Interactions within Groups and Subgroups: The Effects of Demographic Faultlines

Author(s): Dora C. Lau and J. Keith Murnighan

Source: The Academy of Management Journal, Vol. 48, No. 4 (Aug., 2005), pp. 645-659
Published by: Academy of Management
Stable URL: [Link]
Accessed: 12-04-2020 07:10 UTC

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@[Link].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
[Link]

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

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
? Academy of Management Journal
2005, Vol. 48, No. 4, 645-659.

INTERACTIONS WITHIN GROUPS AND SUBGROUPS: THE

EFFECTS OF DEMOGRAPHIC FAULTLINES
DORA C. LAU
Chinese University of Hong Kong

J. KEITH MURNIGHAN
Northwestern University

We investigated the effects of intragroup and cross-subgroup communications in an

experimental field study on demographic faultlines. The results indicated that fault
lines explained more variance in perceptions of team learning, psychological safety,
satisfaction, and expected performance than single-attribute heterogeneity indexes. In
addition, cross-subgroup work communications were effective for groups with weak
faultlines but not for groups with strong faultlines. Overall, this study extends the
original faultline model, documents the utility of the concept of faultlines, and iden
tifies some of their effects on work group outcomes.

The increasing prevalence of diversity in the rather than one at a time. According to our model,
world's workforce presents the practice of manage if group members fall into two distinct, nonover
ment with both challenges and opportunities. A lapping subgroups based on demographic charac
long and active stream of research on the effects of teristics? e.g., young Hispanic women and old
diversity has documented diversity's positive and Caucasian men?a strong "faultline" is present.
negative effects. Findings have shown that, al These "hypothetical lines . . . may split a group into
though diversity has led to improved decision mak subgroups based on one or more attributes" (Lau &
ing and more creative problem solving (Watson, Murnighan, 1998: 328), and they provide an infor
Kumar, & Michaelsen, 1993; Pelled, Eisenhardt, & mal structure for intragroup conflict.
Xin, 1999), it has also reduced interpersonal liking, Early research on the faultline model has been
psychological commitment, intergroup communi supportive. Dyck and Starke's (1999) investigation
cation (e.g., Tsui & O'Reilly, 1989; Tsui, Egan, & of the process that led to the formation of break
O'Reilly, 1992), and group cohesion (Smith, Smith, away organizations used the faultline model as
Olian, Sims, O'Bannon, & Scully, 1994), and it has both a central framework and an explanatory mech
increased group conflict (Jehn, 1995). Recent re
anism. They found that internal group conflict was
search suggests that, over time, the negative effects
activated by outside events and augmented by im
of demographic diversity may diminish, allowing
portant new group members, leading to the forma
diversity's positive effects to be approached if not
tion of ideological rather than demographic fault
achieved (e.g., Earley & Mosakowski, 2000; Harri
lines in groups whose members knew each other
son, Price, Gavin, & Florey, 2002). The ability to
well. Earley and Mosakowski (2000) studied trans
accelerate these effects, to augment diversity's pos
national project teams and, in keeping with the
itive effects and reduce its negative effects, would
faultline model, found that moderate levels of het
be particularly welcome.
erogeneity, rather than low or high levels, led to
In previous work (Lau & Murnighan, 1998), we
continued the diversity literature's emphasis on poorer performance in three separate studies. Gib
intragroup processes by focusing on the effects of
son and Vermeulen (2003) found that teams
group member characteristics in combination learned most when they included moderately
strong subgroups and members who had similar
demographic characteristics. Groups with moder
ate homogeneity but no subgroups learned less,
A much earlier version of this paper was presented at
presumably owing to the barriers of insufficient
the 2001 Society for Industrial and Organizational Psy
chology Conference. We would like to thank the Chinese
interpersonal similarity; groups with very strong
University's Behavioral Interest Group, Kevin Au, Amy subgroups also learned less, as the faultline model
Edmondson, Kate Bezrukova, Bob Lount, Kathy Phillips, suggests. Finally, Thatcher, Jehn, and Zanutto
Niro Sivanathan, and three anonymous reviewers for (2003) created a measure of faultline strength and
their insightful comments and support. reported significant curvilinear effects: groups with
645

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 646 Academy of Management Journal August

moderately strong faultlines experienced less con conflict, scorn, and/or poor performance; with
flict and performed better than strong- or weak weak faultlines, communication should facilitate
faultline groups. performance.
Tests of the faultline model's predictions have Thus, following a long-held, implicit assumption
primarily depended on observational and correla that effective interactions are necessary for groups
tional data. Thus, an experiment that assessed cau to perform (cf. Cartwright & Zander, 1953), in the
sality would provide a much stronger test of the current research we focused directly on the fre
model. So would a comparison of the predictions of quency of work contacts among group members.
the faultline model with previous approaches to We assessed the effects of faultlines, diversity, and
demographic diversity. In addition, the model work-related communications on several group
could benefit from greater specification and more outcomes. We predicted that faultlines will have
pointed testing of its implications. This study ad their effects through work communications: an in
dressed all three of these goals by testing the fault
crease in work communications should have posi
line model's causal predictions, comparing its ex
tive effects for weak-faultline groups but not for
planatory power to that of previous approaches to
strong-faultline groups.
demographic diversity, and extending the initial
Faultlines can form around many attributes, in
model and testing those extensions. To accomplish
cluding demography, skills, personality, and val
these goals, we formulated a model of group fault
lines as moderators of the effects of communication ues. In our earlier research (Lau & Murnighan,
on group outcomes and investigated the impact of 1998), we focused primarily on demographic char
faultlines and communication on four outcome acteristics; according to our model, other, less im
variables (group members' expectations of their mediately obvious variables might also lead to
performance and perceptions of their group learn faultlines, but only after repeated interactions. Re
ing, psychological safety, and satisfaction) in a field search has supported both assumptions: people
experiment that contrasted groups that had either classify themselves quickly on the basis of interper
strong or weak faultlines. sonal characteristics (Fiske, 2000), but the impact
of demographic diversity fades over time (Harrison
GROUP FAULTLINES et al., 2002; Jehn, Northcraft, & Neale, 1999; Pelled
et al., 1999). Thus, because we studied new project
Our faultline model (Lau & Murnighan, 1998) teams, we focused on two highly visible demo
focuses on faultline strength in a group, the size graphic attributes, sex and ethnicity.1 Both have
and status of different subgroups, the effects of been significant predictors of group conflict (Jehn,
adding new members to groups, and the impetus of 1995; Pelled, 1996) and have indirectly led to poor
faultline-related issues. The original presentation group performance (Pelled et al., 1999).
of the model indicated that strong faultlines can A general question, however, is whether fault
have an impact on internal group conflict, politics, lines, which are predicted to have their greatest
and performance, but the exact nature of those re effect when a faultline-related issue arises, will
. lationships was not clearly specified. have an impact when the issues facing a group are
A simple model would include independent and not related to its faultlines. In the current research,
direct effects of faultlines on both conflict and per
the groups' task did not make sex or ethnicity sa
formance. Previous research, however, has shown
lient. Thus, conflict might remain dormant. Theo
that group conflict often has a negative influence
retically, then, this study was a conservative test of
on performance (e.g., Jehn, 1995) and that new
the faultline model: if faultlines had no impact
groups often begin working on a task before con
because the issues didn't bring them into play,
sidering their group process (Hackman & Morris,
1975). Thus, for many groups, conflict may arise the model's expectations would be realized; if
they did have an impact, their potency would be
early. These observations suggest that strong group accentuated.
faultlines will lead to conflict that then reduces
performance and that these effects depend on both
intragroup and cross-subgroup interactions. More
pointedly, following Gibson and Vermeulen's 1 In Jehn's (1995) study, and in Pelled's (1996) work,
(2003) suggestion that heterogeneity and subgroups groups were discriminated on the basis of race as they
can facilitate group learning, we suggest that the comprised African Americans or Caucasian Americans,
key underlying mechanism for these effects is and "race" is the typical cultural descriptor applied to
likely to be communication: with strong faultlines, these ethnic groups. The broader idea of ethnicity was a
communication between subgroups can generate more appropriate descriptor here.

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 2005 Lau and Murnighan 647

SOCIAL IDENTITY, FAULTLINES, AND oriented incentives led to stronger subgroup iden
LEARNING tity ("us versus them") and considerably reduced
individuals' motivations to contribute to a group as
The faultline model (Lau & Murnighan, 1998) a whole.
suggests that, when groups split into subgroups
In contrast, when subgroups are weak, an entire
(even implicitly), the members' group-related iden
group can more easily act as a single unit and learn
tities are associated more with their subgroups than
through "asking questions, seeking feedback, ex
with their entire groups. If this occurs, members of
perimenting, reflecting on results, and discussing
groups with strong faultlines (henceforth, strong
errors or unexpected outcomes of actions" (Ed
faultline groups) are likely to assume that their
mondson, 1999: 353). The exchange of work-re
scripts, values, and assumptions are similar to their
lated information can augment members' task un
fellow subgroup members' (cf. Phillips, Mannix,
Neale, & Gruenfeld, 2004). In contrast, members of derstandings and facilitate group task completion.
Within-group interactions can also facilitate the
groups with weak faultlines (weak-faultline
creation of a safe psychological environment for
groups) are likely to focus more on their entire
interpersonal risk taking (Edmondson, 1999). By
groups and assume that they are broadly similar to
sharing their beliefs, group members can build a
their fellow group members. This description sug
climate of mutual understanding and confidence. A
gests that individuals' group-related identities are
psychologically safe environment also allows
likely to have different sources in strong- and weak
members to exchange sensitive information, iden
faultline groups.
tify team members' mistakes, propose extreme so
Research on social identity theory has shown
lutions, and build trust (Mayer, Davis, & Schoor
that individuals who identify with a group favor
man, 1995) so that they can perform better and
their fellow group members more than individuals boost mutual satisfaction.
of other groups (Abrams, Wetherell, Cochrane,
The faultline model suggests, however, that
Hoggs, & Turner, 1990), and that such favoring
when faultlines are strong, group members are
enhances everyone's group status (Tajfel & Turner,
more likely to communicate and share information
1986). Thus, if individuals identify more with their
within rather than across their subgroups. This seg
subgroup than with the entire group, they are likely
regated communication and information sharing can
to evaluate their subgroup's members particularly
reduce the safety of a group's psychological environ
favorably. This combination of the logics of social
ment and its associated learning, performance, and
identity and group faultline theories suggested the
satisfaction. While information exchange within sub
following hypothesis:
groups can continue to contribute to group pro
Hypothesis 1. Members of strong-faultline cesses and performance, particularly if tasks are
groups evaluate members of their subgroups divisible and group members recognize both the
more favorably than do members of weak group's and the task's structure (e.g., Gibson & Ver
faultline groups. meulen, 2003), cross-subgroup interactions may in
terfere, especially when between-subgroup percep
This general hypothesis, without time demarca tions are negative. Then what might otherwise be
tions, represents part of the basic foundation of the perceived as a constructive critique, for instance,
faultline model. In this research, group members may be seen as an attack (Bartel, 2001). Threats
evaluated each other only toward the end of their from other subgroups can then have the same ef
project. Whether the initial group composition in fects as threats from an out-group (e.g., Sherif &
fluenced evaluations over a group's life span, then, Sherif, 1953), accentuating subgroup boundaries,
was another question addressed by this research. biases, and differentiation, and eroding a group's
Subgroup identification can accentuate mem processes and performance. Thus, when group
bers' awareness of a subgroup's boundary and their members use a subgroup lens to interpret their
feelings of belongingness. The organizational iden communications, doubts and misinterpretations
tity literature has suggested that members use their can thrive, limiting the effective exchange of sensi
identity as a source of self-definition, which helps tive or negative information and reducing the op
them perceive and interpret incoming information portunity for group learning. When psychological
and formulate possible actions (Ashforth & Mael, safety and group learning are adversely affected,
1989). Wit and Kerr's (2002) study of social dilem performance and satisfaction will also suffer. Thus,
mas showed that people focused on subgroups
rather than entire groups when outcomes depended Hypothesis 2. Groups with strong faultlines ex
more on their subgroups' actions than on their en perience more intragroup conflict and poorer
tire groups' actions. They also found that subgroup group outcomes (group learning, psychological

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 648 Academy of Management Journal August

safety, satisfaction and expected group perfor differed from the other subgroup's members on
mance) than do groups with weak faultlines. both demographic dimensions. Thus, strong-fault
line groups included subgroups of Caucasian
Hypothesis 3. Group faultlines moderate the women and non-Caucasian men or non-Caucasian
effects of cross-subgroup communications: for
women and Caucasian men. Each of the subgroups
groups with weak faultlines, more cross-sub
in the strong-faultline groups included at least two
group interactions improve group outcomes; and at most six members. Non-Caucasian sub
for groups with strong faultlines, they do not.
groups most often included individuals of Chinese
heritage, but they also included Koreans, East Indi
METHODS ans, and/or Persians. Analysis of strong-faultline
groups that included only Caucasians and Chinese
Sample versus Caucasians and a mix of non-Caucasian her
Ten undergraduate classes in an introductory or itages differed significantly on only one dependent
ganizational behavior course in a business school at variable, psychological safety, and this effect dis
a university in Canada participated in this study. appeared when we controlled for group size.
Data from ten teams were not included in the anal Groups that did not have strong faultlines had
yses because either less than half of their members more fragmented, more overlapping, and smaller
completed our questionnaire or their demographic subgroups. One group, for instance, included two
information was not complete. Exclusions resulted Caucasian and one non-Caucasian men and two
in a final sample of 504 individuals and 79 groups, non-Caucasian and one Caucasian women. Occa
with 4 to 8 members each (x = 6.4). Of these, 438 sionally, a weak-faultline group included three or
individuals (87%) completed our questionnaire. four people of the same ethnicity and the same sex.
Respondents included 246 women and 192 men; In each of these situations, at least one group mem
223 (51%) were Chinese, 149 (34%) were Cauca ber of the other sex was also of that same ethnicity.
sian, 38 were non-Chinese Asians, 18 were Persian/ (The exact composition of each of the groups is
Iranian, 3 were Aboriginals, and 7 were of mixed available from the authors on request.)
heritage. About half of the sample members (51%) Most of the students did not know each other
had been born in Canada; their average age was 21 before they took the course. We asked them about
years; 390 (89%) had previous work experience; their relationships prior to the class and computed
and 151 (35%) were working at the time of this a ratio of prior ties to group size for each group. A
study. comparison of the ratios for groups with strong and
weak faultlines yielded no significant differences
(F[l, 77] <1). Thus, although some groups may
Composing Groups
have included friends, the proportions of such
The student participants did not vary widely in groups were about the same in the strong-faultline
age or on other obvious demographic factors. Thus, and the weak-faultline groups.
within each class, we randomly assigned students
to groups based on their ethnicity and sex. Al Procedures and Measures
though the students were not randomly assigned to
their different class sections, their random assign The task, a normally required part of the course,
ment to our experimental conditions make this a asked participants to analyze the case of a real
"randomized-block" design. We categorized group estate company that suffered from many maladies,
members as Caucasian or non-Caucasian because including widespread inefficiencies and an ab
Caucasians are considered the dominant racial sence of trust among its employees. It provided an
group in North America. (This may be especially opportunity to apply a variety of analytical lenses
true in the province of Canada in which the study and make many recommendations. At the end of
was conducted.) the course, each group presented its conclusions in
Of the 79 groups, 34 had strong and 45 had weak writing and in class.
faultlines. Analyses indicated that neither the ratio After their presentations but before they had re
of men to women nor the ratio of Caucasian to ceived their grades, we invited the participants to
non-Caucasian members differed in the strong- and complete a questionnaire that included our depen
weak-faultline groups (F[l, 77] = 1.38 and F[l,77] dent and control variables, including the frequency
<1, respectively). of their intragroup and cross-subgroup work con
All of the strong-faultline groups included two tacts and their perceptions of group conflict, team
distinct subgroups whose members were identical learning, psychological safety, satisfaction, and
to one another in terms of ethnicity and sex but performance. Participants were given class time to

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 2005 Lau and Murnighan 649

complete the questionnaires. In addition, we of ert items; an example is, "We regularly take time to
fered two cash prizes (of $50 CD each; about $31 figure out ways to improve our team's processes."
US) that were awarded via lottery in each class, The psychological safety scale included seven
after completion of the questionnaire. Project coor 5-point Likert items; for example, "If you make a
dinators collected completed questionnaires; class mistake on this team, it is often held against you"
instructors did not know whether their students (reverse-coded). The alpha coefficients of these
had completed the questionnaire. Participation was scales, .63 and .71, respectively, were somewhat low.
voluntary, and students were aware that participa The "female faces scale" measured group satis
tion would have no effect on their course grade. faction (Dunham & Herman, 1975; Kunin, 1955).
The questionnaire asked participants to list the Respondents indicated whether they were satisfied
names of their group members and the frequency of with their group in general, their group's members,
work-related communications with each group and their task process using 11 female faces that
member during the final two weeks of the project, ranged from smiling to frowning [a = .88). We used
just prior to completing the questionnaire. Work Jehn's (1995) four-item measures of task conflict
communications included group meetings, sub [a = .80) and relationship conflict {a = .87); both
group meetings, and face-to-face, e-mail, or phone are assessed on five-point Likert scales. Finally,
contacts. Respondents chose among five coding expected group performance was measured via
categories: 1 for group members whom they con group members' ratings of their expected final
tacted less than once a week, and 2,3,4, and 5 for project grade (0-100). We attempted to collect the
1-5, 6-10, 11-15, and 16 or more contacts per instructors' final project grades, but several instruc
week, respectively. We calculated individual cross tors declined to release them. Because the data are
sex contacts by averaging the number of an in incomplete and because a preliminary analysis of
dividual's work contacts with members of the the standardized grades that we did collect led to
opposite sex; individual cross-ethnicity contact no reliable effects, we do not include them here.
measures were calculated the same way. Group Zero-order correlations indicated that instructor's
cross-sex contacts and group cross-ethnicity con evaluations were significantly related to the
tacts were calculated as the average of individual groups' ratings of their expected performance (r =
group members' cross-sex and cross-ethnicity work .40, p < .01, n = 59).
communications. To compare the utility of the traditional diversity
Participants also evaluated their group's mem approach with that of the faultline model, we in
bers on the Individual Process Scale (Barry & Stew cluded diversity indexes for sex and ethnicity as
art, 1997), which consists of ten 5-point Likert separate control variables. Among possible indexes
items on individual behaviors and contributions, that can be used for this purpose, the heterogeneity
including items on task, process, and affect. Task index is the most widely accepted and stable (Tsui
items measured members' task inputs; an example & Gutek, 1999). It is calculated as
is, "pulls his/her weight in the completion of group
tasks." Process items measured how much group
members facilitated the group process; an example
is, "stimulates the group to continue working when where P? is the proportion of the population in
discussion wanes." Affect items measured interper category i.
sonal affect; an example is, "is someone I trust."
Same-sex and same-ethnicity task, process, and af
fect evaluation measures were calculated by aver
Analyses
aging the ratings among men of other men in their Although individuals' responses were the basis
groups, women of other women, Caucasians of for our dependent variables, the items asked par
other Caucasians, and non-Caucasians of other ticipants to evaluate their group's actions, interac
non-Caucasians. tions, and outcomes?for example, "This team fre
Edmondson's (1999) team learning scale mea quently seeks new information that leads us to
sured participants' perceptions that their groups make important changes" (team learning); "Mem
learned through trial-and-error and continuous im bers of this team are able to bring up problems and
provement; we called our measure group learning. tough issues" (psychological safety); and "Please
We used Edmondson's (1999) psychological safety circle the face that best describes how you feel
scale to assess their beliefs that their teams had about your group in general" (group satisfaction).
created a secure environment for expressing their Because this was a randomized-block design, we
opinions, calling the measure group psychological initially analyzed whether the blocks, in this case
safety. The learning scale included six 5-point Lik different classes, were different from one another.

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
650 Academy of Management Journal August

TABLE 1
Means, Standard Deviations, and Correlations among Key Variables'1
Variables0 n" Mean s.d. 6

1. Faultline 438 0.42 0.49 .05 .04 .16 .26* .33**

2. Cross-sex contacts 437 2.55 0.81 .01 .95** .24* .16 .33**
-.05
3. Cross-ethnicity contacts0 400 2.49 0.76 .91** .26* .22+ .40**
' 4. Same-sex task ratings 432 4.17 0.61.06 .08 .09+ .62** .65**
5. Same-sex process ratings 432 3.55 0.71
.13** .58** .06 .09+ .52**
6. Same-sex affect ratings 433 4.21 0.70 .17** .55** .13** .17** .41**
7. Same-ethnicity task ratings 419 4.17 .05 .86** 0.59.12* .07 .46** .50**
8. Same-ethnicity process ratings 420 3.54 .16** 0.69 .06 .45** .05 .83** .39**
.17**
9. Same-ethnicity affect ratings 421 4.20 0.70 .17** .46** .15** .34** .84**
10. Group learning 438 3.64 0.57 .04 .16** .36** .17** .35** .37**
.47**
11. Group psychological safety 438 4.02.09+ 0.62 .14** .36** .14** .31**
12. Group satisfaction 438 8.43 1.79 .14** .15** .36** .13** .40** .51**
.07
13. Expected group performance 438 79.82 6.23 .13** .18** .13** .21** .20**
14. Task conflict 438 2.71 0.71 -.10* .06 .10* .01 .08 .15**
15. Relationship conflict -.16** 438 2.03 .07 .28** 0.81
.05 .15** .41**
16. Sex heterogeneity index 438 1.56 0.50.17**
.06 .04 .04 .13** .09f
17. Ethnicity heterogeneity index -.01 438.09+0.34 .050.47.02 .04 .02

a Group-level correlations are in the upper-right triangle, and individual-level correlations are in the lower-left triangle Cronbach alpha
coefficients are listed in italics on the diagonal.
b Variable codings: 1 == "strong faultline"; 0 = "weak faultline"; 1 = "male"; 2 = "female"; 0 = "non-Caucasian"; 1 = "Caucasian."
c Group correlation coefficients relating to this variable were based on 72, instead of 79, groups.
d Number of individual respondents.
+ p < .10
* p < .05
**p < .01
Two-tailed tests.

To partition the variance between individuals, individual-level predictors because social identity
groups, and classes, we used hierarchical linear theory suggests that groups with superior or subordi
modeling (3) analyses (HLM3; Bryk & Raudenbush, nate identities may react differently to resource allo
1992; Hoffman, 1997), which estimate within cation issues (Tajfel & Turner, 1986). After "partialing
group effects (level 1) and, on the basis of the in out" sex and ethnicity effects at the individual level,
tercepts and the slopes at the individual level, es the results showed that between-group variances
timate the separate effects of group-level predictors were significant for all six evaluation measures
on the intercepts and slopes (level 2), and the in v^OO sex-task ~" *^^? ^00 sex-process ~~ "> ^OO sex-affect ~~
tercepts at the group (here, class) level (level 3). .1U, T0o ethnicity-task ~~ ,^"> ^OO ethnicity-process ~~ '^t
Only intercepts were estimated at the class level Too ethnicity-affect = H> all p's < .01). We used the
because we did find any significant differences for intercept-as-outcomes model to test the hypothesis.
any of the group process and outcome variables Thus, we checked for the presence of systematic be
among the students' different classes. tween-group variance and whether heterogeneity or
In addition, intraclass correlation coefficients faultlines based on group-level sex or ethnicity pre
(ICC[l]s) estimated the proportion of variance in dictors affected within-subgroup evaluations (Byrk &
the outcome variables between groups over the sum Raudenbush, 1992). To investigate whether faultlines
of between- and within-group variance. Significant explained more than traditional approaches to diver
F-tests indicated that aggregations were justified sity, our analyses began by including only sex or
(Klein & Kozlowski, 2000). ethnic heterogeneity. We then assessed whether the
Hypothesis 1 states that individuals in groups addition of faultlines explained additional variance.
with strong faultlines evaluate their subgroup Hypotheses 2 and 3 state that faultlines are pos
members more favorably than do individuals in itively associated with team learning, psychologi
groups with weaker faultlines. Before analyzing the cal safety, group satisfaction, and expected perfor
effects of faultlines on within-subgroup evalua mance. The same two-step test determined whether
tions, we conducted a random-coefficients regres faultlines had incremental utility. Results indi
sion model to evaluate between-group variances. Re cated that the sex and ethnic heterogeneity indexes
spondents' sex and ethnicity were entered as were not related to our dependent variables. To

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 2005 Lau and Murnighan 651

TABLE 1
Continued
7 8 9 10 11 12 13 14 15 16 17

.13 .28* .30** .09 .19 .24* .13 -.15 -.25* .05 .50**
.24* .11 .33** .46** .32** .32** .36** -.01 -.09 .01 -.16
.21+ .13 .35** .47** .32** .32** .39** -.03 -.11 .14 -.20+
.90** .58** .64** .50** .50** .55** .39** -.13 -.30** .19 -.00
.58** .92** .50** .53** .41** .53** .41** -.02 -.10 .23* .09
.55** .48** .90** .50** .59** .65** .44** -.35** -.53** .30** -.04
.62** .61** .47** .44** .49** .35** -.10 -.25* .08 .05
.55** .55** .52** .39** .52** .36** -.06 -.11 .19+ .16
.57** .46** .51** .60** .67** .42** -.36** -.54** .22+ .01
.33** .34** .35** .63 .51** .66** .55** -.10 -.26* .11 -.03
.36** .31** .48** .47** .71 .70** .45** -.48** -.70** .07 -.02
.37** .39** .52** .57** .60** .88 .67** -.50** -.65** .11 -.02
.20** .22** .24** .36** .27** .47** -.22? -.29** .07 -.13
-.10* .08+ .16** -.09+ -.28** -.36** -.11* .80 .77** -.06 .07
-.29** .17** -.40** -.29** -.58** -.58** -.22** .59** .87 -.02 .02
.15** .11* .05 .04 -.01 -.02 -.13** -.02 -.02 .06
.08 .15** .03 -.05 .12* -.06 -.01 .03 -.04 -.06

maximize power, we did not include RESULTS them in subse

quent analyses. After partialing out the individual
Table 1 presents the correlation matrix of the
effects of sex and ethnicity, we obtained results that
primary variables. Cross-sex and cross-ethnicity
showed that the between-group variances were all
work contacts, task, process, and affect ratings, and
Significant (r00 task conflict = '14> T00 relationship conflict =
group process variables were often significantly
??> ^00 group learning ~~ '^' '00 aspsychological
correlated, safety
was expected for task groups. Other
. U O, Tqo group satisfaction ' ' ' '00 expected performance
correlations were infrequently significant.
= 8.90, all p's < .01, except for group learning
[p < .05]).2 We then examined the effects of fault
lines after controlling for group
Preliminary Analyses size in the HLM
analyses. To check the success of the faultline treatment,
Centering is an important method when the ab
we used Thatcher et al.'s (2003) index of faultline
solute zero of a variable is meaningless and when
strength, fau, which measures the percentage of
testing moderating models. Because cross-sub
total variation in overall group attributes accounted
group work contacts ranged from one to five, zeroes
for by a group's strongest possible faultline. To
were meaningless. In addition, we grand-mean obtain the index, one calculates the ratio of the
centered group faultline strength to avoid multicol
between-group sum of squares to the total sum of
linearity between the "main effect" and interaction
variables. squares. Values on this index range from 0 to 1 and
increase with faultline strength. The average fault
line strengths of the weak- and strong-faultline
groups in this study were .71 and .95, respectively;
2 Note that the between-group variances (t00) are met the difference between them was statistically sig
ric-sensitive. Variables that were measured via scales nificant (F [1, 77] = 78.31, p < .001). This result
with wider ranges, such as expected performance, which indicates that the experimental manipulation of
ranged from 0 to 100, and satisfaction, which ranged faultlines was successful. In addition, following
from 1 to 11, had larger variances. procedures recommended by McGraw and Wong

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 652 Academy of Management Journal August

(1996), we found the same pattern of results by garding the task, the data can be aggregated at the
comparing ICC(l)s in the groups as a whole and in group level to explain the between-group variance
the strong- and weak-faultline subgroups. One-way of the outcome variables.
analyses of variance (ANOVAs) on ICC(l)s were Overall, the data indicate that participants con
significantly higher for process and affect evalua tacted each other about their projects about 10
tions for the same-sex and same-ethnicity members times per week. (The average rating was 2.59; a
of subgroups in the strong-faultline groups than the rating of 2 indicated 6-10 contacts per week; a
ICC(l)s in the groups as a whole or the ICC(l)s for rating of 3 indicated 11-15.) Counter to our expec
the weak-faultline groups. (The effects for task eval tations, reports of the frequencies of work commu
uations were consistently not significant.) Thus, nications of strong- and weak-faultline groups did
strong-faultline subgroup members agreed signifi not differ; neither did cross-sex and cross-ethnicity
cantly more on affect and process issues than group subgroups. This suggests that the work itself com
members overall and than members of subgroups pelled group members' interactions.
with weak faultlines, as predicted.
We also assessed whether analyzing the depen
Subgroup Evaluations
dent variables at the group level was justified. For
within-subgroup evaluations, we calculated ICC(l)s We first assessed the influence of the traditional,
by including all group members who evaluated col one-variable-at-a-time approach to diversity before
leagues of the same sex or ethnicity. With one ex assessing the effects of group faultlines. Group
ception (contact frequency among same-sex mem level analyses of the sex heterogeneity index led to
bers regarding the task), the ICC(l)s were all significant, positive effects for the same-sex process
significant (sex-task = .04, F [1, 77] =1.21, p = .13; and affect ratings (see the first row of Table 2) but
sex-process = .07, F [1, 77] = 1.43, p < .05; sex not for the same-sex task ratings. The ethnicity
affect = .14, F [1, 77] = 1.94, p < .01; ethnicity heterogeneity index was negatively related to same
task = .09, F [1, 70] = 1.50, p < .01; ethnicity ethnicity affect evaluations but was not signifi
process = .09, F [1, 70] = 1.52, p < .01; ethnicity cantly related to any of the other same-ethnicity
affect = .15, F [1, 70] = 1.94, p < .01). The ICC(l)s evaluation ratings (the second row of Table 2).
for group outcomes were also significant (group Faultlines had many effects on individuals' eval
learning = .12, F [1, 77] = 1.78; psychological uations of their subgroup members, even after en
safety = .14, F [1, 77] = 1.92; group satisfaction = tering the heterogeneity indexes. The HLM results
.23, F [1, 77] = 2.63; expected performance = .23, F indicate that the members of strong-faultline
[1, 77] = 2.64; task conflict = .29, F [1, 77] = 3.34; groups evaluated their same-sex colleagues more
relationship conflict = .34, F [1, 77] = 3.91, all favorably in terms of group process and affect (row
p's < .01). These findings indicate that, other than 3, columns 2 and 3, in Table 2) than members of
contact frequency among same-sex members re weak-faultline groups. These same effects were also

TABLE 2
Results of Hierarchical Linear Modeling for Within-Subgroup Evaluations41
Same- Same- Same
Same-Sex, Same-Sex, Same-Sex, Ethni
Variable Task Process Affect Task Process Affect

Sex heterogeneity index 0.72 (0.44) 1.33** (0.49) 1.96** (0.32)

Ethnicity heterogeneity -0.20(0.30) -0.16 (0.38) -0.86* (0.38)
index
Faultline 0.08 (0.06) 0.19** (0.07) 0.24** (0.05) 0.10 (0.10) 0.28** (0.10) 0.36** (0.12)

R2h .15 .36 .33 .04 .47 .28

Ai?2c .05 .31 .20 .04 .41 .28

a HLM3 was used. Significance tests are one-tailed for the hypothesized faultline effects, two
parentheses. The degrees of freedom in estimating variance components of same-sex rating models
i & 2 = 67. For the same-ethnicity rating models, they were: d/level 3 = 9 and d/levels 1 & 2 = 60.
b Between-group s proportion of intercept variance explained.
c Incremental proportion of intercept variance explained by faultlines.
* p < .05
** p < .01

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
2005 Lau and Murnighan 653

TABLE 3
Results of Hierarchical Linear Modeling for the Direct Effects of Faultlines3

Expected
Group-Level Task Relationship Group Group Psychological Group
Predictors Conflict Conflict Performance Learning Safety Satisfaction

Faultline (701) -0.14 (0.12) -0.27* (0.13) 0.80 (0.76) 0.04 (0.06) 0.10* (0.05) 0.49* (0.25)
Group size (y02) 0.02 (0.05) 0.03 (0.07) -0.19 (0.60) -0.03 (0.04) -0.07 (0.04) -0.18 (0.13)

R2h .05 .08 .02 .04 .04 .11

AR2c .04 .07 .02 .01 .04 .08

a HLM3 was used. Standard errors are in parentheses. Significance tests are one-tailed for the hypothesized faultline effects, two-tailed
otherwise. The degrees of freedom in estimating variance components were as follows: d/level 3 = 9 and d/levels t & 2 = 67.
b Between-groups proportion of intercept variance explained.
c Incremental proportion of intercept variance explained by faultlines.
* p < .05
** p < .01

significant for same-ethnicity colleagues (row 3, Group Outcomes

columns 5 and 6). No significant effects were found
for task-related issues. These results indicate that The second set of dependent variables included
group members' ratings of their task and relation
strong faultlines led to more interpersonal connec
ship conflict, and their group's learning, psycho
tions within subgroups, rather than within entire logical safety, satisfaction, and expected perfor
groups, supporting the faultline model's predic mance. We tested and compared direct effect and
tions. They also explained considerably more vari moderator models (see Tables 3,4a, and 4b, respec
ance than the heterogeneity indexes on their own tively). Faultlines had a significant main effect for
(row 5 of the table). Finally, neither faultlines nor ratings of relationship conflict, psychological
heterogeneity had strong effects on task-related safety, and group Satisfaction. Contrary to predic
evaluations. Thus, for this sample and this task, tions based on the model, however, these effects
the impact of faultlines on intragroup evaluations indicate that the members of strong-faultline
was primarily on the team members' social groups experienced less relationship conflict, more
connections. psychological safety, and more satisfaction with

TABLE 4a
Sex-Related Results of Hierarchical Linear Modeling for the Moderating Effects of Faultlines

Expected
Group Psychological Group Group
Group-Level Predictors Learning Safety Satisfaction Performance

Faultline (y01) 0.04 (0.05) 0.11+ (0.06) 0.51* (0.23) 0.77 (0.60)
Cross-sex work contacts (y02) 0.27** (0.05) 0.17** (0.04) 0.62** (0.17) 2.47** (0.69)
Faultline X cross-sex work -0.11+ (0.08) -0.24* (0.11) -0.71* (0.37) -3.12* (0.87)
contacts (y03)

i?2b .52 .23 .29 .28

Ai?2c .01 .05 .04 .06

aHLM3 was used. Standard errors are in parentheses. Significance tests are one-tailed for the hypothesized interaction effects,
two-tailed otherwise. The degrees of freedom in estimating variance components were as follows: d/level 3 = 9 and d/levels 1 & 2 = 66.
b Between-groups proportion of intercept variance explained.
c Incremental proportion of intercept variance explained by the interaction term.
+ p < .10
* p < .05
**p < .01

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 654 Academy of Management Journal August

TABLE 4b
Ethnicity-Related Results of Hierarchical Linear Modeling for the Moderating Effects of Faultlines

Expected
Group Psychological Group Group
Group-Level Predictors Learning Safety Satisfaction Performance

Faultline (y01) 0.05 (0.12) 0.13+ (0.07) 0.60* (0.25) 1.12+ (0.64)
Cross-ethnicity work contacts (y04) 0.28** (0.07) 0.17** (0.06) 0.67** (0.22) 2.68** (0.87)
Faultline X cross-ethnicity work contacts (y05) -0.12 (0.12) -0.24* (0.12) -0.44+ (0.29) -2.87* (1.31)

R2b .52 .13 .20 .24

A/?2c .07 .01 .03

a HLM3 was used. Standard errors are in parentheses. Significance tests are one-tailed for the hypothesized interaction effects,
two-tailed otherwise. The degrees of freedom in estimating variance components were as follows: d/level 3 = 9 and d/levels a & 2 = 59.
b Between-group s proportion of intercept variance explained.
c Incremental proportion of intercept variance explained by the interaction term.
+ p< .10
* p < .05
** p < .01

their groups than the members of weak-faultline tively and significantly correlated for the weak
groups. These unexpected findings may have re faultline groups, but none of the eight correlations
sulted from generalizations of the positive social were significant for the strong-faultline groups.
effects within strong-faultline subgroups to the en Fisher's Z-transformation test further confirmed
tire groups, reflecting the importance of proximal that seven of the eight correlations for the weak
relationships; or they may have resulted from faultline groups were significantly larger than the
other, unmeasured, influences. Other direct effects corresponding correlations for the strong-faultline
were not significant. Thus, Hypothesis 2 was not groups (Cohen & Cohen, 1983). Thus, more work
supported. contacts had a variety of positive effects for weak
Hypothesis 3 predicts that faultlines moderate faultline but not for strong-faultline groups. This
the relationship between cross-subgroup communi finding provides strong support for Hypothesis 3.
cations and group outcomes (see Tables 4a and 4b).
Analyses of the main effects of cross-sex (y02) and DISCUSSION
cross-ethnicity (y04) work communications on in
dividuals' perceptions were significantly positive The members of these groups had many things in
for all four dependent variables [p < .01 for each common that could have limited the effects of their
variable), indicating that work communications demographic faultlines. They were all students at
were highly valued. the same university, enrolled in the same course,
The HLM results for the moderating effects of and about the same age. Yet their groups' faultlines
faultlines reveal that all of the interaction terms had a strong impact on their evaluations and per
(yo3 and y05) were negative (see Tables 4a and 4b), as ceptions of group outcomes, even though the task
anticipated. The effects were statistically signifi had nothing to do with sex and ethnicity. This
cant for cross-sex and cross-ethnicity communica result provides considerable support for the origi
tions for psychological safety and expected perfor nal faultline model.
mance; they were also significant for cross-sex The evaluations of the members of the strong
communications for group satisfaction. Combined faultline groups were biased toward individuals
with the positive main effects, these findings indi within their own subgroups in terms of group pro
cate that communication and group outcomes were cess and affect, suggesting that the members of
positively related for the weak-faultline groups but strong-faultline groups identified socially not with
were not significantly related for the strong-fault the group as a whole but with their demographi
line groups. In addition, the zero-order correlations cally similar subgroups. Affect-based biases are
(which may be influenced by problems of indepen easily understandable in groups with strong fault
dence) make these effects very clear (see Table 5): lines because similar people are attracted to each
for each of the four outcome variables, cross-sex other. Process-based biases are also easily under
and cross-ethnicity communications were posi stood, because members may connect more with

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
2005 Lau and Murnighan 655

TABLE 5
Zero-Order Correlations between Cross-Ethnicity and Cross-Sex Contacts
for the Strong- and Weak-Faultline Groups0

Group Psychological Group Expected Group

Variables Learning Safety Satisfaction Performance

Cross-sex work contacts

Strong-faultline groups 34 0.29 0.09 0.08 0.05
Weak-faultline groups 45 0.56** 0.44** 0.46** 0.56**
Fisher's Z 1.39* 1.59* 1.74* 2.43**

Cross-ethnicity work contacts

Strong-faultline groups 28 0.26 0.08 0.15 0.07
Weak-faultline groups 44 0.58** 0.44** 0.42** 0.56**
Fisher's Z 1.40* 1.54* 1.17 2.22**

a The reported Z's are the results of tests for the difference between corresponding correlations.
* p < .05
**p < .01
One-tailed tests.

their subgroup colleagues as they attempt to coor The Faultline Model

dinate their efforts. Because task-based evaluations
focus on members' work inputs, it may be difficult These findings represent the first experimental
or impossible to ignore the contributions of the support for our faultline model (Lau & Murnighan,
other subgroup's members. Whether process and 1998). The results also indicate that faultlines may
affect might ultimately affect actual as well as per have direct and potentially pervasive effects on
ceived performance remains for future research to group processes and outcomes, even without the
determine. trigger of a faultline-relevant issue. This last result
Our most striking finding showed that work com was unexpected theoretically but suggests that
munications were particularly effective for weak faultlines have considerable impact. For instance,
faultline groups but led to no appreciable value for the current findings suggest that, when a work
strong-faultline groups. On each of our four out group's tasks do activate their faultlines, faultlines'
come measures, for both cross-ethnicity and cross effects may be even stronger than what we observed
sex communications, the data revealed exactly the here. The current context might be considered
same pattern (Tables 4a, 4b, and 5). These results "minimal faultline stimulation"; maximal stimula
suggest that a long history of findings on group tion might lead to many deleterious effects.
effectiveness (e.g., McGrath, 1984) needs to be re The analyses also showed that faultlines had
vised: with increased diversity in the workforce, more explanatory power than single-issue hetero
particularly in the United States, intragroup com geneity indexes, in this case for ethnicity and sex.
munications may no longer be generally or uni In addition, faultlines' effects surfaced in a rela
formly helpful, but instead, their effectiveness may tively short time, in one-time project groups. This
depend on whether groups have strong faultlines. finding suggests that the categorical information
These findings may also extend Edmondson's that is embodied in individuals' demographic char
(1999) contention that team structure contributes to acteristics was sufficient to generate faultline ef
team beliefs. In the current context, the presence or fects relatively quickly. Thus, these findings con
absence of strong faultlines created team structures tinue to suggest that demographic characteristics
that influenced the locus of group members' social are potent forces in social contexts.
identifications, which may have contributed to This research also documented some unexpected
their communications and the effectiveness of their consequences of group faultlines. First, contrary to
interactions (Hewstone, Rubin, & Willis, 2002). Ed the model-based prediction, the members of groups
mondson's logic suggests that faultlines' effects on with stronger faultlines were more positive about
group learning and performance may also flow several perceived group outcomes than the mem
through group beliefs. Thus, future studies might bers of groups with weaker faultlines. Second, the
investigate the interplay between faultlines, identi fact that cross-subgroup work communication was
ties, beliefs, and communication affecting impor not effective for strong-faultline groups contradicts
tant group behaviors and outcomes. Allport's (1954) hypothesis, which was supported

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 656 Academy of Management Journal August

by Amir (1969), that contacts between demograph Groups with strong faultlines and subgroups that
ically different individuals will reduce negative vary in size may have strikingly different group
stereotypes, enhance mutual understanding, and dynamics than groups whose subgroups are equal
promote cooperative relationships. It also does not in size (Lau & Murnighan, 1998). Thus, power plays
support an assimilation model (Moghaddam & Sol may surface more in some groups than others, in a
liday, 1991), according to which cross-subgroup systematically predictable fashion. Although
interactions will help to break down subgroup groups and subgroups of different sizes were insuf
boundaries and facilitate group processes and out ficient in number for statistical analysis in this
comes. Instead, it reinforces Sherif and Sherif's research, future research might fruitfully address
(1953) conclusion, that it takes more than contact these dynamics as well.
for groups, and in this case subgroups, to connect We manipulated group composition by creating
positively. Because expectations of positive perfor strong, clear faultlines in new project teams. De
mance increased with communication only for pending on the stage of a group's development,
the weak-faultline groups, strong-faultline groups faultlines in other teams may be based on a multi
might require truly common goals (e.g., defeating tude of potential attributes. Contextual influences,
common enemies) or decidedly integrative tasks to such as organizational culture, the nature of group
overcome their divisive subgroup structures (Aron tasks, or the numerical distinctiveness of a select
son, Blaney, Sikes, Stephan, & Snapp, 1975). attribute (Kanter, 1977) might contribute to the sa
lience of some rather than other attributes. Thus,
Limitations and Future Research further research might also investigate the self-cat
egorization dynamics of different attributes, such
The use of undergraduate project groups and low as occupations and age, to accurately assess fault
financial rewards may limit the generalizability of lines' effects. Advances in the development of for
these results. Because undergraduates have limited mal measures of the theoretical components of the
work experience, their intragroup behaviors and faultline model, such as faultline strength (Shaw,
perceptions may not represent those of individuals 2004; Thatcher, Jehn, & Zanutto, 2003) and fault
in work organizations. The minimal rewards may line distance (Bezrukova, Jehn, & Zanutto, 2005),
have also reduced their motivation. Finally, their may also help to broaden the systematic investiga
youth and relative immaturity may have limited tion of faultlines and their effects.
their ability to handle interpersonal conflict
effectively.
The findings of linear and interaction faultline
Practical Implications
effects differ from the finding of curvilinear effects
in other research (e.g., Earley & Mosakowski, 2000; Because we have conceptualized faultlines as a
Gibson & Vermeulen, 2003). One possible reason is basic structural element in work groups, these find
the somewhat narrow range of faultline strength in ings may have their primary practical impact on the
this study. The undergraduate students were simi design of work groups. Clearly, leaders should be
lar to each other on many dimensions, thus limiting well advised to carefully consider their groups'
the potential range of faultline strength. Thus, this tasks, communication structures, and members' in
study's linear effects may represent small segments dividual characteristics. For instance, managers
of a bigger curvilinear model. who naturally encourage their group members to
These were also relatively short-term rather than work closely with each other might actually be
long-term work groups. Group members might have making a mistake if their groups have strong fault
been affected by knowing that they would not be lines?even if their faultlines and their group task
working together for long. Thus, they may have are not related. Managers may need to limit cross
been more tolerant of their differences, limiting subgroup communication and select tasks that can
overt conflict. They may also have invested less take advantage of groups' within-subgroup inclina
time in getting to know each other, making the tions. Thus, divisible tasks might actually be well
social information that came immediately from de suited for strong-faultline groups, especially if the
mographic attributes have longer effects, accentu groups' final products can be realized by leaders
ating the impact of faultlines (Meyerson, Weick, & who can combine the contributions of their dispar
Kramer, 1996). Although many of these effects are ate subgroups by themselves. This idea supports
also observed in work teams (e.g., Thompson, the dual categorization theory of Hornsey and Hogg
2000), the current results might best be viewed as (2000), who proposed that both group and sub
an early snapshot of the effects of faultlines on group boundaries need to be respected and that the
project teams. assimilation of subgroup members to the larger

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 2005 Lau and Murnighan 657

group may create identity threats and a nonharmo ing who you are: Self-categorization and the nature
nious group atmosphere. of norm formation, conformity, and group polariza
With the popularity of cross-functional and tion. British Journal of Social Psychology, 29: 97
119.
cross-departmental teams, occupational segrega
tion, and the eagerness to correct asymmetric de Allport, G. W. 1954. The nature of prejudice. Reading,
mographic distributions, it may not be uncommon MA: Addison-Wesley.
for teams to have strong faultlines that align sex, Amir, Y. 1969. Contact hypothesis in ethnic relations.
race, ethnicity, occupations and/or tenure. Team Psychological Bulletin, 71: 319-342.
composition should clearly include attention to Aronson, E., Blaney, N., Sikes, J., Stephan, C, & Snapp,
weakening faultlines, an approach that would help M. 1975. The jigsaw route to learning and liking.
take advantage of diversity and blunt the negative Psychology Today, February: 43-50.
impact of potential faultlines. Weak-faultline
Ashforth, B. E., & Mael, F. E. 1989. Social identity theory
groups may also benefit from interpersonal attrac and the organization. Academy of Management Be
tion among themselves (cf. Gibson & Vermeulen, view, 14: 20-39.
2002). But the faultline model suggests that, even
Barry, B., & Stewart, G. L. 1997. Composition, process,
though demographic faultlines may influence a and performance in self-managed groups: The role of
group's early development, demographics may be personality. Journal of Applied Psychology, 82: 62
less influential later, when faultlines based on 78.
other attributes?such as religious or political pref
Bartel, C. A. 2001. Social comparisons in boundary-span
erence or work philosophy, to name just three at
ning work: Effects of community outreach on mem
tributes that might polarize work groups?may bers' organizational identity and identification. Ad
emerge. Thus, the current data and the faultline ministrative Science Quarterly, 46: 379-413.
model suggest that team leaders must be particu
Bezrukova, K., Jehn, K. A., Zanutto, E, & Thatcher, S.
larly attuned to the faultline dynamics of their team
2005. Do workgroup faultlines help or hurt? The
members and design both task and informal com
role of distance and shared identity on individual
munication networks to match their current and
outcomes. Working paper, Rutgers University.
potentially constantly shifting faultline dynamics.
Bryk, A. S., & Raudenbush, S. W. 1992. Hierarchical
When managers activate both superordinate and
linear models: applications and data analysis
subgroup identities, or when they must manage methods. Newbury Park, CA: Sage.
groups that have had those identities simultaneously
Cartwright, D., & Zander, A. 1968. Group dynamics (3rd
activated, other issues may also be pertinent. When
ed.). New York: Harper & Row.
goal completion is crucial, externally induced effects,
such as pressing deadlines, common threats, or com Cohen, J., & Cohen, P. 1983. Applied multiple regres
petition, may actually be necessary to raise group sion/correlation analysis for behavioral sciences.
Second edition. Hillsdale, NJ: Erlbaum.
members' awareness to the level of their entire group.
This idea again suggests that the conclusions from Dunham, R., & Herman, J. 1975. Development of a female
Sherif and Sherif s (1953) classic Robbers' Cave ex faces scale for measuring job satisfaction. Journal of
periments should continue to garner serious consid Applied Psychology, 60: 629-631.
eration. Their studies documented the seemingly nat Dyck, B., & Starke, F. A. 1999. The formation of break
ural competitive tendencies that emerged in groups away organizations: Observations and a process
and the need to create integrative tasks to overcome model. Administrative Science Quarterly, 44: 792
822.
quickly surfacing antipathies. As faultlines become
stronger, subgroup versus subgroup tension may also Earley, P. C, & Mosakowski, E. 2000. Creating hybrid
be relevant within work groups, especially if these team cultures: An empirical test of transnational
competitive tendencies escalate rapidly (as they did team functioning. Academy of Management Jour
in the Sherif and Sherif experiments). The develop nal, 43: 26-49.
ment of an "us versus them" mentality within a group Edmondson, A. 1999. Psychological safety and learning
can make the idea of an effective, cohesive team in behavior in work teams. Administrative Science
creasingly unattainable. Faultlines may be a clear and Quarterly, 44: 350-383.
simple way to diagnose the likelihood of such poten Fiske, S. T. 2000. Stereotyping, prejudice, and discrimi
tially divisive events. nation at the seam between the centuries: Evolution,
culture, mind, and brain. European Journal of So
cial Psychology, 30: 299-322.
REFERENCES
Gibson, C, & Vermeulen, F. 2003. A healthy divide:
Abrams, D., Wetherell, M., Cochrane, S., Hogg, M. A., & subgroups as a stimulus for team learning behavior.
Turner, J. C. 1990. Knowing what to think by know Administrative Science Quarterly, 48: 202-239.

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 658 Academy of Management Journal August

Hackman, J. R., & Morris, C. G. 1975. Group tasks, group Moghaddam, F. M., & Solliday, E. A. 1991. "Balanced
interaction process, and group performance effec multiculturalism" and the challenge of peaceful co
tiveness: A review and proposed integration. In L. existence in pluralistic societies. Psychology and
Berkowitz (Ed.), Advances in experimental social Developing Societies, 3(1): 51-71.
psychology, vol 8: 45-99. New York: Academic Pelled, L. H. 1996. Relational demography and percep
Press.
tions of group conflict and performance: A field in
Harrison, D. A., Price, K. H., Gavin, J. H., & Florey, A. T. vestigation. International Journal of Conflict Man
2002. Time, teams, and task performance: Changing agement, 7: 230-246.
effects of surface- and deep-level diversity on group Pelled, L. H, Eisenhardt, K. M., & Xin, K. R. 1999. Ex
functioning. Academy of Management Journal, 45: ploring the black box: An analysis of work group
1029-1045.
diversity, conflict, and performance. Administrative
Hewstone, M., Rubin, M., & Willis, H. 2002. Intergroup Science Quarterly, 44: 1-28.
bias. In J. T. Spence, J. M. Darley, & J. Foss (Eds.), Phillips, K. W., Mannix, E. A., Neale, M. A., & Gruenfeld,
Annual review of psychology, vol. 53: 575-604. D. H. 2004. Diverse groups and information sharing:
Palo Alto, CA: Annual Reviews.
The effects of congruent ties. Journal of Experimen
Hoffman, D. A. 1997. An overview of the logic and ra tal Social Psychology, 40: 497-510.
tionale of hierarchical linear models. Journal of Shaw, J. B. 2004. The development and analysis of a
Management, 23: 723-744. measure of group faultlines. Organizational Be
Hornsey, M. J., & Hogg, M. A. 2000. Assimilation and search Methods, 7: 66-100.
diversity: An integrative model of subgroup rela Sherif, M., & Sherif, C. W. 1953. Groups in harmony and
tions. Personality and Social Psychology Beview, tension. New York: Harper & Row.
4: 143-156.
Smith, K. G, Smith, K. A., Olian, J. D., Sims, H. P.,
Jehn, K. A. 1995. A multimethod examination of the
O'Bannon, D. P., & Scully, J. A. 1994. Top manage
benefits and detriments of intragroup conflict. Ad ment team demography and process: The role of
ministrative Science Quarterly, 40: 256-282. social integration and communication. Administra
Jehn, K. A., Northcraft, G. B, & Neale, M. A. 1999. Why tive Science Quarterly, 39: 412-438.
differences make a difference: A field study of diver
Tajfel, H, & Turner, J. C. 1986. The social identity theory
sity, conflict and performance in workgroups. Ad of intergroup behavior. In S. Worchel & W. G. Austin
ministrative Science Quarterly, 40: 256-282. (Eds.), Psychology of intergroup relations: 7-24.
Kanter, R. M. 1977. Some effects of proportions on group Chicago: Nelson-Hall Publishers.
life: Skewed sex ratios and responses to token Thatcher, S. M. B., Jehn, K. A., & Zanutto, E. 2003. Cracks
women. American Journal of Sociology, 82: 965 in diversity research: The effects of diversity fault
990.
lines on conflict and performance. Group Decision
Klein, K. J., & Kozlowski, S. W. 2000. From micro to and Negotiation, 12: 217-241.
meso: Critical steps in conceptualizing and conduct Thompson, L. 2000. Making the team: A guide for man
ing multilevel research. Organizational Besearch agers. Upper Saddle River, NJ: Prentice- Hall.
Methods, 3: 211-236.
Tsui, A. S., Egan, T. D., & O'Reilly, C. A. 1992. Being
Kunin, T. 1955. The construction of a new type of atti
different: Relational demography and organizational
tude measure. Personnel Psychology, 8: 65-78. attachment. Administrative Science Quarterly, 37:
Lau, D. C, & Murnighan, J. K. 1998. Demographic diver 549-579.
sity and faultlines: The compositional dynamics of Tsui, A. S., & Gutek, B. A. 1999. Demographic differ
organizational groups. Academy of Management ences in organizations: Current research and fu
Beview, 23: 325-340.
ture directions. Lanham, MD: Lexington Books
Mayer, R. C, Davis, J. H, & Schoorman, F. D. 1995. An Tsui, A. S., & O'Reilly, C. A. 1989. Beyond simple demo
integrative model of organizational trust. Academy graphic effects: The importance of relational demog
of Management Beview, 20: 709-734. raphy in superior-subordinate dyads. Academy of
McGrath, J. E. 1984. Groups: Interaction and perfor Management Journal, 32: 402-423.
mance. Englewood Cliffs, NJ: Prentice-Hall. Watson, W. E., Kumar, K, & Michaelsen, L. K. 1993.
McGraw, K. O., & Wong, S. P. 1996. Forming inferences Cultural diversity's impact on interaction processes
about some intraclass correlation coefficients. Psy and performance: Comparing homogeneous and di
chological Methods, 1: 30-46. verse task groups. Academy of Management Jour
nal, 36: 590-602.
Meyerson, D., Weick, C. K., & Kramer, R. M. 1996. Swift
trust and temporary groups. In R. M. Kramer & T. R. Wit, A. P., & Kerr, N. L. 2002. "Me versus just us versus
Tyler (Eds.), Trust in organizations: Frontiers of us all" categorization and cooperation in nested so
theory and research: 166-195. Thousand Oaks, CA: cial dilemmas. Journal of Personality and Social
Sage. Psychology, 83: 616-637.

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]
 2005 Lau and Murnighan 659

-??\

Dora C. Lau (doralau@[Link]) is an as is the Harold H. Hines Jr. Distinguished Professor of Risk
sistant professor at the Chinese University of Hong Kong. Management at the Kellogg School of Management at
She received her doctoral degree from the University of Northwestern University. He received his Ph.D. in social
British Columbia in Canada. Her research interests include psychology from Purdue University. His research focuses
demographic diversity, team dynamics, relational trust, on negotiation, decision making, groups, and strategic
chain store management, and Chinese management. interpersonal interaction, and currently addresses altru
ism, auctions, cooperation, ethics, fairness, faultlines,
J. Keith Murnighan (keithm@[Link]) norms, penance, power, self-interest, and trust.

-/&

This content downloaded from [Link] on Sun, 12 Apr 2020 [Link] UTC
All use subject to [Link]