At the Margins: a Distinctiveness Approach to the Social Identity and Social Networks of Underrepresented Groups Author(S): Ajay Mehra, Martin Kilduff, Daniel J

Home , Social identity theory

At the Margins: A Distinctiveness Approach to the Social Identity and Social Networks of Underrepresented Groups Author(s): Ajay Mehra, Martin Kilduff, Daniel J. Brass Source: The Academy of Management Journal, Vol. 41, No. 4 (Aug., 1998), pp. 441-452 Published by: Academy of Management Stable URL: http://www.jstor.org/stable/257083 Accessed: 06/01/2009 11:53

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=aom.

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is a not-for-profit organization founded in 1995 to build trusted digital archives for scholarship. We work with the scholarly community to preserve their work and the materials they rely upon, and to build a common research platform that promotes the discovery and use of these resources. For more information about JSTOR, please contact [email protected].

Academy of Management is collaborating with JSTOR to digitize, preserve and extend access to The Academy of Management Journal.

http://www.jstor.org ? Academy of Management Journal 1998, Vol. 41, No. 4, 441-452.

AT THE MARGINS: A DISTINCTIVENESS APPROACH TO THE SOCIAL IDENTITY AND SOCIAL NETWORKS OF UNDERREPRESENTED GROUPS

AJAY MEHRA University of Cincinnati

MARTIN KILDUFF DANIEL J. BRASS The Pennsylvania State University

Using distinctiveness theory, this research showed that the relative rarity of a group in a social context tended to promote members' use of that group as a basis for shared identity and social interaction. Relative to majority group members, racial minorities and women in a master of business administration cohort were more likely to make identity and friendship choices within-group. The marginalization of racial minorities in the friendship network resulted both from exclusionary pressures and from minority individuals' own preferences for same-race friends. By contrast, the marginalization of women resulted more from exclusionary pressures than from their preferences for woman friends.

People are social beings who seek to establish ranks (Brass, 1985). A recent report, for example, ties of identity and friendship with others. In orga- showed that only 57 women (compared to 2,373 nizational settings, diverse groups of people use men) held positions in the highest ranks of Fortune these ties for social support and work accomplish- 500 companies (Catalyst, 1996). ment. But the process of identification and friend- Given the rarity of studies that examine the net- ship formation may unfold differently for members works of both women and minorities (see Ibarra of minority groups and members of majority [1995] for one such study), it remains unclear groups. whether women and members of racial minority Research on the patterning of social relations in groups face similar pressures in informal networks. organizations has suggested the importance of vis- Research does suggest that people tend to interact ible categories such as race and sex as bases for with similar others, and this is particularly true for identification and network formation (e.g., Hughes, relations, such as friendship, that are more expres- For the extent to which women are 1946). example, sive than instrumental (Blau, 1977). Together with a token in a work rather than a presence setting exclusionary pressures from the majority, this pref- substantial of the workforce tends to proportion erence for similar, or "homophilous," others may influence informal interaction patterns (Kanter, contribute to segregation within informal networks Access to informal networks is 1977a). important (Brass, 1985). because, to done in indi- get things organizations, The homophily proposition, however, leaves the viduals must draw on both instrumental resources basis of similarity unspecified. In a social context as work-related advice and and (such sponsorship) that includes men and women of different races, it emotional resources (such as friendship) that infor- is unclear whether people are more likely to iden- mal network contacts offer (Ibarra, 1993). Lack of tify with and select friends on the basis of sex, race, access to informal networks may be one reason that or some other nominal characteristic. We sought to women and minorities, who are entering organiza- clarify the patterns and consequences of such net- tions in unprecedented numbers, are still under- work preferences (1) by examining the extent to represented, especially in upper-management which membership in salient demographic groups influenced social identification and interaction & We thank three reviewers and patterns (cf. Ely, 1995; Tsui, Egan, O'Reilly, 1992) anonymous Angelo and the extent to which members DeNisi for insights and recommendations that helped (2) by examining develop our arguments.Thanks also to Dennis Gioia and of underrepresented groups tended to occupy the Giuseppe (Joe) Labianca for constructive comments on margins of informal social networks. previous drafts. Our sample consisted of individuals enrolled in

441 442 Academy of ManagementJournal August an elite master of business administration (M.B.A.) group. But this prediction still left unanswered the program that functioned as one of the portals to question of which of several possible underrepre- management in corporate America (Kilduff & Day, sented groups any particular individual will tend 1994). These managers-in-training made network to identify with most strongly. For example, when and social identity choices in a campus setting that is an African American woman more likely to feel imposed relatively few of the hierarchical con- strongly African American, and when is she more straints on interaction characteristic of formal or- likely to feel strongly female? Distinctiveness the- ganizations. We compared the identification and ory suggests that race will be a more salient basis friendship patterns of women with those of men, for identity when a person is in a group numeri- compared the patterns of whites with those of ra- cally dominated by those of the same sex as the cial minorities, and examined the structural mar- focal person but of a different race, and sex will be ginality of those groups. a more salient basis for identity when a person is in a group dominated by those of the same race but of the other sex. Distinctiveness theory suggests that THEORYAND HYPOTHESES the salience of a category as a basis for social identification is a function of its relative rarity in a Distinctiveness and Social Identity given context. What determines individuals' identifications 1. The with others? Distinctiveness (McGuire, Hypothesis relative rarity of a social theory in a social will 1984) suggests a parsimonious answer: People in a category particular setting pro- social context tend to with others with mote members' use of that social category as a identify basis social whom they share characteristics that are relatively for identification. rare in that context. Thus, two African Americans In our sample, members of racial minorities were in a of crowd whites will tend to notice and iden- numerically rarer than women. For racial minori- with each other of common tify because their race; ties, we predicted that race would be a stronger when in a of however, group other African Ameri- category for social identification than sex. How- the same two are to notice or cans, people unlikely ever, for whites, the same reasoning suggested that with each other on the of identify basis race. Ac- sex, not race, would be a stronger category for so- cording to distinctiveness theory, the attention- cial identification. salience grabbing of distinctive characteristics is Similarly, we predicted that the salience of race the for social identification. basis Distinctiveness relative to sex would help determine whether extends of theory understanding homophily by people more often chose same-sex or same-race suggesting that similarity is relative to the context. friends. To the extent that an individual is in a In a of the test distinctiveness theory, salience of numerical minority with respect to sex or race, then was for ethnicity higher minority (African Ameri- that category becomes more salient as the basis for can and Hispanic) grade school children than it friendship choice. was for those in the majority (whites). Only 1 percent of the white majority children spontaneously Hypothesis 2. The relative rarity of a social mentioned ethnicity in self-descriptions, compared category in a particular social setting will tend to 17 percent of the African American and 14 per- to promote members' use of that social cate- cent of the Hispanic children (McGuire, McGuire, gory as a basis for friendship formation. Child, & Fujioka, 1978). A follow-up study examined the effect of the sex composition of a group on Marginality the use of sex as a self-identifying characteristic. The likelihood of a child's mentioning his or her Members of underrepresented groups are likely sex in a self-description increased as a function of to be less central in friendship networks than mem- the number of opposite-sex others in the child's bers of well-represented groups because of the household (McGuire, McGuire, & Winton, 1979). former'stendency to select friends from the distinc- Similarly, in an experiment using ad hoc groups, tive groups to which they belong rather than from identification based on sex was more frequent in the social network as a whole. Indeed, the general the spontaneous reports of members of the minority human tendency toward sex and race homophily in sex in mixed-sex groups (Cota & Dion, 1986). friendship choices may work to reduce the central- Drawing on distinctiveness theory, we predicted ity of members of underrepresented groups (who that members of numerically underrepresented have fewer similar others to choose from) relative groups, relative to those in the majority, would to the centrality of members of majority groups (see exhibit a stronger tendency to identify within- the discussion in Ibarra[19931). 1998 Mehra, Kilduff, and Brass 443

Previous theorizing has emphasized exclusion- significantly from respondents with respect to race ary pressures that tend to relegate underrepre- or sex. sented group members to the margins of social networks. Kanter that members of a (1977b) argued Measures group underrepresented in an organization tend to be viewed as tokens, as Os in a series of Xs. Mem- The identity network. The social component of bers of the majority tend to avoid friendship with identity "is the perception of oneness" with others tokens because tokens are viewed through negative (Ashforth & Mael, 1989: 21). These perceptions of stereotypes, because performance failures by to- the social self are necessarily "relational and com- kens tend to attract disproportionate attention, and parative" (Tajfel & Turner, 1985: 16). Thus, we because majority members exaggerate their differ- measured social identity by asking individuals to ences from tokens in order to preserve in-group look down an alphabetical list of second-year distinctiveness. M.B.A. students and place checks next to the Thus, the structural marginality of members of names of those people they considered especially underrepresented groups may well be overdeter- similar to themselves. Individuals were free to mined: it is due both to the friendship choices of make identity choices based on individually salient underrepresented group members and to exclu- criteria, unaffected by researcher-imposed catego- sionary pressures and biases that focus on visible ries (cf. Kelly, 1955). demographic characteristics such as race and sex. The friendship network. We measured friend- The following hypotheses summarize the preced- ship by asking respondents to look down an alpha- ing discussion: betical list of second-year M.B.A. students and place checks next to the names of those people they Hypothesis 3. Members of numerically under- considered to be personal friends. represented groups, relative to members of ma- Homophily. In measuring race and sex homoph- jority groups, are more likely to be structurally ily in the identity and friendship networks, we marginal (less central) in friendship networks. controlled for the relative availability of different Hypothesis 4. Homophily-based friendship ties groups (cf. Ibarra, 1992) because what may appear will be negatively related to centralityfor mem- as a tendency on the part of, for example, women to bers of underrepresented groups and positively form friendships with men, may be attributable to related to centrality for members of majority the proportionally higher number of men in a groups. group. The adjusted homophily index, known as the correlation coefficient Gower and 5. Visible character- point (see Hypothesis demographic [1986] for a review and Krackhardt [1990] istics, such as sex and race, will be Legendre negatively for the from -1 of related to members formula), ranged (indicative centrality for of underrep- extreme to +1 of ex- resented and related to cen- "heterophily") (indicative groups positively treme members homophily). trality for of majority groups. Sex. This was coded as 1 for men and 0 for women. METHODS Race. Using photographs from the school direc- tory and information from publicly available stu- Sample dent r6sumes detailing membership in societies The sample for this study consisted of a class of such as the Black Students Association, two people second-year M.B.A. candidates enrolled in a na- independently coded respondents as either white, tionally ranked M.B.A. program. Nonresidents of African American, Asian American, or Hispanic the United States were excluded from the sample (these were standard categories used by the admin- (and from all questionnaires) because the original istration at this school). Agreement between the research design focused on the job choice process two coders was high (98 percent interrater agree- and included only those eligible to work in the ment). Disputed cases were resolved through dis- United States. The average age of the respondents cussion and a search for further information in the was 27 years. Of the 209 students sampled, 181 (87 r6sume book published by the school. For the ho- percent) completed mailed copies of the sociomet- mophily and regression analyses, we dichotomized ric questionnaire. Missing data reduced the sample race as 0 for whites and 1 for all others. to 159 people, 76 percent of the original popula- As a check on how reliably the coding repro- tion. The final sample included 95 white men, 44 duced individuals' self-coding of race, our coding white women, 10 racial minority men, and 10 racial was compared with the official school records on minority women. Nonrespondents did not differ 113 individuals who had voluntarily reported their 444 Academy of Management Journal August race. Only one person had been misclassified (our value. To check whether the results were affected classification was white, but the self-classification by this definition of the friendship measure, we was Hispanic). There was complete agreement be- also symmetrized the matrix using two alternate tween the two codings for all of those we had rules: (1) replace Xij and Xji by the minimum of (Xij classified as minority group members and for or Xji) and (2) replace Xij and Xji by the average of whom self-report records existed (17 people). We (Xij or Xji). The pattern of results remained the concluded that our coding of race reproduced self- same. ratings at an acceptable degree of accuracy. The Major. Most of the students in the sample had absence of questionnaire items concerning race or chosen one of two majors: finance (56 percent of sex insured that the questionnaire itself did not the sample) or marketing (26 percent), with the trigger salient categories for reporting social iden- remaining students (18 percent) choosing a number tity or friendship. of other possible concentrations. Because we were Structural marginality. Those on the margins interested in the core/periphery structure of the have difficulty accessing the center of a network social world of the M.B.A. students, we dichoto- either through their own friends (direct ties) or mized choice of major to differentiate those stu- through friends of friends (indirect ties). To capture dents choosing popular majors (finance or market- both direct and indirect friendship ties, we used an ing, coded as 1) from those choosing unpopular eigenvector measure (Borgatti, Everett, & Freeman, majors (coded as 0). This dichotomization resulted 1992) that computed centrality as the summed con- in a significant effect for the control variable in our nections to others weighted by the centrality of analyses, whereas a coding representing all possi- those others (see Bonacich [1972] for the formula). ble majors had no significant effects. Marginality was defined as the converse of central- those low on scored on ity: scoring centrality high ANALYSES AND RESULTS marginality. Because the eigenvector analysis program han- The mean homophily values given in Table 1 dled only symmetric data, for this analysis we sym- show that, with availability controlled for, individ- metrized the friendship matrix, using the rule that uals tended to identify with (x = 0.04, s.d. = 0.11) if either member of a pair nominated the other, then and form friendships with (x = 0.04, s.d. = 0.12) the pair was a friendship pair. This operational others of the same race. Similarly, individuals definition preserved information on weak ties (cf. tended to identify with (x = 0.05, s.d. = 0.08) and Granovetter, 1973) and produced the most robust form friendships with (x = 0.04, s.d. = 0.13) others indicator of centrality as measured by the ratio of of the same sex. Individuals tended to establish the largest eigenvalue to the next highest eigen- smaller identity networks (x = 5.06, s.d. = 3.58)

TABLE 1 Descriptive Statistics and Correlationsa

Variable Mean s.d. 1 2 3 4 5 6 7

1. Majorb 2. SexC .02 3. Raced -.09 -.13 4. Centrality 9.09 6.53 .21** .15 -.22*** Race homophily 5. Friendship network 0.04 0.12 .15t -.00 .25*** .16* 6. Identity network 0.04 0.11 .07 -.08 .32*** .25** .47*** Sex homophily 7. Friendship network 0.04 0.13 .12 -.13 -.19* .15t .07 .18* 8. Identity network 0.05 0.08 .03 -.27*** -.06 .07 .05 .17* .34***

aN= 159. b Finance and marketing = 1, other majors = 0. c Men = 1, women = 0. d Minorities = 1, whites = 0. p < .10 * p < .05 ** p < .01 *** p < .001 1998 Mehra, Kilduff, and Brass 445 than friendship networks (x = 16.07, s.d. = 9.43), rows 5 and 6 in Table 2 show that, as predicted, although the two networks were significantly cor- whites were significantly more likely to make related (r = .28, p < .001). friends with others on the basis of sex rather than Recall that Hypothesis 1 suggests that people race (t = -2.62, df = 137, p < .01), whereas minor- tend to identify with those with whom they share a ities were significantly more likely to make friends demographic characteristic that is relatively rare. with others on the basis of race rather than sex The mean homophily values in Table 2 provide (t = 4.63, df= 18, p < .001). support for this hypothesis. The results for the The results presented in Table 3 confirm that identity network presented in the top half of Table these univariate effects of race and sex on the ten- 2 show that the tendency for minorities (x = 0.13) dency to make in-group network choices remained to identify within-group was significantly stronger significant when control variables were introduced (t = -2.03, df = 19.5, p < .05) than that of whites into the analyses. The regression analysis results (x = 0.02). Similarly, the tendency for women (x = presented in the first column (labeled "Sex") show 0.09) to identify within-group was significantly that the tendency to identify with and make friends stronger (t = 3.82, df = 157, p < .001) than that of with members of one's own sex was stronger for men (x = 0.04). Further, the paired comparison women than for men, with an individual's race and t-tests in the first two rows of Table 2 show that, as choice of major controlled for. The regression re- predicted, whites were significantly more likely to sults under the second column (labeled "Race") identify with others on the basis of sex rather than show that minorities were more likely than whites race (t = -3.99, df = 137, p < .001), whereas to identify and make friends within-group, with sex minorities were significantly more likely to iden- and choice of major controlled for. tify with others on the basis of race rather than sex The third hypothesis suggests that members of (t = 1.72, df = 18, p < .05). underrepresented groups are likely to be structur- The patterns of friendship choices paralleled ally marginal in a friendship network. This hypoth- these results, as predicted by Hypothesis 2. Look- esis was supported. Men (x = 9.84, s.d. = 6.89) ing at the bottom half of Table 2, the tendency for were more central than women (x = 7.63, s.d. = members of minority groups (r = 0.16) to make 5.72), and this difference was significant (t = friends within-group was significantly stronger (t = -2.02, df = 157, p < .05). Similarly, whites (x = -3.28, df = 20.3, p < .01) than that of whites (x = 9.59, s.d. = 6.59) were more central than minorities 0.02). Similarly, the tendency for women (x = .06) (x = 5.59, s.d. = 5.47), and this difference was also to make friends within-group was significantly significant (t = 2.58, df = 157, p < .01). stronger (t = 2.11, df = 137, p < .05) than that of The first regression model in Table 4 confirms men (x = 0.02). The paired comparison t-tests in that, with major and sex controlled for, members of

TABLE 2 Mean Homophily Values Showing Tendency to Choose Partners Similar to Selfa

Type of Homophily

Group n Sexb Racec t df

Identity network Whites 139 0.06 (0.72) 0.02 (0.93) -3.99*** 137 Minorities 20 0.04 (0.68) 0.13 (0.36) 1.72* 18 Men 105 0.04 (0.77) 0.03 (0.85) -0.96 103 Women 54 0.09 (0.60) 0.06 (0.85) -1.45 52 Friendship network Whites 139 0.05 (0.65) 0.02 (0.90) -2.62** 137 Minorities 20 -0.06 (0.40) 0.16 (0.27) 4.63*** 18 Men 105 0.02 (0.69) 0.03 (0.86) 0.46 103 Women 54 0.06 (0.41) 0.05 (0.82) -0.68 52 a Unadjusted homophily values are in parentheses. b Men = 1, women = 0. c Minorities = 1, whites = 0. * p < .05 **p < .01 ***p < .001 446 Academy of Management Journal August

TABLE 3 subsample (r = .46, p < .05) was significantly Summary of Regression Analyses Predicting higher (Z = 3.92, p < .001) than the same correla- Homophilya tion for minorities (r = -.46, p < .05). Our fifth hypothesis suggests that visible demo- of Type Homophily graphic characteristics, such as sex and race, will be related to for Variable Sexb Racec negatively centrality underrepresented group members and positively related to Identity network centrality for majority group members. Model 2 in Major 0.01 (0.02) -0.01 (0.02) Table 4 shows that, with a marginally significant Sex -0.05*** (0.01) -0.02 (0.02) (p < .10) effect of sex homophily controlled for, sex Race -0.02 0.10*** (0.02) (0.02) had a significant effect (p < .05) on centrality. This Model F 5.56*** 6.80*** pattern of results suggests that women were less 0.10 0.11 R2 central in the friendship network not so much be- Friendship network cause of their tendency to prefer woman friends, Major 0.03 (0.03) 0.04 (0.02) but more as a result of their exclusion on the basis Sex -0.05*** (0.02) -0.01 (0.01) of Model 3 in Table 4 shows with a Race -0.11*** (0.03) 0.14*** (0.02) gender. that, < .01) effect of race con- F 6.34*** significant (p homophily Model 10.98*** trolled race had a < on R2 0.11 0.18 for, significant (p .01) effect centrality. These results suggest that the marginal- aN = 159. Values represent unstandardized coefficients; ity of members of racial minorities was due both to standard errors are in parentheses. race and to exclusion on the basis of b homophily Men = 1, women = 0. race. One caveat is in order: the of the c Minorities = whites = 0. significance 1, race variable in model 3 (and of the sex * p < .05 similarly, variable in model indicates **p < .01 2) only that an individ- ***p < .001 ual's race (or sex) tends to contribute to the individual's centrality. These results do not allow us to say that race (or sex) was used as a basis for friend- racial minorities tended to be more marginal than ship exclusion by majority group members more whites (p < .05). This same model shows that than it was by underrepresented group members. women were only marginally less central than men To examine the structural network positions of (p < .10), with major and race controlled for. whites and minorities in greater detail, we used The fourth hypothesis suggests that the tendency multidimensional scaling (MDS; Krackhardt, to make in-group (i.e., homophilous) friendship Blythe, & McGrath, 1994) on the unsymmetrized ties will be negatively related to the centrality of 159 X 159 friendship matrix. Figure 1 shows that underrepresented group members and positively the center of the network was occupied exclusively related to the centrality of majority group members. by whites, with a cluster of African Americans lo- The results of subsample analyses offered support cated in the upper right of the graph and other for this hypothesis. The subsample results pre- racial minority members located around the pe- sented in column 4 of Table 4 show that sex ho- riphery. The MDS analysis depicted in Figure 2 mophily (the tendency to choose friends of the shows just the friendship patterns among racial same sex) was positively associated with centrality minorities. African Americans (represented by for men. But the separate analysis for women pre- ovals surrounding "Bill") formed a relatively tight sented in column 5 showed no significant effect for friendship group, with many links between mem- sex homophily. In an analysis not reported in the bers. However, the members of other racial groups table, the positive correlation between sex ho- depended less on cohesive links among themselves mophily and centrality for the male subsample (r = than on the network-spanning activities of particu- .26, p < .01) was significantly higher (Z = 2.57, p < lar individuals. For example, the African American .05) than the negative correlation for the female "Fay" represented a link to the Hispanic commu- subsample (r = -.17, n.s.). nity, and the Hispanic "Jen" linked the African Similarly, the subsample regression results Americans, the Hispanics, and the Asian Ameri- shown in the last two columns of Table 4 show that cans. race homophily (the tendency to choose same-race was associated with friends) positively centrality DISCUSSION for whites, but marginally negatively associated with centrality for minorities. The correlation be- The results show consistent support for a distinc- tween race homophily and centrality for the white tiveness approach to the patterning of social net- 1998 Mehra, Kilduff, and Brass 447

TABLE 4 Summary of Regression Analyses Predicting Centrality in the Friendship Networka

Full Sample Subsamples

Independent Model Model Model Variable 1 2 3 Men Women Whites Minorities

Majorb 3.03* (1.39) 3.07* (1.40) 2.49t (1.37) 1.24 (1.84) 5.39** (1.97) 1.11 (1.47) 1.45 (2.68) Sex homophily 7.13t (3.89) 12.24** (4.68) -7.44 (6.97) Sexc 1.86' (1.07) 2.46* (1.08) Race homophily 15.03** (4.71) 32.77*** (5.79) -12.72t (6.39) Raced -3.33* (1.54) -5.83** (1.64) Model F 4.91** 4.44** 7.48** 4.07* 4.61** 18.50*** 2.43 R2 0.09 0.08 0.13 0.07 0.15 0.21 0.22

a Values represent unstandardized coefficients; standard errors are in parentheses. For the full sample, N = 159. For the subsamples, n's are as follows: men, 105; women, 54; whites, 139; minorities, 20. b Finance and marketing = 1, other majors = 0. c Men = 1, women = 0. d Minorities = 1, whites = 0. p < .10 * p < .05 ** p < .01 *** p < .001 works. The lower the relative proportion of group 1983); however, Segal (1962) presented contrary members in a social context, the higher the likeli- evidence. Although we can predict, from a distinc- hood of within-group identification and friendship. tiveness perspective, that token white males will Previous homophily research (e.g., Tuma & Halli- tend to identify and form friendships with each nan, 1979) has shown that people tend to interact other, we are less sure concerning the conditions with similar others. Our research refined this gen- under which they will experience marginalization. eral proposition by suggesting that perceptions of Further research in settings in which men and are based on distinctiveness within similarity spe- whites are in the minority rather than the majority cific contexts. Further, in extending distinctiveness is needed. from the realm of relations to the theory identity Our research raises the question of the boundary realm of relations, we have shown how friendship conditions of distinctiveness theory. Although this approach can help explain patterns of struc- women constituted 34 of the and tural in percent sample marginalization organizations. racial minorities 13 both From a distinctiveness it is only percent, groups theory perspective, tended to have rates of the of within a context relatively high within-group proportions people specific identification and Future as a work that are even if friendship. research, (such site) important, different of these are different from those of the using percentages underrepresented proportions can the at To the extent that ad- group members, help identify point surrounding society. people which to lose the distinctiveness that just their identities and in to groups begin friendships response enhances the social composition of each locale, the self can identity. The of racial members be considered a work-in-progress rather than a marginalization minority in the network to result both fixed entity (cf. Kondo, 1990). We have emphasized friendship appeared from and from the the potential flexibility of similarity judgments and exclusionary pressures preferences of the minorities for same-race friends. The behaviors, although we were not able to test these of women in ideas because our sample's proportions were simi- marginalization the friendship net- lar to those in the surrounding population. work appeared to result more from exclusionary People's identity and friendship choices may be pressures than from women's preferences for influenced not just by social context, but also by woman friends. Future research could help clarify such factors as the social status of different groups whether some demographic categories, such as in society. For example, the social status of white race, evoke stronger exclusionary pressures than men in society at large may mitigate any exclusion- other demographic categories, such as sex. Future ary pressures they face when present in token num- research could also help clarify whether the mar- bers in a specific context (Fairhurst & Snavely, ginalization of underrepresented groups is more 448 Academy of Management Journal August

FIGURE 1 c Friendship Relations among Individualsa' b

a Letters indicate race of individual: W = white, A = Asian American, B = African American, and H = Hispanic. b To preserve visual clarity, some whites near the center of the sociogram are not shown here. c The sociogram was drawn by Krackplot (Krackhardt, Blythe, & McGrath, 1994). affected by the relative proportions of these groups omy. By forming cohesive groups on the margins of than by the demographic categories themselves. social networks, minority students, such as the Although minorities were structurally marginal tightly knit African Americans, may sustain social in terms of the whole network, they had extensive solidarity while preserving autonomy of action. links among themselves. The African Americans, Thus, on issues of concern to racial minorities, the in particular, formed a tightly knit group on the members of this group may have been free to act as margins of the friendship network. This pattern a cohesive unit relatively unconstrained by binding suggests further research on the possibility that ties to those in the majority. marginality in informal networks may have advan- Further research on the possible costs and bene- tages as well as disadvantages. Previous work has fits of marginality may contribute to a more bal- emphasized the extent to which marginality re- anced understanding of informal social dynamics. duces access to important information (Burt, 1982), For example, preliminary work on the increasingly fuels dissatisfaction with work (Rice & Mitchell, popular minority network groups (such as Xerox's 1973), and generates feelings of isolation (Miller, black caucus) suggests that the positive aspects of 1975). However, from a resource dependence per- belonging to such groups may outweigh the nega- spective (Pfeffer & Salancik, 1978), centrality in a tive effects of symbolic separation (Friedman, network is often gained at the expense of auton- 1996). Members of underrepresented groups may 1998 Mehra, Kilduff, and Brass 449

FIGURE 2 b Friendship Relations among Minoritiesa

a African Americans' names are enclosed in ovals; Asian Americans', in rectangles; and Hispanics', in diamonds. Names are sex-specific. b The sociogram was drawn by Krackplot (Krackhardt, Blythe, & McGrath, 1994). compensate for apparent marginality in any partic- ilar national origin may be particularly evident ular organizational setting by establishing exten- among those groups (such as Korean Americans) sive cross-organizational ties (see the discussion in that include many recent immigrants with strong Thomas and Higgins [1996]). ties to a common homeland. Within the subsample of racial minorities, the Asian Americans appeared to be less cohesive as a Practical Implications group than the African Americans. Future research could examine identity and interaction differences In addition to providing a different lens through among minority groups in organizations. To what which to study patterns of structural marginaliza- extent, for example, do Asian Americans tend to tion, distinctiveness theory generates insights that subscribe to a common identity across different may help organizational policy makers. Many or- subgroups such as Korean American, Chinese ganizations struggle to reconcile pressures toward American, Vietnamese American, and Japanese demographic differentiation and integration. For American? There is some discussion in the ethnic example, in countries that recruit both men and studies literature concerning the emergence of women into the armed forces, policy makers must Asian American panethnicity (Espiritu, 1992), but decide whether or not to follow the example of the other research indicates a continuing preference U.S. Marines and train men and women recruits among some Asian Americans for social interaction separately. Universities try to preserve diversity on based on national origin (e.g., Oh & Kilduff, 1997). campus while striving to maintain an integrative This preference for interacting with people of sim- community. In practice, this means having to de- 450 Academy of Management Journal August cide when, for example, members belonging to re- the small number of minority group members in the ligious or racial groups will be allowed to live sample. Although the statistical results were signif- separately (Denby, 1997). A separate facility for icant despite the small number, further research is members of an underrepresented group may appear needed using samples in which different minority to majority group members as a self-imposed isola- groups are studied in diverse settings, especially tion that threatens to splinter the community. Our work organizations. Such research could also dis- research suggests that pressures toward within- tinguish the importance of visible social markers, group identification and friendship tend to be such as race and sex, from that of group underrep- stronger among members of underrepresented resentation. We cannot determine from the current groups than among members of majority groups. To study whether it was the social category or the achieve integration in organizations, it may be nec- proportion (in relation to the context) that drove essary to first recognize heightened pressures for the results. within-group solidarity among underrepresented group members. Previous research has shown that the lower hier- Conclusion archical rank of women and minorities in many To the extent that people belong to multiple organizations exacerbates the difficulties they en- have bases of on counter in themselves into informal net- groups, they multiple similarity integrating which to build bridges of social identification and works of influential others Ibarra, The (e.g., 1992). Simmel (1955: 125-195) discussed this a friendship. results we report from a sample lacking formal issue. Our demonstrates that the relative rar- that in informal net- study hierarchy suggest segregation ity of a in a social context is likely to promote works even in group may persist "delayered" organiza- members' use of that group as a basis for shared tional forms. Our results into the provide insights and social interaction. All people, at some social networks to in educational identity likely emerge in their careers, are to be M.B.A. point organizational likely organizations, specifically, competitive members of whether this in which a cohort of underrepresented groups, programs relatively large involves race, gender, working in a foreign country would-be executives are socialized together. Simi- as an or a cross-func- lar in for expatriate, simply joining patterns may emerge training programs tional team composed mainly of those with differ- cohorts of new recruits in work organizations. In ent expertise. From this perspective, organizations addition, to the extent that on people depend offer rich environments for identity development friendships formed in M.B.A. programs for job re- based on the shared characteristics individuals can ferrals and their careers, support throughout pat- discover. The discovery and promotion of shared terns established in these programs may have long- bases of identification be one of the most chal- effects. Our research raises may lasting practical lenging tasks of management. questions concerning whether network patterns formed in graduate or company training programs have lasting effects on interaction patterns in work REFERENCES settings. Ashforth, B. E., & Mael, F. 1989. Social identity theory and the organization.Academy of Management Re- Limitations view, 14: 20-39. A limitation this study shares with most net- Blau, P. M. 1977. Inequality and heterogeneity: A prim- itive social structure. New York: Free work research is the lack of multiple items for theory of Press. assessing the reliability of the identity and friendship measures. Recall that we gathered data Bonacich, P. 1972. Factoring and weighting approaches on these variables by using a roster method that to status scores and clique identification. Journal of Mathematical 2: 113-120. required individuals to look down an alphabeti- Sociology, cal list of fellow students and place checks next Borgatti, S. P., Everett, M. G., & Freeman, L. C. 1992. to the appropriate names. Repeating this proce- UCINET IV. Columbia, SC: Analytic Technologies. dure with multiple items would be a time-con- Brass, D. J. 1985. Men's and women's networks: A study suming and cognitively taxing process for re- of interaction patterns and influence in an organiza- spondents. Fortunately, research suggests that tion. Academy of Management Journal, 28: 327- single-item network measures are largely reliable 343. when the roster method is used to facilitate indi- Burt, R. S. 1992. Structural holes: The social structure viduals' recall (Marsden, 1990). of competition. Cambridge, MA: Harvard University Another limitation of the current study involves Press. 1998 Mehra, Kilduff, and Brass 451

Catalyst. 1996. The 1996 Catalyst census of the women discourses of identity in a Japanese workplace. board of directors of the Fortune 500. New York: Chicago: University of Chicago Press. Catalyst. Krackhardt, D. 1990. Assessing the political landscape: Cota, A. A., & Dion, K. L. 1986. Salience of gender and Structure, cognition, and power in organizations. sex composition of ad hoc groups: An experimental Administrative Science Quarterly, 35: 342-369. test of distinctiveness Journal theory. of Personality Krackhardt, D., Blythe, J., & McGrath, C. 1994. Krackplot and Social 50: 770-776. Psychology, 3.0: An improved network drawing program. Con- Denby, D. 1997. Passion at Yale: Does the constitution nections, 17(2): 53-55. care about coed dorms? New 22: Yorker, September Marsden, P. V. 1990. Network data and measurement. In 21-22. W. R. Scott & J. Blake (Eds.), Annual review of Ely, R. J. 1995. The power in demography: Women's sociology, vol. 16: 435-463. Palo Alto, CA: Annual social construction of gender identity at work. Acad- Reviews. 38: 589-634. emy of Management Journal, McGuire, W. J. 1984. Search for the self: Going beyond Espiritu, Y. L. 1992. Asian American panethnicity: self-esteem and the reactive self In R. A. Zucker, Bridging institutions and identities. Philadelphia: J. Aronoff, & A. I. Rabin (Eds.), Personality and the Temple University Press. prediction of behavior: 73-120. New York: Aca- demic Press. Fairhurst, G. T., & Snavely, K. 1983. A test of the social isolation of male tokens. Academy of Management McGuire, W. J., McGuire, C. V., Child, P., & Fujioka, T. Journal, 26: 353-361. 1978. Salience of ethnicity in the spontaneous self- as a function of one's ethnic distinctiveness Friedman, R. A. 1996. Defining the scope and logic of concept in the social environment. Journal minority and female network groups: Can separation of Personality and Social 36: 511-520. enhance integration? In G. R. Ferris (Ed.), Research Psychology, in personnel and human resource management, McGuire, W. J., McGuire, C. V., & Winton, W. 1979. vol. 14: 307-349. Greenwich, CT: JAI Press. Effects of house-hold sex composition on the salience of one's in the self-con- Gower, J. C., & Legendre, P. 1986. Metric and euclidean gender spontaneous Journal Social Psychology, properties of dissimilarity coefficients. Journal of cept. of Experimental 15: 77-90. Classification, 3: 5-48. Miller, 1975. Isolation in Alienation Granovetter, M. 1973. The strength of weak ties. Ameri- J. organizations: from control, and relations. can Journal of Sociology, 6: 1360-1380. authority, expressive Administrative Science Quarterly, 20: 260-271. Hughes, E. C. 1946. The knitting of racial groups in Oh, H., & Kilduff, M. 1997. in a land: industry. American Sociological Review, 11: 512- Strangers strange 515. Korean entrepreneurs and the Presbyterian church. Paper presented at the annual meeting of the Ibarra, H. 1992. and differential returns: Sex Homophily Academy of Management, Boston. differences in network structure and access in an Pfeffer, & Salancik, G. 1978. The external control advertising firm. Administrative Science Quarterly, J., of 37: 422-447. organizations. New York: Harper & Row. L. & T. R. 1973. Structural determi- Ibarra, H. 1993. Personal networks of women and minor- Rice, E., Mitchell, nants of individual behavior in Ad- ities in management: A conceptual framework. organizations. ministrative Science 18: 56-70. Academy of Management Review, 18: 56-87. Quarterly, B. E. 1962. Male nurses: A case in status Ibarra, H. 1995. Race, opportunity, and diversity of social Segal, study contradiction and loss. Social 41: circles in managerial networks. Academy of Man- prestige Forces, 31-38. agement Journal, 38: 673-703. Simmel, G. 1955. and the web Kanter, R. M. 1977a. Some effects of proportions on Conflict of group-affili- ations H. Wolff & R. Bendix, New York: group life: Skewed sex ratios and responses to token (K. trans.). Free Press. women. American Journal of Sociology, 82: 965- 990. Tajfel, H., & Turner, J. C. 1985. The social identity theory of behavior. In S. Worchel & W. G. Austin Kanter, R. M. 1977b. Men and women of the corpora- intergroup tion. New York: Basic Books. (Eds.), Psychology of intergroup relations (2d ed.): 7-24. Chicago: Nelson-Hall. Kelly, G. A. 1955. The psychology of personal con- structs. New York: Norton. Thomas, D., & Higgins, M. 1996. Mentoring and the boundaryless career: Lessons from the minority ex- Kilduff, M., & D. V. 1994. Do chameleons ahead? Day, get perience. In M. B. Arthur & D. M. Rousseau (Eds.), The effects of on careers. self-monitoring managerial Boundaryless career: A new employment principle Journal, 37: 1047-1060. Academy of Management for a new organizational era: 269-281. New York: Kondo, D. 1990. Crafting selves: Power, gender, and Oxford University Press. 452 Academy of Management Journal August

Tsui, A. S., Egan, T. D., & O'Reilly, C. A. 1992. Being Martin Kilduff (Ph.D., Cornell University) is an associate different: Relational demography and organizational professor of organizational behavior at The Pennsylvania attachment. Administrative Science Quarterly, 37: State University. Three current research projects focus on 549-579. how social distance affects perceptions of balance in Tuma, N. B., & Hallinan, M. T. 1979. The effects of sex, friendship networks, on how self-monitoring differen- race and achievement on schoolchildren's friend- tially affects the promotion of men and women to top ships. Social Forces, 57: 1286-1309. management, and on how individuals strategically man- age identity in organizations. Ajay Mehra is an assistant professor of management at the University of Cincinnati. He is currently completing his Daniel Brass is a doctorate in organizational behavior at The Pennsylvania J. professor of organizational behavior State University. Ongoing projects include an examination in the Smeal College of Business Administration at The of the effects of demographic and cognitive diversity on top Pennsylvania State University. He received his Ph.D. management team performance and an investigation into from the University of Illinois. His research interests the effects of social networks and personality on perfor- include the antecedents and consequences of social net- mance, promotions, and stress in a high-tech firm. works in organizations.