Está en la página 1de 15

Organizational Behavior and Human Decision Processes 99 (2006) 1–15

www.elsevier.com/locate/obhdp

Decision performance and diversity structure: Comparing faultlines


in convergent, crosscut, and racially homogeneous groups 夽
John E. Sawyer a,¤, Melissa A. Houlette b,1, Erin L. Yeagley c,1
a
Department of Business Administration, Alfred Lerner College of Business and Economics, 236 Alfred Lerner Hall,
University of Delaware, Newark, DE 19716, USA
b
College of Mount St. Joseph, USA
c
University of Delaware, USA

Received 25 July 2003


Available online 28 September 2005

Abstract

Group diversity structure (the composition of racial and job-function diversity) and pre-discussion decision eVects on group deci-
sion accuracy were tested in three-person groups. Evidence supported the social categorization model and the notion of multiple
faultlines (i.e., subgroup boundaries). Crosscut diversity structure, where racial and job-function subgroup boundaries are crossed,
weakened faultlines, enhanced information sharing and improved decision-making. Our data also supported the common knowledge
eVect. Groups in which members made pre-discussion choices arrived at incorrect decisions consistent with majority members’ pre-
discussion preferences, based on a biased subset of information. Crosscut groups in which members did not make pre-discussion
choices performed the best. Analyses of video taped interactions and attitudes toward the group help to explain these diVerences. We
discuss the implications for managing demographically diverse groups and for future research on the impact of various attributes of
diversity in groups.
 2005 Elsevier Inc. All rights reserved.

Keywords: Group decision; Social categorization; Faultlines; Diversity structure; Common knowledge eVect; Attentional focus

In modern businesses, groups are frequently com- group piecing together a more accurate and complete
posed of members with diverse backgrounds and from information base from which to make decisions (Stasser
multiple business functions. Members meet to share and & Titus, 1985). Theoretically this interaction results in
discuss information relevant to the problem at hand. In higher quality decisions than any individual would
principle, such discussion is believed to accomplish a achieve working alone. Having people from multiple
corrective function whereby the biased and incomplete business functions and backgrounds work together in
information of individuals is compensated for by the groups is expected to provide a broader range of per-
spectives (i.e., knowledge, skills, information, and prob-

This research was supported by a CERT grant from the College of
lem solving approaches) than would come from
Business and Economics, University of Delaware. Thanks to three homogeneous groups of individuals.
anonymous reviewers, David A. Harrison and Deborah C. Andrews However, prior research on diversity in groups has
for their suggestions and assistance developing the manuscript. shown mixed results. A meta-analysis of empirical studies
*
Corresponding author. Fax: +1 302 831 4196. of diversity found no consistent eVect of diversity (job
E-mail address: sawyerj@lerner.udel.edu (J.E. Sawyer).
1
Melissa Houlette and Erin Yeagley were both doctoral students in
related, gender or race) on either performance or group
the Psychology Department at the University of Delaware during data cohesion (Webber & Donahue, 2001). This result did not
collection. diVer between top management versus lower level teams.

0749-5978/$ - see front matter  2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.obhdp.2005.08.006
2 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

Webber and Donahue oVer a number of possible explana- Group level diversity translates into individual attributes
tions for their Wndings. One possibility is the inconsistent that are task-related or relations-oriented (Jackson,
Wndings may occur because researchers ignore diVerent May, & Whitney, 1995). Task-related diversity may
eVects of diVerent types of diversity (e.g., surface versus occur when members of diVerent job-functions bring
deep level) and the interaction of time with types of diver- diVerent information to bear on a problem. Resulting
sity (Harrison, Price, & Bell, 1998). Furthermore, diVerent information diVerences (Jehn, Northcraft, & Neale,
attributes of diversity may aVect performance only to the 1999) provide diVerent perspectives (Williams &
extent that they aVect social integration in groups (Harri- O’Reilly, 1998), but may result in conXicting viewpoints
son, Price, Gavin, & Florey, 2002). and opinions pertaining to a group task (Jehn & Man-
Diversity activates social categorization (Kramer, nix, 2001). While researchers expected task-related and
1991) in which individuals identify themselves as insiders relations-oriented diversity to have diVerential eVects on
or outsiders (Milliken & Martins, 1996). As a result, performance and group cohesion, a recent meta-analysis
groups may not adequately integrate information across across 24 diVerent studies found no consistent eVects of
job-function or racial subgroups. Self-categorization on diversity (Webber & Donahue, 2001).
salient diVerences like race and job-function leads to Time may moderate the eVects of surface level versus
boundaries that seriously inXuence group processes, limit deep level diversity. Surface level attributes such as race
communication, and diminish group cohesion (Lau & and gender are immediately evident (generally immuta-
Murninghan, 1998). Lau and Murninghan refer to these ble) characteristics (Jackson, 1992; Jehn et al., 1999;
ingroup versus outsider boundaries as “faultlines.” In the Maznevski, 1994; Pelled et al., 1999). Deep level diversity
current study, we focus on two types of faultlines; those relates to those attributes of the individual such as atti-
deWned by demographic diversity and those deWned by tudes, beliefs and values that take time to emerge from
informational diVerences imbedded in diVerent job-func- interpersonal interaction. Harrison et al. (1998) found
tions. We refer to the relationship between these attributes that time neutralized the negative eVects of surface level
of diversity as a group’s Diversity Structure. diversity on workgroup cohesion as group members got
Given the inconsistent impact of diversity across to know each other. In contrast, the recognition of deep
studies, the unanswered question is; under what circum- level diversity emerges over time as people learn about
stances does diversity in teams matter? Our study others attitudes and beliefs. Thus, while deep level diver-
focuses on when and how diversity eVects might be oper- sity had no eVect on cohesion in the short term, for those
ating with the goal of learning how to leverage the pre- teams with greater experience together, deep level diver-
dicted, but elusive eVects of diversity. We postulate that sity (variance in job satisfaction and organizational
multiple attributes of diversity manifest themselves in commitment) reduced group cohesion over time.
diVerent conWgurations leading to various diversity Lack of consistent Wndings suggests that the eVect of
structures. We argue that diversity structure, not amount diversity is dependent on the social processes groups
of diversity, is a key driver of the eVect of diversity. We adopt. Diversity may have positive impact only to the
examine the eVects of diVerent combinations of job- extent that groups engage in debate (Simons, Pelled, &
function and demographic diversity on group decision- Smith, 1999), and social integration may determine group
making eVectiveness by varying group composition and performance and cohesion (Harrison et al., 2002). Thus, to
information distribution. We test group level perfor- understand the eVects of diversity we must develop and
mance and the cross-level eVects on behavior and per- understanding of the how diversity eVects social integra-
ceptions to help us understand why diversity structure tion and the interaction processes that allow groups to
diVerences aVect group performance. We also examine assimilate information to make eVective decisions.
the eVect of pre-discussion decisions among group mem-
bers to allow us to provide comprehensive recommenda- Diversity structure: The conWguration of faultlines
tions for improving decision outcomes in diverse groups.
Social identity theory (Tajfel & Turner, 1986) and self-
categorization theory (Turner, 1982) are the predominate
Group level eVects of diversity structure and prior choices frameworks for understanding the eVects of diversity on
social integration. Self-categorization theory suggests that
While management practitioners often tout the individuals use various attributes of others to determine to
advantages of a diverse workforce, research has sug- what groups they belong. Surface level characteristics such
gested that diversity is a double edged sword, bringing as race and gender are immediately evident (generally
beneWts and limitations. In an attempt to understand the immutable) characteristics that lead to identiWcation of
eVects of diversity, researchers have diVerentiated job- subgroups (Jackson, 1992; Jehn et al., 1999; Maznevski,
related diversity such as attitudes, viewpoints, and 1994; Pelled et al., 1999). While deep level diversity (Harri-
knowledge from less job-related diversity such as race, son et al., 1998) often takes time to emerge through group
gender and age (e.g., Pelled, Eisenhardt, & Xin, 1999). interaction, visual or other sensory cues can sensitize
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 3

members to diVerences and similarities quickly (Hertel & ninghan (1998) developed the faultline analogy. They
Kerr, 2001). Typically, in a business context group mem- suggested that subgroups based on salient attributes of
bers know of each other’s job-function aYliations both group members form faultlines. Individual attributes
through labeling and through surface manifestations such such as race, gender, job roles, and other salient diVer-
as dress or mannerisms. Informational diVerences may act ences lead to self-categorization and thus form multiple
as a deep level diversity that can emerge over time, or can homogeneous subgroups. In an otherwise homogeneous
be cued by known job-function diVerences (e.g., “I’m from group, job-function diVerences will result in a single
the marketing department,” or sales people dress diVer- faultline because the members of one job-function see
ently than manufacturing people) so that job related themselves as a separate subgroup from others in the
information diVerences act more like surface level diver- group. This single subgroup faultline is illustrated in the
sity. Job-function distinctions (e.g., marketing versus pro- middle image of Fig. 1, where two members of a single
duction) can lead to social categorization within an job-function within a racially homogeneous group see
organization (Kramer, 1991). Social Identity Theory sug- themselves as diVerent and separate from a third mem-
gests that any salient characteristic can lead to subgroup ber who represents a diVerent job-function.
identity in which members see themselves as insiders to The alignment of multiple subgroups determines the
their subgroup, and view others as outsiders. Such sub- strength of the faultline separating a subgroup from the
group boundaries generate beliefs and attitudes about rest of the group. When the faultlines of racial and job-
insiders versus outsiders that can form a barrier to com- function subgroups converge, such that members of the
munication and collaboration. job-function subgroup are also members of a racial sub-
Similarly, when a single individual, or a numerically group, the subgroup distinction is sharpened. Multiple
small subgroup of members hold diVerent information converging subgroups (Lau & Murninghan, 1998)
they may become minority opinion holders. Minority opin- strengthen the faultline, further blocking social and infor-
ions can increase group creativity in problem solving and mation integration across the subgroup boundary. The
improve the quality of solutions (Nemeth & Wachtler, convergence of race and job-function subgroups is illus-
1983). However, majority opinion holders are more likely trated in the bottom right image of Fig. 1. The faultline
to accept and integrate minority opinions when the minor- analogy is consistent with the social categorization model
ity opinion holder is a member of the same social category (Brewer & Miller, 1984), which predicts that convergence
as majority (i.e., Clark & Maass, 1988; Mugny & Papasta- of race and job-function will promote category-based
mou, 1982). Those within one’s own social group will be responding because race and job-function cues coincide.
more inXuential (Brock, 1965; Goethals & Nelson, 1973; The convergence of racial with job-function sub-groups
Lau & Murninghan, 1998) because people respond better may be a limiting condition on the potential beneWts of
to messages that come from within their subgroup than job-function (task-related) diversity (Pelled et al., 1999;
from an outsider (Turner, 1982; Van Knippenberg & Simons et al., 1999) because it impedes social integration.
Wilke, 1992; Wiegman, 1985; Wilder, 1990). In contrast, crosscutting race with diVerent job-func-
Thus, the amount of diversity matters less than how tions undercuts the salience of racial categorization by
that diversity is conWgured in the group. Lau and Mur- changing the patterns of who is “in” and who is “out”

C
Cross-cut

A B C
Same Race

C
A B

A B

• Fill pattern indicates different job functions resulting in Convergent


unique information.
• Background color indicates race.
• The three partitions of each triangle indicate three
group members, persons A, B, & C.
Fig. 1. Diversity structure: Distribution of information across racial subgroups.
4 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

and promotes individual as opposed to subgroup-based ment is high, if cross-group alignment is also high fault-
interaction (Marcus-Newhall, Miller, Holtz, & Brewer, lines are weak. Shaw (2004) validated his approach by
1993). When race crosses job-function, some racial sub- demonstrating that the calculations of cross subgroup
group members will align with the job-function identiW- alignment were consistent with Lau and Murninghan’s
cation of members outside their racial subgroup. The top prototypical groups. He also showed that the measures
left image of Fig. 1 illustrates this crosscutting of race had convergent and discriminant validity in two sam-
and job-function subgroups. Crosscutting also results in ples. Shaw cautioned that his study did not test the
members within the same racial subgroup representing eVects of diversity on group process or performance.
diVerent job-functions, thus holding diVerent informa- Our conception of diversity structure is consistent
tion from each other, while having job-function in com- with Shaw’s conceptualization of faultline strength as
mon with members outside their racial subgroup. Since the extent of internal versus cross subgroup alignment.
people expect to agree with similarly categorized others, The convergent subgroups described above and illus-
discrepant views from in-group members violate expec- trated in Fig. 1 have strong faultlines because there are
tations and are likely to induce active processing of two attributes of alignment (shared race and job-func-
information (Baker & Petty, 1994; Petty & Cacioppo, tion) within the subgroup. When job-function diversity
1986). converges with racial diversity (i.e., racial subgroup
There has been little empirical research directly test- members have similar information), the job-function
ing Lau and Murninghan’s (1998) faultline concept. faultline reinforces the racial faultline thus strengthening
Gibson and Vermeulen (2003) measured team heteroge- the subgroup distinction, reducing cohesion with mem-
neity as the overlap of attributes across all pairs of team bers outside the racial subgroup, and greatly impeding
members summed across the team. They deWned sub- communication (O’Reilly, Caldwell, & Barnett, 1989). In
group strength as the standard deviation of the overlap this conceptualization, the same race group has a moder-
across members. Thus, if some group member pairs had ate faultline because there is a subgroup identiWed by the
many traits in common while others had few traits in internal alignment of job-function, but there is no salient
common, subgroup strength was high indicating there cross subgroup alignment. In contrast, the crosscut
was at least one subgroup whose members shared many group, thou it has more diversity than the same race
traits. Gibson and Vermeulen (2003) found an inverted group (race and job-function versus merely job-func-
U relationship between subgroup strength and team tion) has a weaker faultline. When members of a sub-
learning behavior, with team learning occurring only group also share salient attributes with other group
when subgroup strength was moderate. In their formula- members outside their subgroup, there is cross-subgroup
tion, subgroup strength is moderate when members of a alignment of attributes. This cross subgroup alignment
subgroup also share attributes in common with group reduces the subgroup identiWcation and thus the fault-
members outside their subgroup. line strength. Thus, social integration is more likely to
Thatcherr, Jehn, and Zanutto (2003) used a similar occur (Marcus-Newhall et al., 1993). The crosscut struc-
calculation of faultline strength. They tested the eVect of ture with its weakened faultlines should allow members
faultline strength on conXict (relationship, task, and pro- to share and integrate information better than groups
cess) and group outcomes (performance and morale). with clear-cut boundaries, thus arriving at better deci-
Their analysis did not support the hypothesis that fault- sions more often than same race groups.
line strength is inversely related to performance and task We hasten to point out that this is a more Wne-grained
conXict. Their own reanalysis indicated that groups with approach, and leads to a diVerent hypothesis than the
medium faultlines in which members had overlapping minority inXuence literature. The minority inXuence lit-
membership across multiple faultlines had higher levels erature would suggest that same race groups would per-
of performance and morale. Thus, the critical factor in form better because the minority-opinion holder shares
enhancing performance and morale is cross-subgroup race with both other members (Phillips, 2003) while in
alignment. crosscut groups the minority-opinion holder shares race
Shaw (2004) also presented a method of determining with only one member of the majority-opinion group.
faultline strength. He measured internal alignment by Following Shaw’s (2004) measurement of faultline
determining the extent to which members of a subgroup strength, convergent groups will have the strongest faultline
are more similar than chance distribution with other because both race and job-function deWne a subgroup.
members on all other possible attributes. Summing Same race groups have a subgroup deWned by only job-
across attribute categories determines the extent of inter- function and thus their faultline will be weaker than con-
nal subgroup alignment. Cross-subgroup alignment is vergent groups. Crosscut groups have the weakest faultline
the extent to which members of a subgroup share attri- because one member of the racial subgroup aligns with the
butes with others outside their subgroup. If internal job function attribute of a member outside the racial sub-
alignment is high and cross-group alignment is low, the groups, and one member of the job-function subgroup
faultlines are strong. In contrast, even if internal align- aligns racially with a member outside the job-function
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 5

subgroup. Thus, we predict that crosscut groups will arrive will aVect all of those members’ pre-discussion judgments
at the correct decision more frequently, and convergent and thus will have a larger inXuence on the group judg-
groups less frequently, than racially homogeneous groups. ment than information that a minority of the members
hold. When a subgroup of the members hold the same
Hypothesis 1. Groups in which attributes crosscut sub information, it is likely that they will arrive at the same pre-
group boundaries will achieve greater decision accuracy discussion preference, assuming they weight the informa-
than same race groups and groups with convergent sub- tion similarly. The informational subgroup with the major-
group boundaries will under perform same race groups. ity of the members will be able to pressure the member(s)
holding the minority opinion to adopt its preference
Prior choices and the common knowledge eVect (Davis, 1973). Thus, shared information should have more
inXuence on the group judgment even if unshared informa-
Several researchers have studied information sharing tion is discussed fully. When the knowledge held by the
and group judgment. A number of scholars have used informational outsider is needed to realize the best option,
hidden proWle tasks to investigate the information-sam- informational subgroup boundaries make it less likely that
pling model (Stasser, 1988). Hidden proWles occur when the minority information is considered and the group will
commonly held information supports an inferior deci- likely adopt the inferior majority choice. In contrast, when
sion alternative, thus the use of unique information held there is no pre-discussion decision, members consider more
by an individual or a small subset of the group is information. With no pre-discussion decisions, the atten-
required to uncover the best choice. The information- tional focus is taken oV the pre-discussion preferences and
sampling model proposes that the probability of an item more attention is given to exploring the information, thus
of information entering group discussion is a function of leading to a more optimal decision.
the number of individuals in the group who hold that
Hypothesis 2. When members do not state pre-discussion
item. As the pool of shared items is depleted, more
preferences groups will make the best decision more
unshared items are discussed (Larson, Foster-Fishman,
often than when members make a pre-discussion choice.
& Keys, 1994). However, other processes as noted in the
comprehensive review by Stasser and Dietz-Uhler (2002) We expect that pre-discussion preferences will act in
aVect the likelihood that information will enter group an additive fashion with diversity structure. That is, by
discussion and thus inXuence decisions. stating a pre-discussion preference, group members are
Research on conformity (Deutsch & Gerard, 1955) and likely to focus on individual decisions rather than
jury decision-making (Davis et al., 1993) has shown that exploring information. Given that pre-discussion prefer-
once individuals have made a public commitment, they ences are related to job-function informational diVer-
resist changing their judgment even under social pressure. ences, and each of our conditions includes a job-function
In a decision task, the act of forming and formally stating faultline, all groups stating pre-discussion preferences
a judgment may bias individuals to seek information to will under perform those groups not stating pre-discus-
conWrm that judgment and contribute only information sion preferences. Thus, we do not expect an interaction
that supports that judgment. The attentional focus model among these factors.
(Kelly & Karau, 1999) suggests that initial preferences
serve as a Wlter through which new information is consid-
ered. Thus, member preferences may inXuence the content Cross-level eVects of group structure on individuals
of group discussion such that members adopt a hypothe-
sis-conWrming search strategy where individual attention Group interaction processes
and group discussion focuses on information supportive
of pre-discussion preferences and ignores or undervalues To reach the best decision, the job-function outsider
contradictory information. must participate in the group discussion and members of
However, even if the information is considered it may the job-function subgroup must pay attention to him or
have less impact on the group’s decision than the mem- her. However, interaction problems associated with the
bers’ pre-discussion preferences. Common knowledge diversity structure of the group often leads to poor atti-
has a greater impact on group judgment than unshared tudes and unproductive behaviors (Maznevski, 1994).
information because it can serve as a common reference Because token, or skewed distribution of minority
point for all group members. Gigone and Hastie (1993) membership in groups has negative eVects on the token
found that group members use information they hold member (Kanter, 1977) we compare two critical posi-
prior to group discussion to make individual judgments. tions within the group. In all three types of diversity
Individual judgments have a greater inXuence on group structure, there is always an information outsider. In the
judgments than the original information. convergent condition the information outsider is not a
One implication of the common knowledge eVect is that member of either the racial or job function subgroup. In
information held by multiple members prior to discussion the crosscut condition the information outsider is a
6 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

member of the racial majority subgroup. In the same Hypothesis 4. The job-function (information) outsider will
race condition the information outsider diVers from be attended to more in crosscut groups than in same race
other group members only by job-function (see position groups and less in convergent diversity structure groups.
C in each image in Fig. 1). Comparing the crosscut and
convergent conditions the racial outsider is a member of Group member perceptions of the group
the job-function subgroup in the crosscut but not the
convergent condition (see position A in Fig. 1). Compar- Members’ identiWcation with the group may be an
ing across conditions, one person is always both a racial important factor in group decision-making since strong
and information insider (see position B in each image of group attachment has been associated with a high level
Fig. 1). Focusing on position C allow us to make infer- of activity (Kirchmeyer & Cohen, 1992). Informational
ence about the impact of diVerent diversity structures on diVerences within the group may result in the informa-
the information outsider. Comparing position A and C tion outsider not identifying with the group. When diver-
allows us to determine if this eVect is due to being a sity structure is convergent, the informational outsider is
racial minority, versus being an information outsider. further isolated due to both job-function and racial
The faultline analogy (Lau & Murninghan, 1998) sug- diVerences. In contrast, the crosscut diversity structure
gests that the stronger the faultline the less participation weakens the faultline boundaries. In the absence of
will cross the faultline. In convergent groups, the infor- strong faultlines the informational outsiders are more
mation outsider is constrained by both the uniqueness of likely to see themselves as belonging to the group. In the
his or her job-function and the racial boundary. Thus, same race groups, though all members are of the same
the boundary to participation is stronger in convergent race, the job-function faultline is salient resulting in less
groups than in same race groups where only the job- identiWcation with the whole group.
function faultline impedes participation. In crosscut
Hypothesis 5. The information outsider will identify with
groups, the crossing of the racial and job-function fault-
the group more in crosscut groups than same race groups
lines will reduce subgroup categorization opening
and less in groups with convergent subgroup boundaries.
greater participation by the information outsider than in
the same race condition. Diversity leads to higher performance when members
are able to combine and build on each other’s ideas.
Hypothesis 3. The job-function (information) outsider
Studies have shown that groups high on both diversity
will participate more in crosscut diversity-structure
and integration have positive attitudes and improved
groups than in same-race groups and less in groups with
information-processing (Harrison et al., 2002; Maznev-
convergent subgroup boundaries.
ski, 1994). Since crosscutting weakens faultlines and
For Hypothesis 3, we compare the participation of should therefore promote integration of information
the job-function (information) outsider with the job- across job-function and racial diversity, information
function majority member who is a racial minority in the outsiders in the crosscut condition should perceive
crosscut and a racial majority in the convergent condi- greater opportunity to inXuence the group.
tion (person A in each image of Fig. 1). We did this to
Hypothesis 6. Information outsiders will perceive their
control for variability in discussion time across groups
ability to inXuence the group more in crosscut groups
and to assess if race versus information minority is the
than same race groups and less in convergent groups.
determining factor of this eVect. We propose that infor-
mation minority is more important than race in deter- Following the same logic, the information outsider
mining level of participation. will perceive greater group process eVectiveness in cross-
Majority group members will pay attention to similar cut diversity structure and less eVectiveness in the con-
others and allow their information to inXuence the vergent condition than among same race groups.
group’s decisions more than information coming from
dissimilar persons (Baker & Petty, 1994; Phillips, 2003). Hypothesis 7. Information outsiders will perceive the
Thus, we predict the information outsider will be group process as more eVective in in crosscut groups
attended to more in the crosscut and less in the conver- than same race groups and less in convergent groups.
gent than in the same race condition.2

2
Prior research (Harrison et al., 2002) suggests that the eVect of diver- Method
sity structure on group performance is mediated by social integration.
This suggests a mediation hypothesis. However, the experimental para- Participants worked on a drug marketing decision
digm used in this study makes it impractical to generate a large enough
task in three-person groups. The study employed a 3
sample size to eVectively test a mediation hypothesis controlling for non-
independence among group members and experimental condition. Thus, Diversity Structure (crosscut, same race, or
we consider the co-occurrence of participation, attitudes and perfor- convergent) £ 2 Pre-discussion Decision (yes, no)
mance by condition as consistent with the Harrison et al., Wndings. between-groups design. In racially diverse groups, two
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 7

members of the same race formed a racial subgroup strength and decision accuracy. Comparing to Lau and
while the other member was a racial outsider. We focus Murninghan’s (1998) classiWcations, crosscut teams can
on racial diversity because it is among the most obvious align into two subgroups, two ways (i.e., the race sub-
and socially signiWcant attributes of diversity. However, group is not the same as the job-function subgroup)
we expect that our research would apply to social cate- making their faultline the weakest. The same race diVer-
gorization based on inherent characteristics that are ent job-function groups can align into only one sub-
highly salient and imply attitudinal, belief, value or sta- group with one diVerentiating attribute. In contrast, the
tus diVerences among group members (Hollingshead, convergent groups can align into two subgroups, but
1996). The racial majority was composed of White only one way. Because two attributes deWne the single
Americans because of the larger availability of White subgroup, their faultline is the strongest of the three
versus minority participants, which approximates the group structures.
demographics in the US workforce (Bureau of Labor The Pre-discussion Decision manipulation was
Statistics, 2002). We determined diversity structure by the whether participants made and recorded a personal deci-
assignment of group members to job-function through sion about which drug to market before group discus-
labeling and information supplied to members of the sion. In the pre-discussion decision condition, prior to
groups. In the convergent condition, both racial majority entering the group discussion, we asked participants to
members (persons A and B in Fig. 1) represented the decide which drug was the best to develop for market
same job-function and held the same information while based on the information they learned. We gave them a
the racial outsider (person C) held diVerent information form on which to record their choice and their conW-
representing a diVerent job-function. In the crosscut con- dence in this choice. The participant handed this form
dition, the racial outsider (person A) represented the personally to the experimenter. In the no pre-discussion
same job-function as one member of the racial majority decision condition, we did not ask individuals to make a
(person B), while the other member of the racial majority decision prior to discussion.
(person C) represented the diVerent job-function (the
information outsider). In the same race groups, two Procedure
members (persons A and B) represented the same job-
function while the third member (person C) was the job- Prospective participants completed a pre-experimen-
function (information) outsider. tal questionnaire describing their demographic charac-
We use a laboratory context in order to isolate two teristics approximately one week prior to the
attributes of diversity into speciWc structural conWgura- experiment. They were then assigned to experiment
tions. While we acknowledge the limitations of using times in order to compose groups of appropriate demo-
contrived groups in a laboratory context, this methodol- graphic mix. Participants were given a scenario in which
ogy allows us to directly test hypotheses about diversity a pharmaceutical company was faced with a decision of
structure without other confounds that occur in Weld set- which of two cholesterol-reducing drugs to develop for
tings. We chose to use three-person groups and compare market. They were told they would later be assigned to a
same race/diVerent job-function to crosscut and conver- three-person group that would be required to identify
gent diversity structures in order to provide the simplest, which drug was the best to develop for market. We deliv-
most straight-forward test of the faultline hypothesis. ered all instructions via audiotape for standardization.
We use same sex groups to avoid gender diversity within Participants were given an introduction sheet listing
groups. background information about cholesterol and the
Using Shaw’s (2004) computation,3 the faultline issues that they should consider when making a decision
strength for the crosscut groups is 0.1875, for same race about marketing a given drug. This information was
diVerent job-function groups is 0.4375, and for conver- designed to orient the participants to the task and pro-
gent groups is 0.875. Thus, our ordering of conditions is vide a context for the facts that they would subsequently
consistent with a linear relationship between faultline be given. We delivered the job-function assignment
manipulation by telling participants that they would
3
Shaw (2004) computes Faultline Strength as the average of the learn information about the two drugs and that the
faultline strength for each diversity attribute represented within the
information they receive would be either marketing or
group. Faultline strength is the Internal Alignment on an attribute mul-
tiplied by the inverse of the Cross Group Alignment. Internal Alignment medical facts related to each drug. They were told that
is the 2 ratio of the observed frequency of membership within a sub- they should be prepared to discuss the information they
group compared to the expected value assuming random distribution. were about to learn and that they would have to rely on
Cross-Subgroup Alignment is the weighted average of the cross prod- their memory during discussion. Participants had 20
ucts of membership across each pair of subgroups. Faultline Strength
minutes to study the 22 items given to them on a printed
varies from 0.0 to 1.0 with 1.0 being the strongest. Strong faultlines can
result from high internal alignment or from low cross sub-group align- fact sheet. The facts were in random order. We gave par-
ment while low scores can result from little internal alignment or from ticipants a worksheet to help them study the information
high cross sub-group alignment. by sorting the facts for the two drugs, which also served
8 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

to emphasize the job-function assignment manipulation. eVectively use the information outsider’s facts to arrive
We then collected all material and gave participants at the correct choice of Drug 1.
10 min to list as many facts as they could recall about Because our participants were in their workgroups for
each drug. The recall task served as a control to assure only a short time, we mimicked the cueing of informa-
that there were no diVerences across conditions in the tional diVerences by visually labeling every member of
preparation of the participants. We followed the recall the workgroup as per their job-function. Tajfel, Billig,
task with the pre-discussion decision manipulation. Bundy, and Flament (1971) showed that such mere
We then assigned participants to groups and moved labeling of people into diVerent categories is suYcient to
them to private discussion rooms. Audiotaped instruc- produce ingroup–outgroup boundaries, dubbed the min-
tions informed the groups that they would have as much imal group eVect. Many studies have produced ingroup
time as they needed to choose, by consensus, which drug favoritism and outgroup bias using the minimal group
to develop for market. We videotaped the discussions. paradigm (e.g., Diehl, 1988; Gaertner & Insko, 2000;
After reaching consensus, each group recorded its deci- Hertel & Kerr, 2001). Thus, we expect this manipulation
sion and wrote a justiWcation for its choice. Afterwards, to produce a faultline. We told participants that they
participants were individually administered a post-dis- would receive diVerent information and gave them bad-
cussion questionnaire. ges indicating which set of information they received to
emphasize the job-function assignment manipulation.
Participants To control for possible eVects of the speciWc category of
information held by the members, we used a counterbal-
Two hundred nineteen students (102 males and 117 anced design with two diVerent Stimulus Sets of infor-
females) enrolled at a large university in the Eastern mation such that the information outsider held medical
United States were assigned to 73 three-person groups. information in half the groups and marketing informa-
We recruited 153 advanced undergraduates from business tion in the other half.
courses and minority-student support programs. Addi- Pilot testing veriWed all intended drug choices. Addi-
tionally 66 MBA students participated as part of a busi- tionally, during the experiment, in 79% of the groups in
ness course. Each group was composed of members of the which members made pre-discussion decisions, all three
same gender and education level (i.e., MBA students never members began with the intended pre-discussion prefer-
participated along with undergraduate students). The ences. The greater the number of members entering group
racial outsiders were African American in 39 groups, His- discussion with the correct preference, the greater the like-
panic in four groups and Asian in 12 groups. We evenly lihood of groups reaching the correct decision option,
distributed the race of the outsider among the Diversity 2(3, N D 35) D 8.69, p D .03. This is consistent with Gruen-
Structure conditions. Undergraduate students received a feld, Mannix, Williams, and Neale (1996). However, this
ten-dollar incentive or extra credit for participation. relationship was not inXuenced by the Diversity Structure
Incentives were for participation, not performance. Analy- manipulation, 2(6, N D 35) D 0.46, p D 1.00, thus diVer-
sis showed that type of incentive (cash or course credit) ences in pre-discussion preferences were not responsible
did not inXuence performance. Additionally, there were for diVerential eVects across conditions.
no signiWcant diVerences in group structure or dynamics
comparing MBA to undergraduate student groups. Dependent measures

Decision task material and information assignment Decision accuracy


We coded the selection of the best option (Drug 1) as
The decision task was adapted from a drug marketing correct.
decision task used by Kelly and Karau (1999). We
selected a sample of their items to form two diVerent full Videotaped behaviors
information sets (32 items per set). Each participant We coded videotaped group discussions for talking
received partial information consisting of 12 general and attending behavior as discussed in Hypotheses 3 and
information facts (identical for all participants), and 10 4 using the CAMERA behavioral coding system (Kruk
category-speciWc facts consisting of either medical or & Geuze, 1992) which uses a computer interface to elec-
marketing issues regarding two diVerent drugs. In each tronically time stamp the video taped sessions. When the
group, we assigned identical sets of information to two target behavior occurs, coders press assigned keys pro-
of the members, forming an “information subgroup” grammed for each of the behavior sets. The computer
preferring Drug 2. We assigned one member a unique set uses the time stamp to record the duration and fre-
of category speciWc information designed to lead to a quency of the button presses. We showed six coders (two
preference for Drug 1, constituting an “information out- for each behavior) examples of the behaviors, rated sev-
sider.” The complete set of 32 items led to a preference eral practice sessions taken from pilot studies, discussed
for Drug 1 in 77% of pilot test individuals. Groups must discrepancies in the ratings, and continued training until
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 9

an inter-rater reliability of at least .80 was achieved. All ence the processing objective of the participants. Addi-
coders were blind to the experimental conditions and tionally, the association between recall and drug choice
hypotheses and worked independently once trained. was not signiWcant, F (1,99) D 0.92, p > .10.
Some video segments (distributed evenly across condi-
tions) could not be coded for one or more of the behav- Post-experiment questionnaire
iors due to poor audio quality (degrees of freedom are
reported for each analysis). Manipulation checks
The CAMERA system produced total discussion We included questions to verify that participants
time for each session. Raters coded talking as the dura- accurately perceived the informational and racial com-
tion of time that one member of the information sub- position of the group. Participants identiWed their own
group and the information outsider talked. We chose the race and indicated which member(s), if any, were of the
information subgroup member who was a racial major- same race. Participants also identiWed their own job-
ity member in the convergent condition and a racial out- function (medical or marketing) and indicated which
sider in the crosscut condition (person A in Fig. 1) in member(s), if any, represented the same job-function.
order to make a direct comparison of the talking behav- Seventy-one groups remained after removing two
ior of information outsiders versus insiders when they groups in which one or more members incorrectly identi-
are racial minorities versus white. The information out- Wed the racial and/or job-function composition.
sider was a racial majority member in the crosscut condi-
tion but a racial outsider in the convergent condition Control variables
(person C in Fig. 1). Inter-rater agreement for the dura- Participants indicated the extent to which they knew
tion of talking behavior was assessed by correlating rat- other members of the group to control for the eVect of
ings of the two coders for the information outsider prior acquaintance. Two groups had members who
(r D .97) and the information insider (r D .94). The second reported being close or intimate friends with another
behavior coded was the amount of time the racial major- member, confounding our Diversity Structure manipula-
ity member, who was also an information insider, spent tion. We removed these groups from further analyses leav-
attending to the information outsider (inter-rater agree- ing 69 groups in the experimental design. Table 1 identiWes
ment, r D .94). In all conditions this was always person B the number of groups remaining in each condition.
(see Fig. 1). We chose person B because that individual
always shared job-function with one member and race Measures of perceptions of the group
with the other group member (or with both members in We administered three scales to identify constructs
the same-race groups).4 We are interested in determining presumed to reXect group processes. The Group IdentiW-
if this person attends to unique information more when cation scale was nine items with a seven-point Likert
it comes from a person of the same or diVerent race. We scale (Hinkle, Taylor, & Fox-Cardamone, 1989) (e.g., “I
operationalized attending as directing head movement see myself as an important part of this group.”). Oppor-
and gaze toward the information outsider. tunity to InXuence consisted of six questions scored on a
nine-point Likert scale measured one’s ability to inXu-
Control variables ence the group decision (Johnson & Johnson, 1994) (e.g.,
“How understood and listened to did you feel in your
Two independent raters content coded recall tasks group?”). The group decision Process EVectiveness scale
(inter-rater r D .92). We tested to assure that the amount (Reagan & Rohrbaugh, 1990) consisted of 11 items,
of information correctly recalled by individuals was not scored on a seven-point Likert scale with three adapted
confounded with the Pre-discussion Decision manipula- to apply to our task (e.g., “The group considered all rele-
tion. A repeated measures ANOVA comparing recall of vant information in making its decision”).
items related to Drug 1 versus Drug 2 crossed with the
Pre-discussion Decision manipulation showed that par-
ticipants recalled more items relating to Drug 2 Results
(M D 10.10) than Drug 1 (M D 9.45), F (1, 187) D 10.33,
p < .01, but there was no eVect due to the Pre-discussion Decision analysis
Decision manipulation, F (1, 187) D 0.22, p > .10. Thus,
the Pre-discussion Decision manipulation did not inXu- We Wrst conducted a logistic regression of decision
accuracy on stimulus set to assure that the stimulus set
4
did not aVect the decision outcome. The test indicated
We seated person B facing the camera so that we could clearly
there was no eVect of information condition on decisions
monitor her or his attention to the other two members. Because the
other members had their sides turned toward the camera, we were not (Wald test D 0.35, df D 1, p D .55). We then conducted the
able to reliably code their attention (see coding methods later in the logistic regression on decision accuracy by Diversity
manuscript). Structure (crosscut, same, convergent) by Pre-discussion
10 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

Table 1
Logistic regression of decisions on diversity structure and pre-discussion decision conditions
Number correct/total N (proportion correct)
Pre-discussion decision No pre-discussion decision Diversity structure eVect
Crosscut 6/12 (.50) 11/13 (.85) 17/25 (.68)
Same race 2/10 (.20) 5/10 (.50) 7/20 (.35)
Convergent 4/13 (.31) 4/11 (.36) 8/24 (.33)
Pre-discussion decision eVect 12/35 (.34) 20/34 (.59)

EVect B SE Wald Test df SigniWcance EVect size


2 ¤
Parameter estimates, signiWcance tests, and Nagelkerke R measure of eVect size
Diversity structure 6.95 2 .03 .14
Crosscut vs. Same Race ¡1.45 .66 4.79 1 .03
Convergent vs. Same Race 0.03 .66 0.00 1 .96
Pre-discussion decision 1.07 .53 4.04 1 .04 .08
Constant ¡1.41 .83 2.886 1 .09
¤
The Nagelkerke R2 for the combined structure and pre-discussion decision eVect size is .21.

Decision (yes, no). We conducted planned contrasts for information outsider (whose race also varied across con-
convergent versus same-race and crosscut versus same- dition). There was a signiWcant main eVect for the
race conditions. The logistic regression is appropriate repeated measures factor such that the information out-
because all variables in the analysis are categorical. In sider (M D .33) talked more than the information insider
Table 1, we present the count and percent of correct and (M D .21), F (1, 25) D 13.95, p < .01, 2 D .36. There was a
incorrect responses by Diversity Structure and Pre-dis- signiWcant Speaker by Diversity Structure interaction,
cussion Decision conditions. Additionally, we report the F (2, 25) D 4.24, p < .05, 2 D .25 (see Fig. 2). In the same
parameter estimates for each hypothesis test along with race groups, information insiders and outsiders talked
Nagelkerke R2 as a measure of eVect size for each eVect about the same. In the crosscut condition, the informa-
and for the full model (Tabachnik & Fidel, 1989). The tion outsider talked more than the insider, supporting
signiWcant Diversity Structure parameter for crosscut Hypothesis 3. However, disconWrming Hypothesis 3, the
compared with same race groups supports Hypothesis 1, information outsider also talked more in the convergent
indicating that crosscut performed better than same race Diversity Structure groups. Recall in this condition the
groups. However, contrary to our expectation, same race information outsider is also the racial outsider. We note
did not perform better than convergent groups, partially that averages within the information subgroups are
disconWrming Hypothesis 1. Post hoc contrast indicated closer than averages within the racial subgroups, with
that crosscut performed signiWcantly better than conver- information outsiders talking more than information
gent ( D 1.48, SE D .63, Wald D 5.48, p < .05) diversity insiders do, independent of race. Thus, the Wndings
structure. Consistent with Hypothesis 2, the signiWcant regarding the amount of information outsider talking is
parameter estimate for pre-discussion decisions indicates due to diversity structure, not race alone.
that groups in the no Pre-discussion Decision condition
made the correct choice more frequently than groups in
the pre-discussion decision condition. 0.45
Proportion of Total Discussion Time

0.4 Racial
Majority
Analyses of talking and attending behavior 0.35
Racial
0.3 Minority
We conducted a 2 Pre-discussion Decision by 3 0.25
Diversity Structure by 2 Speaker (information 0.2 Racial Racial
insider[person A] vs. outsider[person C]) repeated mea- 0.15 Majority Minority
sures5 ANOVA on the arcsine transformed proportion6 0.1
of total discussion time the information insider (whose 0.05
race varied across condition) talked compared to the 0
Convergent Same Crossed
O 0.31 0.28 0.39
5
We used repeated measures ANOVA because we are comparing the I 0.19 0.27 0.17
talking time of two individuals who are members of the same group. Be- Diversity Structure Condition
cause they are members of the same group, they are not independent.
6 Fig. 2. Proportion of discussion time the information outsider (O)
We used the arsine transformation because the data are propor-
tions. This transformation approximates the normal distribution as- talked compared to the information insider (I) who’s race varies as a
sumed for the ANOVA statistics and does not alter the interpretation. function of diversity structure condition.
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 11

Table 2
Test of post-test measures for the information outsider: culture condition by pre-decision
Source Dependent variable F df p Partial 2
a
Corrected model Group identiWcation 3.75 5/54 .01 .26
Opportunity to inXuenceb 4.14 5/54 .01 .28
Process eVectivenessc 1.68 5/54 .16 .14
Intercept Group identiWcation 2390.57 1/54 .00 .98
Opportunity to inXuence 2762.64 1/54 .00 .98
Process eVectiveness 1263.36 1/54 .00 .96
Diversity structure Group identiWcation 8.62 2/54 .00 .24
Opportunity to inXuence 8.83 2/54 .00 .24
Process eVectiveness 2.41 2/54 .10 .08
Pre-discussion decision Group identiWcation 0.41 1/54 .53 .01
Opportunity to inXuence 2.11 1/54 .15 .04
Process eVectiveness 2.03 1/54 .16 .04
Diversity structure by pre-discussion decision Group identiWcation 0.23 2/54 .80 .01
Opportunity to inXuence 0.08 2/54 .92 .00
Process eVectiveness 1.10 2/54 .34 .04
a
R2 D .26 (adjusted R2 D .19).
b
R2 D .28 (adjusted R2 D .21).
c
R2 D .14 (adjusted R2 D .06).

We conducted a 2 Pre-discussion Decision by 3 and the standard errors for each dependent variable by
Diversity Structure ANOVA on the arcsine transformed Diversity Structure. Pair-wise comparisons for eVects of
proportion of time the racial/information insider spent Diversity Structure (convergent versus same and cross-
attending to the information outsider. There was a main cut versus same) indicated that the information outsider
eVect for Diversity Structure on attending, reported more Group IdentiWcation, greater Opportu-
F (2, 35) D 3.90, p < .05, 2 D .18. Consistent with Hypoth- nity to InXuence, and greater Process EVectiveness in the
esis 4, the crosscut groups (41%) were signiWcantly diVer- crosscut than in the same race condition, partially sup-
ent from same race (25%) (p D .04). However, same race porting Hypotheses 5–7. However, the convergent condi-
groups did not diVer from convergent groups (27%) tion also produced greater opportunity to inXuence than
(p D .82). Thus, while the information outsider talked did the same race condition. The convergent diversity
more than the insider in both convergent and crosscut structure condition did not diVer from same race on the
conditions did: the racial/information insider attended other measures.
to the information outsider more in the crosscut than In order to rule out the explanation that this eVect is
convergent or same race conditions. due to being a racial minority (versus information out-
sider) we conducted the same analysis with the atti-
Analysis of perceptions about the group tudes of the racial minority member toward the group.
Since there is no racial minority in the same race condi-
We conducted a 2 Pre-Discussion Decision by 3 tion this analysis compared crosscut with convergent
Diversity Structure MANOVA on information outsid- Diversity Structure and Pre-discussion Decision condi-
ers’ Group IdentiWcation, Opportunity to InXuence, and tions. The multivariate tests were not signiWcant. The
Process EVectiveness.7 The information outsider shared only univariate test that was signiWcant was for Diver-
race with both other members in the same race condi- sity Structure on perception of the decision process
tion, shared race with one other member who held diVer- eVectiveness (F(1, 48) D 5.59, p D .02, 2 D .11). Racial
ent information in the crosscut condition, and shared minority members were more satisWed with the deci-
neither race nor information with any other member in sion process in the crosscut (M D 5.60, SE D .20) than in
the convergent condition. The multivariate test was sig- the convergent condition (M D 4.94, SE D .20). Thus,
niWcant for Diversity Structure (F (3,53) D 6.95, p < .01, information outsiders feel more group identiWcation
2 D .28). Table 2 shows the univariate test results for and ability to inXuence the group in crosscut versus
each of the dependent variables. Table 3 shows the other conditions, while racial minorities perceive less
coeYcient- reliability estimates, grand means and stan- decision process eVectiveness in the convergent condi-
dard deviations, inter-scale correlations, as well as means tion compared to the crosscut condition. Racial minor-
ities did not diVer on group identiWcation or perceived
7
inXuence on the decision in the crosscut compared with
Based on a prior reviewer’s comments, we conducted an analysis
including all members’ perceptions of the group across all conditions.
convergent conditions. Thus, the Wndings relating atti-
We found no diVerence across conditions except for the information tudes to diversity structure condition are not due
outsider. strictly to racial minority status.
12 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

Table 3
Descriptive statistics for post-test measures
Mean, standard deviation, coeYcient- reliability and inter scale correlations
Mean SD Coef- GI OI
Group identiWcation (GI) 5.54 .90 0.86
Opportunity to inXuence (OI) 5.81 .89 0.84 .71
Process eVectiveness 5.06 1.06 0.81 .40 .42
Dependent variable Diversity structure Main eVect for diversity structure
Mean post-test measures for the information outsider by diversity structure condition
Group identiWcation Crosscut 6.02 (.16)b
Same Race 4.93 (.23)
Convergent 5.32 (.17)
Opportunity to inXuence Crosscut 6.21 (.16)b
Same Race 5.04 (.23)
Convergent 5.77 (.17)a
Process eVectiveness Crosscut 5.40 (.21)a
Same Race 4.66 (.30)
Convergent 4.94 (.22)
Parentheses indicate standard errors. Comparisons are Convergent and Crosscut compared to Same Race.
a
p < .05.
b
p < .01.

Discussion and conclusions of the information outsider, or attending to the informa-


tion outsider, conWrms that diversity structure, not race,
Groups in which race crosscut job-function had is the critical factor in our Wndings. In our study, cross-
greater decision accuracy than racially homogeneous cutting allows the job-function outsider to share racial
groups while convergent and same race group perfor- group membership with one of the job-function sub-
mance did not diVer. The groups in this study whose group members allowing them to be more persuasive
members did not make a pre-discussion decision outper- (Van Knippenberg & Wilke, 1992).
formed groups whose members did record an individual Our study showed that when race is crossed with job-
decision prior to discussion. Crosscut Diversity Struc- function the information outsiders identiWed more with
ture with no Pre-discussion Decision was the only condi- the group, perceived more opportunity to inXuence the
tion in which groups signiWcantly outperformed random group, and rated the process as more eVective than when
chance (.50) and outperformed individuals who had full all members were of the same race, or when race and job-
information in the pilot testing. Only crosscut Diversity function converged. The results of the video analysis
Structure with no Pre-discussion Decision resulted in the indicated that these perceptions were accurate. The
facilitative communication necessary to achieve the crosscut condition was the only condition in which the
results desired of cross-functional groups. information outsider talked signiWcantly more than the
The Wndings for talking and attending behavior help insider, and was attended to the most. It was in the cross-
us understand the role of racial and information outsid- cut condition that information outsiders had a link to
ers. Recall that in the same race groups the information the job-function subgroup via racial similarity with a
outsider is a majority race member, while in the conver- subgroup member. These Wndings are consistent with the
gent condition the information outsider diVers in race social categorization model (Brewer & Miller, 1984;
from the information insiders. In the crosscut condition, Marcus-Newhall et al., 1993) and Lau and Murnin-
the information outsider is a majority race member ghan’s (1998) faultline analogy. The overlapping of
while one of the information insiders is the minority race group boundaries that exists with crossed categoriza-
member. In both crosscut and convergent conditions, the tions weakens faultlines.
information outsider talked more than the information Contrary to our hypotheses, our research found no
insider of the opposite race. Racial minority and major- diVerence between convergent diversity structure and
ity members are more similar within job-function assign- same race groups in decision accuracy or the processes
ment than are participants of the same race across job- that aVect it. Thus, convergence eliminated the potential
function assignments. Similarly, attending diVered only positive eVects of diversity but did not lead to poorer
by Diversity Structure, in which the racial/information performance than same race groups. We suspect this
insider attended to the information outsider more in the occurred because the locus of poor performance was the
crosscut than same race condition, while convergent ver- inability to share information. In our study, job-function
sus same race did not diVer. Thus, the fact that conver- (and thus information) was the source of one faultline. In
gent and same race groups did not diVer on participation the same race groups, the lack of a salient characteristic
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 13

crossing the job-function faultline may have allowed the that moderately diverse groups, deWned as those in
preferences generated by the faultline to be as strong as which characteristics overlap subgroups, perform better.
the race plus job-function faultline of the convergent This structure should allow for better group communi-
groups. This suggests that race is not the critical dimen- cation and ultimately a better group product. Failure to
sion in our Wndings: Diversity Structure is. structure the diversity dimensions of the group with
Thus, crosscutting reduces category-based informa- cross category characteristics wastes the richness of the
tion processing and decreases the likelihood of commu- variation in knowledge and perspectives that diversity
nication barriers formed by clear-cut group boundaries. can bring to the group.
The fact that job-function roles vary within race may Another way to improve group functioning is to
reduce the assumed similarity among group members establish a group norm that prevents members from
who share race and prime group members to expect coming to the group with a predetermined decision and
diVerences in perspective within the racial pair. If job- focuses on exploring all relevant information by the
function and demographic subgroups converge, diversity group. Maier (1963), in noting the propensity of individ-
may actually limit pooling of the information that diver- uals and groups to generate solutions prior to analyzing
sity can bring to the group. Our same race groups were the problem, cautioned individuals and groups not to
similar to Gruenfeld et al. (1996) unfamiliar groups. make choices prior to a thorough discussion of the prob-
Gruenfeld et al., found familiarity among group mem- lem and the facts at hand. Our Wndings support his prop-
bers allowed the group to transcend preferences due to osition. The negative eVect of pre-discussion decisions
information diVerences. Familiar groups pooled infor- found in our study is consistent with prior research on
mation while unfamiliar groups aggregated judgments. the common knowledge eVect and the attentional focus
We believe that our crosscut condition had a similar model. That research has shown that individual judg-
eVect. Crosscutting appears to have facilitated the social ments are given more weight in the group decision than
integration (Harrison et al., 2002) and open communica- the information available to the group as a whole
tion (Simons et al., 1999) necessary for diversity to ben- (Gigone & Hastie, 1993) and that successful groups pool
eWt group decisions. information rather than aggregating individual judg-
Our Wndings also provide a possible explanation for ments (Gruenfeld et al., 1996). Individual judgments
why prior research using the faultline analogy has found a become an attentional Wlter through which information
non-linear relationship between faultline strength and is processed (Kelly & Karau, 1999). Thus, the absence of
group outcomes (Gibson & Vermeulen, 2003; Thatcher a pre-discussion decision is one of the necessary condi-
et al., 2003). In their measurement system, groups with a tions for optimal group performance.
high degree of cross-subgroup alignments have moderate
faultline strength. In our conceptualization, and Shaw’s Limitations and recommendations for future research
(2004) measurement system, cross-subgroup alignment
results in low faultline strength. The critical aspect in group In this research, we used a decision task for which
performance and cohesion may not be the amount of there is a binary choice. We did so because we wanted to
diversity or homogeneity. The critical factor may be those assess decision accuracy on choice tasks similar to that
salient attributes that cross, or transcend, social subgroups done by Stasser and colleagues. However, judgment
that allow the group members to think of themselves as a tasks in which the accuracy of judgments or decision
whole, rather than as subgroups within the group. tasks with multiple decision choices would provide a
greater range of responses and potentially provide
Implications for practice increased power to address similar questions. We also
focus on just one element of diVerence in job-function
Our research shows that one way to reduce the barri- diversity, that of information. Job functions bring with
ers to communication is to assure that attributes align them a multitude of diVerences. This is why it would be
members across subgroups. Thus, in a group composed impossible to conduct this research in a Weld setting. The
of marketing and manufacturing representatives, having ability to isolate a single characteristic of job-function,
a racial minority who shares job-function with a racial information relative to decision-making, allows direct
majority member, or having a manufacturing represen- tests of our hypotheses. In the Weld, the many job-func-
tative who shares race with a marketing representative tion diVerences would confound the method and make
may serve to open lines of communication. In groups speciWc conclusions impossible.
where diverse members bring potentially diVerent infor- Our Wndings are limited in that they come from
mation or viewpoints, our research and the theorizing of groups with short duration and with no history among
Lau and Murninghan (1998) suggests that the group their members. We expect that some of the eVects of sub-
would beneWt by identifying some characteristic that group boundaries, relationship, and task conXict on
crosses subgroup faultlines. Gibson and Vermeulen group performance change over time (Jehn & Mannix,
(2003) echo this recommendation because they found 2001). Time may neutralize the eVects of surface level
14 J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15

diversity (Harrison et al., 1998). However, we postulate Clark, R. D., & Maass, A. (1988). The role of social categorization and
that when surface level characteristics converge with perceived source credibility in minority inXuence. European Journal
of Social Psychology, 18(5), 381–394.
deep level attitude or value diVerences, time may accen- Davis, J. H. (1973). Group decision and social interaction: A theory of
tuate those eVects as deep level diVerences are uncovered social decision schemes. Psychological Review, 80, 97–125.
which reinforce the surface level categorization. Future Davis, J. H., Stasson, M. F., Parks, C. D., Holbert, L., Kameda, T.,
research needs to investigate information sharing in Zimmerman, S. K., & Ono, K. (1993). Quantitative decisions by
groups over time or in which members have a history groups and individuals: Voting procedures and monetary awards
by mock civil juries. Journal of Experimental Social Psychology,
together. Unfortunately, this is a diYcult task to accom- 29, 326–346.
plish in the laboratory. Field research provides a better Deutsch, M., & Gerard, H. B. (1955). A study of normative and infor-
context to test time-related eVects. However, Weld mational social inXuence upon individual judgment. Journal of
researchers seldom have the freedom to compose groups Abnormal and Social Psychology, 51, 629–636.
to test speciWc group composition hypotheses or ran- Diehl, M. (1988). Social identity and minimal groups: The eVects of
interpersonal and intergroup attitudinal similarity on intergroup
domly assign groups to tasks. Researchers need to look discrimination. British Journal of Social Psychology, 27, 289–300.
for settings in which they can assess group composition Gaertner, S. L., & Insko (2000). Intergroup discrimination in the mini-
variations and communication. mal group paradigm: Categorization, reciprocation, or fear? Jour-
Another limitation of our research is that our groups nal of Personality and Social Psychology, 79, 77–94.
were composed such that the racial majority was always Gibson, C., & Vermeulen, F. (2003). A healthy divide: Subgroups as a
stimulus for team learning behavior. Administrative Science Quar-
White Americans and the minority member was always a terly, 48, 202–239.
social minority. Racial minorities bear the burden of Gigone, D., & Hastie, R. (1993). The common knowledge eVect: Infor-
assumed status and power diVerences. Research using the mation sharing and group judgment. Journal of Personality and
minimal group paradigm (Gaertner & Insko, 2000; Hertel Social Psychology, 65, 959–974.
& Kerr, 2001) suggests that convergent subgroups have Goethals, G. R., & Nelson, E. R. (1973). Similarity in the inXuence pro-
cess: The belief-value distinction. Journal of Personality and Social
negative eVects even when there is no presumption of sta- Psychology, 25, 117–122.
tus or power diVerences, but what happens when the sta- Gruenfeld, D. H., Mannix, E. A., Williams, K. Y., & Neale, M. A.
tus or power diVerences reverse? What would happen if (1996). Group composition and decision making: How member
the racial minority in our groups were a social majority familiarity and information distribution aVect process and perfor-
member (e.g., a single White among a group of African- mance. Organizational Behaivor and Human Decision Processes, 67,
1–15.
Americans)? This is an excellent question for future Harrison, D. A., Price, K. H., & Bell, M. P. (1998). Beyond relational
research. While we believe that our Wndings will generalize demography: Time and the eVects of surface- and deep-level diver-
to a variety of groups in which social categories connote sity on work group cohesion. Academy of Management Journal, 41,
diVerences in status, we need further research to test the 96–107.
generalizability of our Wndings to other compositions such Harrison, D. A., Price, K. H., Gavin, J. H., & Florey, A. T. (2002). Time,
teams, and task performance: Changing eVects of surface- and
as where White Americans are the cultural outsiders, or deep-level diversity on group functioning. Academy of Management
where other inherent characteristics such as gender form Journal, 45, 1029–1045.
the basis of subgroup diVerences. Hertel, G., & Kerr, N. L. (2001). Priming in-group favoritism: The
As researchers continue to examine the tendency of impact of normative scripts in the minimal group paradigm. Jour-
decision-making groups to fall short of optimal perfor- nal of Experimental Social Psychology, 37, 316–324.
Hinkle, S., Taylor, L. A., & Fox-Cardamone, L. (1989). Intragroup
mance, they should identify factors that could allow identiWcation and intergroup diVerentiation: A multicomponent
groups to excel. The future of group decision-making approach. British Journal of Social Psychology, 28, 305–317.
research and the study of demographic and functional Hollingshead, A. B. (1996). Information suppression and status persis-
diversity need to converge to continue this process of tence in group decision making: The eVects of communication
discovering the true potential of diverse groups. media. Human Communication Research, 23, 193–219.
Jackson, S. E. (1992). Team composition in organizational settings:
Issues in managing an increasingly diverse workforce. In S. Wor-
chel, W. Wood, & J. A. Simpson (Eds.), Group processes and produc-
References tivity (pp. 138–173). Newbury Park, CA: Sage.
Jackson, S. E., May, K. E., & Whitney, K. (1995). Understanding the
Baker, S. M., & Petty, R. E. (1994). Majority and minority inXuence: dynamics of diversity in decision-making teams. In R. A. Guzzo &
Source–position imbalance as a determinant of message scrutiny. E. Salas (Eds.), Team eVectiveness and decision making in organiza-
Journal of Personality and Social Psychology, 67, 5–19. tions (pp. 204–261). San Francisco: Jossey-Bass.
Brewer, M. B., & Miller, N. (1984). Beyond the contact hypothesis: Jehn, K. A., & Mannix, E. A. (2001). The dynamic nature of conXict: A
Theoretical perspectives on desegregation. In N. Miller & M. B. longitudinal study of intragroup conXict and group performance.
Brewer (Eds.), Groups in contact: The psychology of desegregation Academy of Management Journal, 44, 238–251.
(pp. 281–302). Orlando, FL: Academic Press. Jehn, K. A., Northcraft, G. B., & Neale, M. A. (1999). Why diVerences
Brock, T. C. (1965). Communicator-recipient similarity and decision make a diVerence: A Weld study of diversity, conXict, and perfor-
change. Journal of Personality and Social Psychology, 1, 650–654. mance in workgroups. Administrative Science Quarterly, 44, 741–
Bureau of Labor Statistics (2002). Employed persons by detailed occu- 763.
pation, sex, race and Hispanic origin. Available from: http:// Johnson, D. W., & Johnson, F. P. (1994). Joining together. Needham
www.bls.gov/cps/cpsaat11.pdf. Heights, MA: Allyn & Bacon.
J.E. Sawyer et al. / Organizational Behavior and Human Decision Processes 99 (2006) 1–15 15

Kanter, R. M. (1977). Some eVects of proportions on group life: Reagan, P., & Rohrbaugh, J. (1990). Group decision process eVective-
Skewed sex ratios and responses to token women. American Journal ness: A competing values approach. Group and Organization Stud-
of Sociology, 82, 965–990. ies, 15, 20–43.
Kelly, J. K., & Karau, S. J. (1999). Group decision making: The eVects Shaw, J. B. (2004). The development and analysis of a measure of
of initial preference and time pressure. Personality and Social Psy- group faultlines. Organizational Research Methods, 7, 66–100.
chology Bulletin, 25, 1342–1354. Simons, T., Pelled, L. H., & Smith, K. A. (1999). Making use of diVer-
Kirchmeyer, C., & Cohen, A. (1992). Multicultural groups: Their per- ence: Diversity, debate, and discussion comprehensiveness in top
formance and reactions with constructive conXict. Group and Orga- management teams. Academy of Management Journal, 42, 662–673.
nization Management, 17, 153–170. Stasser, G. (1988). Computer simulation as a research tool: The DIS-
Kramer, R. M. (1991). Intergroup relations and organizational dilem- CUSS model of group decision making. Journal of Experimental
mas: The role of categorization processes. In B. M. Staw & L. L. Social Psychology, 24, 393–422.
Cummings (Eds.), Research in organizational behavior (Vol. 13, pp. Stasser, G., & Dietz-Uhler, B. (2002). Collective choice, judgment and
191–228). London: JAI Press Ltd. problem solving. In M. A. Hogg & S. Tindale (Eds.), Blackwell
Kruk, M. R., & Geuze, R. H. (1992). CAMERA: A system for collecting handbook of social psychology: Group processes (pp. 31–55). Oxford,
and correcting behavioral data. The Netherlands: University of UK: Blackwell.
Groningen. Stasser, G., & Titus, W. (1985). Pooling of unshared information in
Larson, J. R., Foster-Fishman, P. G., & Keys, C. B. (1994). Discussion group decision making: Biased information sampling during dis-
of shared and unshared information in decision-making groups. cussion. Journal of Personality and Social Psychology, 48, 1467–
Journal of Personality and Social Psychology, 67, 446–461. 1478.
Lau, D. C., & Murninghan, J. K. (1998). Demographic diversity and Tajfel, H., Billig, M. G., Bundy, R. P., & Flament, C. (1971). Social cat-
faultlines: The compositional dynamics of organizational groups. egorization and intergroup behaviors. European Journal of Social
Academy of Management Review, 23, 325–340. Psychology, 1, 149–178.
Maier, N. R. F. (1963). Problem-solving discussions and conferences: Tajfel, H., & Turner, J. (1986). The social identity theory of inter-
Leadership methods and skills. New York: McGraw-Hill. group behavior. In S. Worchel & W. G. Austin (Eds.), The social
Marcus-Newhall, A., Miller, N., Holtz, R., & Brewer, M. B. (1993). psychology of intergroup relations (pp. 7–24). Chicago: Nelson-
Cross-cutting category memberships with role assignment: A Hall.
means of reducing intergroup bias. British Journal of Social Psy- Tabachnik, B. G., & Fidel, L. S. (1989). Using multivariate statistics
chology, 32, 125–146. (second ed.). New York: Harper and Row.
Maznevski, M. L. (1994). Understanding our diVerences: Performance Thatcherr, S. M. B., Jehn, K. A., & Zanutto, E. (2003). Cracks in diver-
in decision-making groups with diverse members. Human Relations, sity research: The eVects of diversity faultlines on conXict and per-
47, 531–552. formance. Group Decision and Negotiation, 12, 217–241.
Milliken, F., & Martins, L. (1996). Searching for common threads: Turner, J. (1982). Toward a cognitive redeWnition of the social group.
Understanding the multiple eVects of diversity in organizational In H. Tajfel (Ed.), Social identity and intergroup behavior (pp. 15–
groups. Academy of Management Review, 21, 402–433. 40). Cambridge, UK: Cambridge University Press.
Mugny, G., & Papastamou, S. (1982). Minority inXuence and psycho- Van Knippenberg, D., & Wilke, H. (1992). Prototypicality of argu-
social identity. European Journal of Social Psychology, 12, 379–394. ments and conformity to ingroup norms. European Journal of
Nemeth, C. J., & Wachtler, J. (1983). Creative problem solving as a Social Psychology, 22, 141–155.
result of majority vs minority inXuence. European Journal of Social Webber, S. S., & Donahue, L. M. (2001). Impact of highly and less job-
Psychology, 13(1), 45–55. related diversity on work group cohesion ande performance: a
O’Reilly, C., Caldwell, D., & Barnett, W. (1989). Workgroup demogra- meta-analysis. Journal of Management, 27, 141–162.
phy, social integration, and turnover. Administrative Science Quar- Wiegman, O. (1985). Two politicians in a realistic experiment: Attrac-
terly, 34, 21–37. tion, discrepancy, intensity of delivery, and attitude change. Journal
Pelled, L. H., Eisenhardt, K. M., & Xin, K. R. (1999). Exploring the of Applied Social Psychology, 15, 673–686.
black box: An analysis of work group diversity, conXict, and per- Wilder, D. A. (1990). Some determinants of the persuasive power of
formance. Administrative Science Quarterly, 44, 1–28. ingroups and outgroups: Organization of information and attribu-
Petty, R. E., & Cacioppo, J. (1986). The elaboration likelihood model of tion of independence. Journal of Personality and Social Psychology,
persuasion. In L. Berkowitz (Ed.), Advances in experimental social 59, 1202–1213.
psychology (Vol. 19, pp. 123–205). New York: Academic Press. Williams, K. Y., & O’Reilly, C. A. (1998). Demography and diversity in
Phillips, K. W. (2003). The eVects of categorically based expectations organizations. In B. M. Staw & L. L. Cummings (Eds.), Research in
on minority inXuence: the importance of congruence. Personality Organizational Behavior (Vol. 20, pp. 77–140). London: JAI Press
and Social Psychology Bulletin, 29(1), 3–13. Ltd.

También podría gustarte