thatAsian students are considerably more likelyto nominate an Asian student than a whitestudent as a friend, relative to the number ofAsian and white schoolmates. We concludefrom these results that Asian students appearto be following a path of group solidarity intheir primary group relations, displayingsome affinity for white friends, but withoutassimilation to white or black patterns. Thisis consistent with the suggestion of Portesand Rumbaut (1996) that Asian immigrantsmay be following a path of "selective accul-turation" in which partial assimilation is ac-companied by the simultaneous maintenanceof ethnic ties.

Hispanic students are in the unique posi-tion in our study of having their friendshipselections divided by crosscutting racial andethnic (Hispanic origin) lines. We find thatHispanic origin is an important basis forfriendship matching, most notably betweenwhite Hispanic students and Hispanic stu-dents who indicate their race as "other." Yetwe also find that race remains highly salientto the friendship choices of Hispanic stu-dents—the friends of white Hispanics aremostly whites, white Hispanics, and otherHispanics; the friends of black Hispanics aremostly blacks and black Hispanics. Theseresults suggest that, while ethnic solidarityexists, white and black Hispanics are assimi-lating in different ways, with racially whiteHispanics joining white peer groups and ra-cially black Hispanics joining black peergroups.

We take these results as contributing to theevidence that racial difference is importantin understanding processes of immigrant ad-aptation and should be included in theoriesof immigration adaptation. The influence ofblack and white race on friendship patternsamong Hispanics provides an especiallyclear example of how the racial categoriza-tion of an immigrant group may have a last-ing influence on the incorporation of thegroup members into the host society.

Although we view the stability of racialhomophily across generations as mostly in-consistent with traditional assimilationtheory as applied to racial groups, we note


that traditional assimilation theory may stilldescribe other dimensions of immigrant ad-aptation. An analysis of generational changein friendship homophily based on national-origin ethnicity, for instance, may well bemore consistent with traditional assimilationtheory. That is, the persistent racial and His-panic-origin homophily we find here couldrepresent homophily on the basis of national-origin groups for early generations and racialor panethnic selection in later generations.

Consistent with past research, we findclear evidence of propinquity effects for stu-dents of all races; The share of cross-racefriends depends strongly on the share of po-tential friends in the school context who areother-race. We also find, however, that stu-dents in schools in which they are membersof a small racial minority substantially in-crease their odds of own-race friend selec-tion. We believe this to be because studentsdesire several friends of their own racialgroup for reasons of social support and theyalter their friend-making behavior to achievethis goal when there are only a few same-race friends in their school. Future researchincorporating direct measures of the atti-tudes, beliefs, or feelings that are relevant toracial differences in friendship selectionwould be useful to provide more definitiveevidence regarding this interpretation of the"intensification" effect.

We note three final caveats. First, our re-sults reflect the organizational and racial en-vironments present in our sample of juniorand senior high schools, and racial relationsin American society more generally. Ifschool officials were to manage integratedenvironments differently, or racial relationsin American society were to significantlychange, our conclusions would need to becorrespondingly adjusted. Second, we havesaid little about multiracial students becausethe complexity of multiracial identity re-quires a separate analysis. The inclusion ofmultiracial students would somewhat in-crease the extent of cross-race friendship inour results. Finally, adolescents are at a stagein the life cycle in which racial identity andracial awareness crystallize. A study ofadults in social contexts with racial compo-sitions similar to our sample of schoolsmight well find lower levels of racial homo-phily in friend selection.

DISCUSSION

If current trends continue. Hispanic studentswill soon become the largest minority groupin schools, and Asians will make up 10 per-cent of all students by mid-century (Orfieldand Yun 1999). Will the increasing preva-lence of these new immigrant minoritiesbreak down high levels of social segregationamong students of different races? How willthe growing presence of these new immi-grant minorities alter the structure of cross-race networks among students?

We tlnd evidence consistent with reductionin the extent of racial segregation in friend-ship, but we also find continuity in severalrespects. Our results indicate that interracialand interethnic friendships in schools willincrease as the share of students who are ofAsian and Hispanic origins increases, be-cause of propinquity effects and because raceis less of a barrier to friendship among somecombinations of non-Hispanic white, whiteHispanic, other Hispanic, and Asian studentsthan between white students and black stu-dents. The clear own-race preference of allgroups in their friendship nominations alsosuggests, however, that own-race friends willcontinue to be overrepresented in the socialnetworks of most students relative to theirschool's racial composition.

We also find that black-white divisionswill continue to be important in the increas-ingly Hispanic schools of the future, becausewhite Hispanic and black Hispanic studentsare joining, respectively, white and blackpeer groups. Disturbingly, we find especiallyhigh levels of segregation of blacks, includ-ing black Hispanics, from all other racialgroups. Our results suggest that black-non-black is an important dividing line in multi-racial schools, a division that will likely be-come more evident as the share of schoolpopulations who are neither white nor blackincreases.

The increased own-race favoritism infriendship among students who are membersof small racial minorities suggests that dis-persing minority students in very small con-centrations would not necessarily maximizethe extent of cross-race ties in a school dis-trict (see Feld and Carter 1998). Instead,maintaining more balanced racial propor-tions in some schools might actually lead to

FRIENDSHrP SEGREGATION 5 6 1

more reciprocated cross-race friendships be-cause it will not result in the intensificationof homophily that we find among students inespecially small minorities. More detailedconclusions sorting out own-race intensifica-tion and propinquity effects will require adetailed formal model that is beyond thescope of this paper. Our tentative conclusionis that a low degree of clustering of a racialminority in schools may be desirable to in-crease reciprocated cross-race friendships insituations in which the minority represents asmall share of the overall population of aschool district.

Although race will remain an importantinfluence on student friendship choices inmultiracial schools, our results do not meanthat the current levels of segregation infriendship are inevitable. On the contrary, theschool propinquity effects indicate that in-creasing school racial diversity is likely toincrease friendships across racial lines, espe-cially when combined with the greater likeli-hood of cross-race friendship for Asian andHispanic students and the possible weaken-ing of "intensification" effects as numbers ofAsian and Hispanic students grow beyond atoken few in more schools. Our results sug-gest that a substantial increase in racial di-versity in schools will lead to a notable in-crease in cross-race friendships, relative tocurrent levels.

Yet this potential will only be realized ifthe increasing diversity of student popula-tions translates into increasing racial diver-sity within schools. We believe tbat demo-graphic change makes at least small in-creases in within-school racial diversity very

likely in the future, but we are uncertain ofhow closely within-school diversity willmirror increasing diversity in the populationat large. One recent study finds that a largeshare of Hispanic students in the Southwestare educated in almost entirely Hispanicschools, raising the possibility that growingdiversity in the school-aged populationmighi be accompanied by increased school-level segregation (Frankenberg, Lee, andOrfield 2003). All else equal, tbe increasingdiversity of American society should trans-late into greater within-school diversity—but we cannot assume that all else will nec-essarily be equal. Continued attention to ra-cial proportions in schools remains impor-tant if schools are to fulfill their promise asinstitutions in which Americans learn tomanage racial diversity and interact acrosslines of racial and cultural difference.

Lincoln Quillian is Associate Professor of Soci-ology at the University of Wisconsin-Madison.His research focuses on inequality, racial atti-tudes, and urban population distribution. Hiscurrent research projects investigate the causesand consequences of urban spatial segregationon the basis of income and the friendship net-works of multiracial adolescents.

Mary E. Campbell is a Ph.D. candidate in Soci-ology at the University of Wisconsin-Madison.Her research focuses on racial identity and ra-cial stratification in education and employment.She is currently completing her dissertation,which investigates the patterns of racial identifi-cation for multiracial Americans who are forcedto choose a single race and explores stratifica-tion outcomes for multiracial adolescents andadults.

A P P E N D I X A

Mean Number uf Same-Sex Friends and Student/School Counts, by Race and School RacialComposition: Add Health In-School Questionnaire, 1994 to 1995

Race of Sender/School Racial Composition

White Respondents (N = 38.512)

School is<10% while

School is 10-30% white


School is >60% white

Mean Nutnber of:

Own-RaceFriends in School

1.02

1.02

2.06

2.39

Other-RaceFriends in School

IMIM.56.36

Number ofStudents

397

2,038

12,994

23,083

(Number ofSchools)^

(14)

(17)

(48)

(54)

(Continued on next page)


(Appendix A continued from previous page)

Race of Sender/School Racial Composition

Mean Number of:

Own-Race Other-RaceFriends in School Friends in School

Number ofStudents

(Number ofSchools)*

Black Respondents (N = 11,243)

School is<IO% black

School is 10-30% black


School is >60% black -•

While Hispanic Respondents (N = 2,606)

School is <10% white



School is >60% white

Black Hispanic Respondents (N = 1,027)

School isclO^t black



School is >60% black

Other Hispanic Respondents (N = 8.428)School is <tO%' Other Hispanic

School is 10-30% Other Hispanic

School is 3 0 ^ 0 % Other Hispanic

Asian Respondents (N = 3.479)

School is <10% Asian

School is 10-30% Asian

Other respondents (N = t .564)

Multiracial respondents (N = 6,113)

.79

1.15

1.35

1.73

.65

.69

1.36

2.10

.53

.72

.90

1.17

.29

.88

1.19

.37

1.34

.14

.37

1 (

.81

.66

.58

.83

1.16

.68 .

1 .

[J^ -.53

1.72

.99

.57 . •• • • n . : - .

1.64 .

.92

2.00^

1.74

1,673

3.406

3.571

2,593

1.012

271

694

629

197

315

341

219

1.161

2,473

4.794

1.134

2.345

_

(61)

(3!)

(14)

(7)

(11)

(16)

(48)

(51)

(43)

(29)

(14)

(7)

(87)

(23)

(12)

(95)

(13)

—

_

Notes: Own race friend.s are while for whiles, 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 forpersons of other race, and multiracial for multiracial persons.

'Includes all schools with at least one student of the indicated race/Hispitnic origin group.

APPENDIX BCoefTicients Trum the Regression of All Race and School Racial Composition Interactions in Modelswith Race Composition Effects: Add Health In-School Questionnaire, 1994 to 1995


Race of Receiver

White

Black

White Hispanic

While

[Reference]

-3.42"*(.17)

-1 .88"*(.54)

Black

^ . 2 7 " '(.43)

[Reference]

-3.32"*(.39)

WhileHispanic

-.15(.55)

-1.62"*(.35)

[Reference]

BlackHispanic

-2.97**(1.09)

.36(.70)

-2.27***(.61)

OtherHispanic

- . 6 3 '(.26)

-1.47"*(.32)

- . 7 0 "(.24)

Asian

-3 .65""(.30)

-2.33"*(.29)

- 2 . 4 6 " '(.69)


FRIENDSHIP SEGREGATION 5 6 3

(Appendi.x B continued from previous page)


Race of Receiver (Continued)

Black Hispanic

Other Hispanic

Asian

Other race

1

Multiracial

Wdte

-3.10"*(.19)

-2.75"*(.16)

-2 .53"*(.15)

- 2 . 4 9 " '(.15)

-2 .51"*(.14)

School Racial Composition xRate

0-10% white XWhite receiver


30-60% white xWhite receiver


0-10% black XBlack receiver

10-30% black xBlack receiver

30-60%. black xBlack receiver

60-100% black XBlack receiver

0-10% whiLex WhiteHispanic receiver

10-30% white x WhiteHispanic receiver

30-60% white x WhiteHispanic receiver

60-100% white X WhiteHispanic receiver

0-10% black X BlackHispanic receiver


30-60% black x BlackHispanic receiver


[Reference [

-1.46"*(.21)

-1.92*"(.15)

-2.38'*'(.14)

1 Reference [

- . 6 6 "(.20)

-.65*'(.24)

-.76*(.38)

[Reference[

-.05(.50)

-.29(.50)

-.59(.49)

[Reference]

- 1 . 6 1 " "(.34)

-1.38""(.41)

- 1 . 7 1 "(.61)

Black

-.21(.20)

-2.45*""(.20)

-2.75*"'(.18)

-2.34""(.14)

-2.00*"(.11)

[Reference]

.83(,44)

.51(.44)

.61(.44)

[Reference]

- . 8 7 ' "(.12)

-1.47""(.13)

-2.04***(.12)

[Refcrcntjc[

.69(.45)

.10(.42)

.22(.43)

[Reference]

-.78*'(.26)

-1.26""(.24)

-1.94""(.24)

WhiteHispanic

-.68(.35)

- .62*"(.18)

-.50*"(.15)

- .70*"(.12)

- .54*"(.11)

[Reference]

-.07(.59)

-.27(.58)

-.47(.58)

[Reference]

-.10(.42)

.23(.42)

.59(.57)

[Reference]

.10(.31)

.32(.25)

-.58(.30)

[Reference]

-.78(.62)

-.48(.82)

1.33(.71)

BlackHispanic

[Reference]

-1.60'(.80)

-1 .84"(.65)

-2.34***(.69)

-1.53*(.66)

[Reference]

-.13(.86)

-.24(.82)

-.07(.80)

[Reference]

-.88**'(.17)

- 1 . 4 5 " '(.19)

-2.04"*(.16)

[ Reference]

-.09(.77)

-.29(,86)

-.41(.81)

[Reference]

P -.15>' (.69)

-1.16(.68)

- I . I5(.67)

OtherHispanic

-1 .32"(.44)

[Reference]

- 1 , 2 1 " "(.27)

-1.13*"(.25)

-1.04*"(.24)

[Reference]

- . 8 1 " "(.18)

-1.10*"(.25)

- . 9 3 " '(.13)

[Reference]

-.35(.30)

-.53(.33)

.55(.30)

[Reference]

-.13(.16)

.00(.16)

- .46 '(.18)

[Reference]

-.13(.39)

-.57(.43)

-.30(.77)

Asian

-2 .81"*(.61)

-1.48"*(.26)

[Reference]

-2 .01*"(.22)

-1.99*"(19)

[Reference]

1.26*"(.33)

1.23***(.27)

1.45"*(.25)

[Reference]

-.84*(.35)

-1 .17"(.36)

.23(.25)

[Referencel

.12(.70)

-.24(.70)

.33(.71)

[Reference]

-.65(.97)

-.73(1.25)

1.54(1.15)



{Appendix B continued from previous page)

[ndependent Variable White Black

School Racial Composition x Race (Continued)

0-10% other Hispanic xOther Hispanic receiver



0-10% Asian x Asianreceiver

tO-30% Asian x Asianreceiver

Number of dyads

1 Reference]

-.24(.15)

- .44'(.22)

[ Reference 1

-.34*(.14)

708,510

(Reference]

-.31(.27)

-1.14**"(.21)

[Reference]

- . 8 1 * "(.24)

300.595

WhiteHispanic

[Reference]

.47(.31)

.40*(.19)

[Reference!

- . 6 8 "(.24)

459.200

BlackHispanic

(Reference]

-.06(.53)

-.89(.51)

[Reference]

-1.34"*(.36)

88,865

OtherHispanic

[Reference]

.13(.27)

-.52*(.25)

(Reference |

- , 7 9 "(.30)

268.316

Asian

1 Reference]

-1.26*"(.26)

-1.56"*(.34)

[Reference]

- .97""

221.020

Notes: Standard errors are adjusted for school clustering. Models are estimated using pseudo-maximum-likeU-hood methods and thus standard likelihood-ratio tesEs are invalid. Also included but not shown: all variables inTable 3. Also see notes to Table 3.

if' Shaded cells indicate composition effects on own-race friend selection. Race effects and diagonal (own-race) effects are also shown in Table 3.

'p < .05 "p < .01 '"p < .001 (two-tailed tests)

REFERENCESAllport, Gordon. 1954. The Nature of Prejudice.

Cambridge. MA: Addison-Wesley.Anderson, Carolyn J., Stanley Wasserman, and

Bradley Crouch. 1999. "A P* Primer: LogitModels for Social Networks." Social Networks21:37-66.

Bankston. Carl L.. Ill and Min Zhou. 1995. "Ef-fects of Minority-Language Literacy on theAcademic Achievement of Vietnamese Youthsin New Orleans." Sociology of Education68:1-17.

Bearman, Peter S., Jo Jones, and J. Richard Udry,1997. "The National Longitudinal Study ofAdolescent Health: Research Design." Re-trieved July 29, 2003 (http://www.cpc.unc.edu/addhealth).

Blau, Peter. 1977. Inequality and Heterogeneity:A Primitive Theory of Social Structure. NewYork: Free Press.

Chan, Kenyon and Shirley Hune. 1995. "Racial-ization and Panethnicity: From Asians inAmerica to Asian Americans." Pp. 205-56 inToward a Common Destiny: Improving Raceand Ethnic Relations in America, edited by W.Hawley and A. Jackson. San Francisco. CA:Jossey-Bass.

Clotfelter. Charles. 2002. "Interracial Contact inHigh School Extracutricuiar Activities." Ur-ban Review 34:25-46.

Dawkins, Marvin P. and Jomills Henry Braddock

II. 1994. "The Continuing Significance of De-segregation: School Racial Composition andAfrican American Inclusion in American So-ciety." Journal of Negro Education 63:394-405.

Denton, Nancy and Douglas Massey. 1989. "Ra-cial Identity among Caribbean Hispanics: TheEffect of Double Minority Status on Residen-tial Segregation." American Sociological Re-view 54:790-808.

Feld. Scott L. 1982. "Social Structural Determi-nants of Similarity among Associates." Ameri-can Sociological Review 47:797-801.

Feld, Scott L., and William C. Carter. 1998."When Desegregation Reduces InterracialContact: A Class Size Paradox for Weak Ties."American Journal of Sociology 103:1165-86.

Festinger, Leon. Stanley Schachter. and KurtBack. 1950. Social Pressures in InformalGroups: A Study of Human Factors in Hous-ing. Stanford, CA: Stanford University Press.

Fink, Benedykt and Klaus-Peter Wild. 1995."Similarities in Leisure Interests: Effects ofSelection and Socialization in Friendships."Journal of Personality and Social Psychology135:471-82.

Frankenberg, Erica. Chungmei Lee, and GaryOrfield. 2003. "A Multiracial Society withSegregated Schools: Are We Losing theDream?" A report of the Civil Rights Project

FRIENDSHIP SEGREGATION 56S

of Harvard University. Harvard University,Cambridge, MA. Retrieved July 29, 2003(httpL//w ww.civilrightsproject.harvard.edu/research/resegO3/reseg-full.php).

Gerard, Harold and Norman Miller. 1975. SchoolDesegregation: A Long-Term Study. NewYork: Plenum.

Glazer, Nathan. 1993. "Is Assimilation Dead?"Annals of the American Academy of Politicaland Social Science 530:122-36.

Glazer, Nathan and Daniel Moynihan. 1963. Be-yond the Melting Pot. Cambridge. MA: MITPress.

Gordon, Milton. 1964. Assimilation in AmericanLife: The Role of Race. Religion and NationalOrigins. New York: Oxford University Press.

Haeker, Andrew. 1993. Two Nations: Black andWhite, Separate, Hostile. Unequal. New York:Ballantine Books.

Hallinan. Maureen. 1974. The Structure of Posi-tive Sentiment. Amsterdam, The Netherlands:Elsevier Scientific.

. 1982. "Classroom Racial Compositionand Children's Friendships." Social Forces61:56-72.

Hallinan, Maureen T. and Stevens S. Smith.1985. "The Effects of Claiisroom Racial Com-position on Students' Interracial Friendliness."Social Psychology Quarterly 48:3-16.

Hallinan, Maureen T., and Richard A. Williams.1989. "Interracial Friendship Choices in Sec-ondary Schools." American Sociological Re-view 54:67-78.

Hirschman, Charles and Matthew Snipp. 1999."The State of the American Dream: Race andEthnic Socioeeonomic Inequality in the UnitedStates, 1970-1990." Pp. 89-107 in A NationDivided: Diversity. Inequality, and Communityin American Society. New York: ColumbiaUniversity Press.

Johnson. James H., Jr. Walter C. Farrell Jr., andChandra Guinn. 1999. "Immigration Reformand the Browning of America: Tensions, Con-flicts and Community Instability in Metropoli-tan Los Angeles." Pp. 390-411 in The Hand-book of International Migration. New York:Russell Sage Foundation.

Joyner, Kara and Grace Kao. 2000. "School Ra-cial Composition and Adolescent RacialHomophily." Social Science Quarterly 81:810-25.

Kandel, Denise. 1978. "Homophily, Selection,and Socialization in Adolescent Friendships."American Journal of Sociology 84:427-36.

Kubitsehek, Warren and Maureen Hallinan.1998. "Tracking and Students' Friendships."Social Psychology Quarterly 61:1-15.

Lieberson, Stanley. 1961. "A Societal Theory ofRace and Ethnie Relations." American Socio-logical Review 26:902-10.

Little, Roderick. 1992. "Regression with MissingX's: A Review." Journal of the American Sta-tistical Association 87 (420):1227-37.

Massey, Douglas S. and Nancy Denton. 1987."Trends in the Racial Segregation of Blacks,Hispanies, and Asians: 1970-1980." AmericanSociological Review 52:802-25.

McPherson, Miller, Lynn Smith-Lovin, andJames M. Cook, 2001. "Birds of a Feather:Homophily and Social Networks." Annual Re-view 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 MexicanAmericans." Sociology of Education 69:276-89.

Orfield, Gary and John T. Yun. 1999. "Resegre-gation in American Schools." A report of theCivil Rights Project of Harvard University.Harvard University, Cambridge, MA. Re-trieved July 29, 2003 (http://www.civilrightsproject, harvard.edu/research/resegO3/reseg-full.php).

Patchen, Martin. 1982. Black-White Contact inSchools: Its Social and Academic Effects. WestLafayette, IN: Purdue University Press.

Portes. Alejandro and Ruben C. Rumbaut. 1996.Immigrant America: A Portrait. Berkeley, CA:University of California Press.

. 2001. Legacies: The Story of the Immi-grant Second Generation. Berkeley, CA: Uni-versity of California Press.

Portes, Alejandro and Min Zhou. 1993. "TheNew Second Generation: Segmented Assimi-lation and Its Variants among Post-1965 Im-migrant Youth." Annals of the American Acad-emy of Political and Social Science 530:74-96.

Powers, Daniel A., and Christopher G. Ellison.1995. "Interracial Contact and Black RacialAttitudes: The Contact Hypothesis and Selec-tivity Bias." Social Forces 74:205-26.

Qian, Zhenchao. 1997. "Breaking the Racial Bar-riers: Variation in Interracial Marriage Be-tween 1980 and 1990." Demography 34:263-76.

Raftery, Adrian. 1995. "Bayesian Model Selee-tion in Social Research." Sociological Method-ology 25:11)-63.

Rumbaut, Ruben G. 1999. "Assimilation and ItsDiscontents: Ironies and Paradoxes." Pp. 172-95 in The Handbook of International Migra-tion. New York: Russell Sage Foundation.

St. John, Nancy H. and Ralph G. Lewis. 1975."Race and the Social Structure of the Elemen-tary Classroom." Sociology of Education 48:346-68.

Schofield, Janet and H. A. Sagar. 1977. "Peer In-teraction Patterns in an Integrated Middle

see AMERICAN SOCIOLOGICAL REVIEW

School." Sociometry 490:130-38.Schofield, Janel Ward and Bernard E. Whitley,

1983. "Peer Nomination vs. Rating Scale Mea-surement in Children's Peer Preferences." So-cial Psychology Quarterly 46:242-51.

Sigelman, Lee and Susan Welch. 1993. "TheContact Hypothesis Revi.sited: Black-White[nteraction and Positive Racial Attitudes." So-cial Forces 7 \:l^\-95.

Sowell, Thomas. 1981. Ethnic America. NewYork: Basic Books.

StataCorp. 2001. Stata Statistical Software 7.0User's Guide. College Station, TX: Stata Cor-poration.

Tatum, Beverly Daniel. !987. AssimilationBlues: Black Families in a White Community.New York: Basic Books.

. 1999. Why Are All the Black Kids SittingTogether in the Cafeteria? New York: BasicBooks.

Tuan, Mia. 1998. Forever Foreigners or Honor-ary Whites? The Asian Ethnic Experience To-

day. New Brunswick, NJ: Rutgers UniversityPress.

Wamer, Lloyd and Leo Srole. 1945. The SocialSystems of American Ethnic Groups. New Ha-ven, CT: Yale University Press.

Wasserman, Stanley, and Philippa Pattison. 1996."Logit Models and Logistic Regression for So-cial Networks: I. An Introduction to MarkovGraphs and P*." Psychometrica 61:401-25.

WeiLs, Amy Stuart and Robert Crain. 1994. "Per-petuation Theory and the Long-Term Effectsof School Desegregation." Review of Educa-tional Research 64:531-53.

Winship. Christopher, and Larry Radbill. 1994."Sampling Weights and Regression Analysis."Sociological Methods and Research 23:230-57.

Zhou, Min and Carl L. Bankston IIL 1994. "So-cial Capital and the Adaptation of the SecondGeneration: The Case of Vietnamese Youth inNew Orleans." International Migration Review28:821^5.

beyond black and white: the present and future of multiracial ... - kenneth … · 2019. 11....

Documents