Upload
mjans71
View
216
Download
0
Embed Size (px)
Citation preview
8/8/2019 Networks, Diversity and Performance
1/33
RESEARCH PAPER SERIES
GRADUATE SCHOOL OF BUSINESS
STANFORD UNIVERSITY
Research Paper No. 1585
Networks, Diversity, and Performance:
The Social Capital of Corporate R&D Units
Ray Reagans
Ezra W. Zuckerman
October 1999
8/8/2019 Networks, Diversity and Performance
2/33
Data for this research were collected with the support of NSF grant ISI-8304340, Division of Industrial Science and
Technological Innovation, Bernard P. Cohen, Principal Investigator. We are in debt to Professor Cohen, Lisa Troyer,
and Shaul Gabbay for help with obtaining and facilitating use of the data. We have also benefited from the comments
of Linda Argote, Ronald Burt, Kathleen Eisenhardt, Paul Goodman, and Barbara Lawrence. We alone are responsible
for any remaining mistakes.
Research Paper No. 1585
Networks, Diversity, and Performance:
The Social Capital of Corporate R&D Units
Ray E. Reagans
Ezra W. Zuckerman
October, 1999
8/8/2019 Networks, Diversity and Performance
3/33
Networks, Diversity, and Performance:
The Social Capital of Corporate R&D Units
Abstract
Paralleling debates in the social capital literature, existing theory is of two minds
regarding the performance implications of demographic diversity. One view sees
diversity as problematic for organizational teams because of the strains that plague
relationships across social divides. According to the second view, diverse teams are
enriched through linkages between individuals with different skills, resources, and
perspectives. This debate is usefully framed in terms of a teams social network: do
teams achieve a higher performance when the network among members connects diverse
individuals or when ties are localized among members of the same demographic
category? In this study, unique data on the social networks, organizational tenure, and
performance of 223 corporate R&D units are analyzed to address this question. We find
that, controlling for its tenure distribution, a team is less productive when its network
remains concentrated among members of neighboring cohorts. In addition, we find that
teams with more dense patterns of communication are more productive than units with
more sparse structures.
8/8/2019 Networks, Diversity and Performance
4/33
1
Introduction
Do organizational teams achieve higher performance when their membership is diverse
or when members are homogeneous? According to one line of thinking, diversity is
problematic because it introduces social divisions that hinder effective teamwork. In his
classic statement on organizational demography, Pfeffer (1983) illustrates this view with
the example of tensions between members of different organizational cohorts. He argues
that informal social networks and a sense of shared identity take root among individuals
who enter the organization at the same time. This leads to an increased capacity for
intracohort communication but a potential for strain in intercohort relations.
Homogeneous groups are thus expected to perform at a higher level because such groups
coordinate their actions more easily than do diverse teams (cf., McCain et al., 1983;
OReilly et al., 1989; Zenger and Lawrence 1992).
A second approach to the question of diversity argues that a heterogeneous
membership improves a teams performance. Proponents of this view also invoke
organizational tenure to illustrate their thinking. For example, Ancona and Caldwell
(1992, p. 355) write that teams that draw their members from various cohorts achieve
higher performance because Members who have entered the organization at different
times know a different set of people and often have different technical skills and different
perspectives on the organizations history. That is, since greater diversity entails
relationships among people with different sets of contacts, skills, information, and
experiences, diverse teams enjoy an enhanced capacity for creative problem solving. By
contrast, while homogeneous groups may in fact be more harmonious, the performance of
such teams is limited by the relative redundancy of members perspectives, information,
8/8/2019 Networks, Diversity and Performance
5/33
2
and resources (e.g., Ancona and Caldwell 1992; Bantel and Jackson 1989; Pelled et al.,
1999).1
These opposing views on the value of diversity have parallels in the burgeoning
literature on social capital. While work falling under this heading is quite diverse (see
e.g., Portes 1998; Adler and Kwon 1999; Gabbay and Leenders 1999), most relevant to
debates on organizational diversity are the two rival conceptions of social capital that
emerge from social network theory. The first view emphasizes the benefits that arise
when social networks are characterized by closure (Coleman 1988; 1990); i.e., when
relations are embedded in a dense web of third-party connections (cf., Granovetter1
985).
Such closure is thought to foster identification with the group (Portes and Sensenbrenner
1993) and a level of mutual trust that facilitates exchange (Coleman 1988). A second
approach, advocated by Burt (1992; cf., 1982), understands social capital as value derived
from structural holes or the absence of social closure. Burt argues that an actor who
bridges disconnected social circles enjoys better access to information (cf., Granovetter
1973); such an actor also profits when she brokers between a pair of rivals or
disconnected exchange partners, thereby controlling the terms of trade (cf., Simmel 1955;
Pfeffer and Salancik1978; Cook et al., 1983).
It is evident that, aside from such issues of brokerage and control2, the network
structures discussed as providing social capital underlie the two views on diversity. In
particular, those who stress the drawbacks of diversity do so precisely because
heterogeneity hinders the emergence of the dense social networks that lead members to
identify with one another and thereby facilitate mutual coordination. Indeed, just as
discussions of social closure tend to focus on the collective trust found in cohesive, ethnic
8/8/2019 Networks, Diversity and Performance
6/33
8/8/2019 Networks, Diversity and Performance
7/33
4
same degree of tenure diversity. That is, controlling for the diversity of a group, we may
ask whether the groups performance is degraded or enhanced by an increase in contact
across demographic categories.
We believe that such a direct examination of networks across demographic
categories represents an important step in advancing the diversity-performance debate. In
the present study, we undertake such an analysis using a unique data set of corporate
R&D units. The survey of 223 R&D units we study includes detailed information on the
communication networks among members of each unit. In addition, as the data set also
contains information on organizational tenure, unit performance, and other relevant
factors, it frees us from simply positing that a particular communication pattern mediates
the affect of tenure diversity on performance and enables us to take a direct look at how
relations within and across cohorts affects a teams performance.
We proceed as follows. First, we describe the setting for our research. Next, we
present the hypotheses that we test in our analysis. In the subsequent sections, we
describe the methods used and then present our results. To preview our main findings,
we observe that R&D units that have a high frequency of cross-cohort ties are more
productive than those whose networks are concentrated among members of neighboring
cohorts. This result supports the view that teams perform at a higher level when they
bring into contact individuals who have a variety of backgrounds and experiences. In
addition, we find that teams with a higher density of interaction are more productive, a
result which confirms the importance of social closure. Finally, our analysis indicates
that demographic composition of the teams has no direct impact on performance; rather,
the diversity effects we find are apparent only when seen as a property of the teams
8/8/2019 Networks, Diversity and Performance
8/33
5
informal network structure. We conclude with a discussion of implications of these
results for the study of diversity and performance.
The Setting: Corporate R&D Units
A consistent theme sounded by studies of corporate R&D over the past thirty years
highlights the importance of informal communication networks as a critical means by
which scientists keep up with technological and scientific developments as well as
organizational directives (e.g., Allen 1977; Katz and Tushman 1979; Katz 1982; Zenger
and Lawrence1
992; Hansen1
999). Furthermore, while much research focuses on the
communication links that bind R&D units to one another, within-unit interaction has been
shown to be important as well (Allen and Cohen 1969; Tushman 1977). Thus, to the
degree that the network patterns posited by the opposing perspectives on diversity have
greater effects on performance in contexts where networks are more salient, such effects
should be evident in a study of corporate R&D units.
The survey data we use in our analysis are uniquely suited to shed light on the
issues at hand. The survey of 223 R&D units in 29 corporations was administered in
1985-1986. The sample was designed to gain a broad coverage of industries and tasks
rather than a representative sample of all R&D units (see Cohen and Zhou 1991; Shenhav
1991). It included a wide variety of R&D units, ranging from teams focused on basic
research to those working on more applied projects such as product and process
development and improvement. Unit leaders or managers were asked a series of
questions regarding their unit, including its productivity. Unit members were asked a
wide variety of questions on their work and how it relates to that of other members.
8/8/2019 Networks, Diversity and Performance
9/33
6
Most relevant to the issues at hand, the data include information on the members
tenure in the organization as well as their level of contact with fellow unit members.
Organizational tenure is a useful demographic variable in this context because its
meaning is largely invariant across organizations and subunits and because it allows us to
link our research with a large number of previous studies on the correlates of diversity, in
which organizational tenure has figured quite prominently (e.g., Ancona and Caldwell
1992; OReilly et al., 1989; Pfeffer1983; Wagner et al., 1983; Zenger and Lawrence
1992).3
Our network analysis focuses on the information generated by asking each
scientist to indicate how frequently he communicates with each colleague (0 = never,1
= less than once a month, 2 = 1 to 3 times a month, 3 = 1 to 3 times a week, 4 =
daily). Together with data on organizational tenure, these network data afford a rare
opportunity to assess directly whether team performance is enhanced or degraded by ties
that cross demographic boundaries.
Hypotheses
In the analysis that follows, we focus on a teams network homogeneity, the extent to
which within-team networks link members of proximate cohorts rather than members of
distant cohorts. Note that the demographic diversity of a team necessarily conditions the
level of network homogeneity it displays: at the limit, when all members belong to the
same demographic category, team networks will be completely homogeneous as well.
However, following Lawrence (1997), we posit that the network structure of a team is not
reducible to its demographic composition. For instance, two teams with the same tenure
distribution may differ quite substantially in the extent to which members of different
8/8/2019 Networks, Diversity and Performance
10/33
7
cohorts interact with one another. Thus, our strategy is to examine the impact on
performance of a teams network while controlling for its demographic composition. In
addition, our measure of network homogeneity is conditioned on a scientists opportunity
for interacting with members of distant cohorts, which varies by her tenure and the tenure
distribution of the group. Thus, we examine the impact of network homogeneity
independent of a teams demographic composition.
In particular, we aim to assess which of the two theories discussed above best
explains the relationship between network homogeneity and performance. According to
the first view, teams are most effective when they are characterized by dense,
homogeneous networks (e.g., Pfeffer1983; McCain et al., 1983; OReilly et al., 1989;
Zenger and Lawrence 1992). By contrast, coordination among members of different
demographic categories should be more difficult, thereby lowering team performance.
Following scholars who understand social capital as a property emergent from closed,
community-like networks (e.g., Coleman 1988, 1990; Portes and Sensenbrenner1993),
this perspective expects the relatively inharmonious relations typical of a diverse task
group to limit its effectiveness. That is,
H1a: The greater the network homogeneity of a team, the higher its performance.
The second view emphasizes the benefits of ties that cross social boundaries,
arguing that encounters among people with differing skills, information, and outlooks,
enhance a teams capacity for creative action (e.g., Ancona and Caldwell 1992; Bantel
and Jackson 1989; Pelled et al., 1999; March 1991, pg. 74-81). Consistent with those
who see social capital as a matter of bridging disconnected social circles (e.g., Burt
8/8/2019 Networks, Diversity and Performance
11/33
8
1992), this approach expects a teams performance to improve when its network includes
numerous links between members of different cohorts. That is, while such relations may
produce greater tension, they enhance the teams capacity for creative problem solving to
the extent that such a team tends to be more productive than teams with lower network
diversity. Thus, this perspective entails that:
H1b: The greater the network homogeneity of a team, the lower its performance.
The assumption in the foregoing discussion is that a groups demography has no
direct impact on a groups performance. Rather, any effect of demography is mediated
through the teams network structure. Indeed, it is difficult to understand how the
diversity of a team could affect its performance without supposing that diverse teams
differ from those that are more homogeneous in how team members relate to one another.
It is possible, however, that such mediation occurs not through changes in the level of
interaction among team members but through other social processes. For example,
following social categorization theory (cf., Tajfel 1981; Turner1987), one variant of the
perspective that sees diversity as problematic expects individuals to attribute positive
characteristics to members of their own demographic category and negative traits to other
categories. As a result, greater diversity may heighten social tensions via an increase in
the ratio of negative to positive attributions but without changing interaction patterns
(e.g., Ely, 1994; Pelled 1997; Riordan and Shore 1997). Similarly, Ancona and Caldwell
(1992) suggest that diversity may impede performance not by lowering social cohesion
but by degrading the teams capacity for such group processes such as goal setting.
Alternatively, it may be that the benefits of diversity arise not through the development of
relationships among individuals with different perspectives but by creatively resolving
8/8/2019 Networks, Diversity and Performance
12/33
9
conflict between such actors in a team context (Eisenhardt et al., 1997). That is, whether
diversity promotes or erodes team performance, such an effect may not occur through its
social network.
Unfortunately, the data set we use does not allow us to measure social
categorization or group processes. However, to the extent that such processes affect
performance independent of changes in a teams social network, a relationship between
demographic diversity and performance should be observed independent of the network
structure of the team. Thus, in parallel with the two hypotheses presented regarding the
effect of network homogeneity, we submit two contradictory hypotheses regarding the
impact of demographic diversity on performance. That is,
H2a: The greater the demographic diversity of a team, the lower its performance.
H2b: The greater the demographic diversity of a team, the higher its
performance.
Method
Analytic Framework
We test these hypotheses with a set of regression analyses, which assess the effects on
team productivity of the diversity variables and other factors discussed below. Since our
data set has a nested structure in that multiple units within the same firm are considered,
we estimate the models as fixed-effects regression analyses whereby a dummy variable
for each firm is included in each model. Coefficients in these equations reflect how a
covariate affects within-firm variation in performance (see, e.g., Hannan and Young
1977).
8/8/2019 Networks, Diversity and Performance
13/33
10
Measures
Dependent Variable: Team Productivity We construct our measure of team performance
from answers to the following question, asked of unit managers regarding the eleven
items listed in table 1:
Consider each of the following written products and/or prototypes that could have
resulted from the work of this unit during the last three years. How many of each
has this unit produced? For each product, choose one of the listed alternativesand enter its number in the space provided.
Note that, since this question asks managers only about the amount of work a unit has
produced in various areas,4
it can speak only to the groupsproductivity but not other
aspects of performance. In an effort to discern a single measure of productivity, we
perform a least-squares regression factor analysis (see Jobson 1992, ch.9) on the
correlation matrix among the eleven items. Overall, three factors explain 44% of the
common variance and the first factor explains 24% of this variance. Table 1 gives
summary statistics for the eleven items and their loadings on this first factor. As we can
see, this factor represents a basic productivity dimension in that it reflects the extent to
which a unit has generated many papers, proposals, patents, and reports. While all eleven
items load positively on this factor, those that have low factor loadings such as books and
computer programs are relevant only to a small number of units. In the analysis that
follows, we treat the first factor score as our dependent variable, keeping in mind that,
while this measure appears to capture key aspects of team productivity, there may be
other dimensions of team performance that are not reflected in this variable.
Table 1 About Here
8/8/2019 Networks, Diversity and Performance
14/33
11
Main Independent Variables
Tenure Diversity We measure diversity in organizational tenure with the Gini coefficient
of mean difference (CMD; Kendall and Stuart, 1977; pg. 48):5
The CMD differs from the more familiar Gini index in that the latter divides the
CMD by twice the group mean. The typical motivation for scaling a measure of
inequality such as the Gini index--or, its near equivalent, the coefficient of variation-- by
a function of the mean is to tap the intuition that, holding constant the dispersion on some
resource, an increase in the level on that resource lowers the degree of felt inequality
(Allison 1977, p. 867). For instance, as the average level of income in a population
increases, absolute differences in income become less important. However, while this
rationale for scaling by the mean makes sense for inequality, it is not clear whether it
applies to the case of diversity measure. For instance, with respect to organizational
tenure, this would imply that work groups with higher mean tenure are less diverse than
those with lower mean tenureindependent of the dispersion in tenure. It is possible that
differences in tenure become less salient as average tenure increases; however, it is
equally reasonable to think that tenure is less salient when most team members are recent
arrivals. Thus, rather than assume a particular relationship between mean tenure and the
experience of diversity, we consider the CMD and mean tenure as separate variables so
that we may disentangle empirically the effects of each.Note that similar issues apply to
the coefficient of variation.6
=
I
ji
J
ij kk
jkik
kNN
ttCMD
)1(
|| where tikand tjkindicate to the individual i and jstenure in the organization, and Nkis the number of
individuals in R&D unit k.
8/8/2019 Networks, Diversity and Performance
15/33
12
Network Measures We are primarily concerned with how a units productivity varies by
its level of network homogeneitythe extent to which interaction is concentrated among
members of neighboring cohorts or is spread out to include contact among members of
distant cohorts.7
Clearly, the overall level of interaction among members of a teami.e.,
the density of the teams networkestablishes a baseline for network homogeneity and
thus should be included as a control variable. Thus, we derive two measures from the
based on the networks of communication frequency generated from the question cited
above.8
Density First, network density is the average level of communication between any two
members of unit k,
Density varies from 0 (no relations between team members) to 1 (maximum strength
relations between all team members).
Note that, beyond its role as a control variable for network homogeneity, the
effect of density on performance is of interest because previous research suggests that
this association should be positive. Indeed, as emphasized above, the perspective that
sees diversity as problematic understands this negative effect to be mediated by a decline
in group cohesion. This view is supported by a long line of research that generally finds
a positive effect of cohesion on performance (see Evans and Dion 1991 for review).
Thus, while our results on network homogeneity speak directly to the specific question of
)1(
)max(/
=
kk
I
ji
J
ij
ijkijk
kNN
zz
Density
where zijkis the tie from team member i to teammember j, max(zik) is the largest of is relation
to anyone, and Nkis the number of members in
unit k.
8/8/2019 Networks, Diversity and Performance
16/33
13
whether cross-boundary ties enhance or degrade group performance, the effect of density
has implications for more general questions regarding the types of network structures that
facilitate or hinder success.
Network Homogeneity To measure network homogeneity9, we begin with a measure of
the similarity in organizational tenure between two individuals i and j in unit k, ijk, that
is conditioned on scientist is position in the units tenure distribution:
Next, we measure the strength of relationship between actors, pijkas a proportion of the
total volume of contact in which i engages:
=
I
ji
ijkijkijk zzp /
Using these terms, network homogeneity is a density measure where relations are
weighted by how proximate ego and alter are in their number of years of service in the
organization:
)1(
*
=
kk
I
ji
ijk
J
ij
ijk
kNN
p
NH
The higher a teams network homogeneity, the more concentrated is interaction among
scientists of similar years of service in the corporation; a low score, by contrast, indicates
that the team has achieved a high level of network diversity.10
=I
ji
ijkik
ijkik
ijk
dd
dd
)max(
)max(
where dij is the absolute distance in years of
tenure in the organization and dmaxk is the
maximum distance between actor i and anyother actor in unit k.
8/8/2019 Networks, Diversity and Performance
17/33
14
Control Variables
Task The R&D units under study vary in the type of research they perform. It is
particularly important to control for the type of unit because the productivity measure we
use may vary in salience across such types. Following Cohen et al., 1986, we classify the
units as devoted to basic research (17% of all units), product development (42%), product
improvement (6%), process improvement (24%), and research targeted toward fixing a
specific problem with a product or process (11%). Units are classified in one of these
categories based on the mean response by unit members to a question that asked them to
choose which of these types best describes their unit (Cohen et al.,1
986).
Competition Prior research has indicated that competition in the surrounding market
affects team performance (Ancona and Caldwell, 1992). We measure the
competitiveness of the market context with the managers or unit leaders response to the
follow question:
What is the competitive pressure in this product area?
The manager or unit leader could respond to the question on a four-point scale ranging
from Not keen (uncontested market available) to Prohibitive (any sales increase
highly contested).
Size Finally, team size represents an important control variable. The units vary
considerably in size, ranging from one team, which contains three members to one that
contains thirty-four scientists. Since the dependent variable is a function of the volume
of work produced by group members, it should be significantly related to size. In
8/8/2019 Networks, Diversity and Performance
18/33
15
addition, since the density of a groups networkas well as its degree of network
diversity-- is generally a negative function of its size, it is important to include size as a
control in our analysis.
Results
Table 2 contains summary statistics and table 3 produces a correlation matrix for the
covariates in the analysis. Several relationships in the latter table are noteworthy. First,
we see confirmation of the importance of size as a covariate: large teams have
significantly less dense and less diverse networks than do small teams. Second, note the
insignificant correlation between tenure diversity and network homogeneity (r=.-07).
This reflects the construction of the network homogeneity measure in that it is
conditioned on each actors opportunity for interaction with scientists of distant cohorts.11
Tables 2 & 3 About Here
Table 4 presents the fixed-effects regression results in stepwise fashion. In the
first column, we enter the covariates that capture key control factors-- the size of the
team, the type of research performed by the unit, and the competitive intensity
experienced by the unit. As suggested above, since the dependent variable essentially
measures the volume of output of a unit, the positive coefficient on team size should not
be interpreted as a meaningful effect. However, team size is an important control
variable for that very reason. Research type is measured by a series of dummy variables
for four of the five types with units focused on process improvement treated as a
reference category. We see that units that are oriented towards fixing particular problems
and those engaged in improving an existing product score lower on the productivity
8/8/2019 Networks, Diversity and Performance
19/33
16
measure than do other units. These results make sense in that the elements that form the
core of this measurethe number of many papers, proposals, patents, and reportsare
likely to be less relevant for units that are not engaged in the development of a new
product or process or in basic research. Finally, the finding that units operating in more
competitive environments are more productive corresponds to that found by other
researchers (Ancona and Caldwell, 1992). It may be that competitive intensity
concentrates the attention of researchers; it is also possible that more organizational
resources are directed towards efforts that face greater competition.
Tables 4 About Here
The second model introduces the measures of demographic compositionthe
mean tenure of the group and the degree of diversity in tenure. We see that teams that are
more senior in their membership achieve a higher degree of productivity, which suggests
the importance of experience for performance. However, we see no effect for tenure
diversity. Diversity in firm tenure appears neither to enhance nor to degrade team
performance.
The third model adds density as a covariate. We see a significant positive
relationship, which indicates that teams that average more frequent contact among their
members achieve higher productivity. This result lends support to those who understand
social capital as a function of a groups capacity for collective action (e.g., Coleman
1988; Portes and Sensenbrenner1993). Better communication links among members of a
group enable its members to achieve a greater degree of coordination and hence a level of
performance that is unattainable by teams that are less well-connected.
8/8/2019 Networks, Diversity and Performance
20/33
17
However, the final model in table 4 suggests support for the second view of social
capital as it has been applied to thinking on diversity. In particular, we see that teams
that experience more extensive links among members of different cohorts achieve a
higher level of productivity than teams with high network homogeneity. That is, as
suggested by those who see value in diversity and as reflected in hypothesis 1b, the
formation of links across demographic boundaries-- and the different sets of information,
experiences, and outlooks that such boundaries divideenriches the research process and
promotes greater productivity. By contrast, this result indicates a lack of support for
hypothesis1
a: it does not appear that increased levels of intra-cohort contact improve
team performance. Finally, it is noteworthy that we find a negative effect for network
homogeneity but no effect for tenure diversity. This confirms our strategy of exploring
the impact of the network processes that underlie theories of the diversity-performance
relationship. A direct examination of the frequency of interaction across organizational
cohorts sheds light on this relationship in a way that would have been unattainable had
we merely focused on measures of demographic composition.
Summary and Discussion
We regard the preceding results as an important first step in gaining a better
understanding of the social processes that link the demographic composition of teams and
their performance. As argued above, existing theory on this relationship may be usefully
classified in terms of two views on how social structure affects a teams capacity for
effective actioni.e., its social capital. In particular, corresponding to those who see
social capital as emerging from the dense networks emblematic of close-knit
8/8/2019 Networks, Diversity and Performance
21/33
18
communities, one view on diversity worries that demographic diversity introduces
potential bases for social cleavage, which prevent such cohesion from developing. We
have found support for such a view of social capitalbut not as it applies to diversity. In
particular, we find that R&D units that have more dense networks of interaction achieve a
higher level of productivity than do those with sparse networks. However, we find that
demographic diversityper se has no effect on productivity and that, in fact, teams that
display greater levels of contact within cohorts are less productive than teams where ties
link members of distant cohorts. The latter results reflects the orientation of a second
view on diversity, corresponding to a second perspective on social capital, which
emphasizes the importance of interchange among individuals with a wide range of skills,
information, and experiences, for maximizing a groups capacity for creativity and
effective action.
That we find some support for two very different perspectives on social capital
should not be surprising: both views capture important elements of what it takes for a
task group to achieve success in reaching its goals. A team that does not develop the
network connections among their members that enables it to coordinate effectively
clearly faces an uphill battle. However, when such networks remain concentrated among
homogeneous sets of individuals, the team fails to generate the creative tensions that can
come only from interaction among different individuals.
We hope that our results set the stage for future research on the performance of
organizational teams, which may vary both in their demographic composition and
network structure. In particular, a consideration of three important limitations in the
current study suggests directions that future studies might direct their focus. First, the
8/8/2019 Networks, Diversity and Performance
22/33
19
patterns we find pertain to firm tenure but may not be as relevant for other demographic
variables. In particular, while firm tenure is often a salient characteristic because of the
way it relates to hiring practices and organizational seniority systems (Pfeffer1983), it is
likely that the relationship between diversity, networks, and performance is somewhat
different in the context of such societally-salient traits as gender and race. For instance,
while the benefits of bridging members of distant cohorts appears to outweigh the social
strains that such relations often entail, it may be that the social strains involved in
transgender or interracial ties are more difficult to overcome.
Second, the findings presented here may be limited to R&D teams. As discussed
above, research and development is an area where having communication links with
others who are engaged in similar or related research is critical for achieving success. As
such, it may be that achieving a high degree of network diversity may be less important
in other types of organizational teams.12
Thus, only with similar analyses of other
settings may we build a general theory of the relevant processes.
Finally, while we find that network diversity increases productivity, it may be that
extensive inter-category links have negative implications for other outcomes. In
particular, those who see diversity as problematic stress the conflict that arises from the
introduction of social divisions into a group (e.g., Pelled et al., 1999). Thus, greater
network homogeneity may help reduce the level of conflict among members and thereby
improve its performance on outcomes other than productivity. As Williams and OReilly
(1997, p. 98) emphasize, the key question is whether the enrichment in information and
skill than derives from diversity outweighs the negative consequences that result from
possible increases in conflict. Following the strategy we have proposed and adopted
8/8/2019 Networks, Diversity and Performance
23/33
20
here, we suggest that our understanding of the social mechanisms involved in such
processes must begin with a focus on the social networks that do or not occur within and
across demographic categories.
8/8/2019 Networks, Diversity and Performance
24/33
21
REFERENCES
Adler, Paul S. and Seok-Woo Kwon
1999 Social Capital: The Good, the Bad, and the Ugly. Unpublished manuscript;
Marshall School of Business, University of Southern CaliforniaAllen, Thomas J.
1977 Managing the flow of technology: Technology transfer and the dissemination of
technological information within the R&D organization. Cambridge,Massachusetts: MIT Press.
Allen, Thomas J. and Stephen I. Cohen.
1969 Information flow in research and development laboratories. Administrative
Science Quarterly, 14: 12-19.
Allison, Paul D.
1978 Measures of inequality. American Sociological Review, 43: 865-880.
Aldrich, Howard E. and Catherine Zimmer
1986 Entrepreneurship through social networks. Pp. 3-24 in The Art and Science ofEntrepreneurship, edited by D.L. Sexton and R.W. Smilor. Cambridge, Mass.:
Ballinger.
Ancona, Deborah G., and David F. Caldwell
1992 Demography and design: Predictors of new product team performance.
Organization Science, 3: 321-341.
Bantel, Karen A., and Susan E. Jackson
1989 Top management and innovations in banking: Does the composition of the top
team make a difference? Strategic Management Journal, 10: 107-124.
Burt, Ronald S.
1982 Toward a structural theory of action: Network models of social structure,
perception and action. New York: Academic Press.
1991 Measuring age as a structural concept. Social Networks, 13: 1-34.
1992 Structural Holes: The Social Structure of Competition. Cambridge, Mass.:
Harvard University Press.
Cohen, Bernard P, and Xueguang Zhou1990 Status processes in enduring work groups. American Sociological Review, 51:
179-188.
8/8/2019 Networks, Diversity and Performance
25/33
22
Coleman, James S.
1988 Social capital in the creation of human capital. American Journal of Sociology 94: S95-
S120
1991 Foundations of Social Theory. Cambridge, Mass.: Harvard University Press.
Cook, Karen S., Richard M. Emerson, Mary R. Gillmore, and Toshio Yamagishi.
1983 The distribution of power in exchange networks: theory and experimental
results. American Journal of Sociology 89: 275-305.
Elder, Glenn H., Jr.
1974 Children of the Great Depression. Chicago: University of Chicago Press.
Eisenhardt, Kathleen M., Jean L. Kahwajy, L.J. Bourgeois III.
1997 How Management Teams Can Have a Good Fight. Harvard Business Review,
75: 77-85.
Ely, Robin J.
1994 The effects of organizational demographics and social identity on relationshipsamong professional women. Administrative Science Quarterly, 39: 203-238.
Evans, D.R. and K.L. Dion
1991 Group cohesion and performance: A meta-analysis. Small Group Research, 22:
175-86.
Gabbay, Shaul M. and Roger Th. A.J. Leenders.
1999 CSC: The structure of advantage and disadvantage. Pp. 1-14 in Roger Th.A.J.Leenders & Shaul M. Gabbay. (Eds.), Corporate social capital and liability.
Boston: Kluwer Academic Publishers.
Gabbay, Shaul M. and Ezra W. Zuckerman
1998 Social capital and opportunity in corporate r&d: The contingent effect of contact
density on mobility expectations. Social Science Research 27: 189-217.
Granovetter, Mark S.
1973 The strength of weak ties. American Journal of Sociology, 78: 1360-1380.
1985 Economic Action and Social Structure: The Problem of Embeddedness.
American Journal of Sociology, 91:481-510.
Hannan, Michael T. and Alice A. Young1977 Estimation in panel models: Results on pooling cross-sections and time series.
Pp. 53-87 in David Heise ed., Sociological Methodology 1977 San Francisco:Jossey-Bass.
Hansen, Morten T.
1999 The search-transfer problem: The role of weak ties in sharing knowledge across
organization subunits. Administrative Science Quarterly, 44: 82-111.
8/8/2019 Networks, Diversity and Performance
26/33
23
Jobson, J.D.
1992 Applied Multivariate Data Analysis: Volume II: Categorical and Multivariate
Methods. New York : Springer-Verlag.
Katz, Ralph and Robert Tushman.
1979 Communication patterns, project performance, and task characteristics: Anempirical evaluation and integration in an R&D setting. Organizational Behavior
and Human Performance, 23: 139-162.
Kendall, Maurice and Alan Stuart
1977 The Advanced Theory of Statistics. Vol. 1. 4th
ed. New York: Macmillan.
Lawrence, Barbara S.
1997 The black box of organizational demography. Organization Science, 8: 1-22.
March, James G.
1991 Exploration and exploitation in organizational learning. Organization Science,1: 71-87
McCain, Bruce E., Charles OReilly, and Jeffrey Pfeffer
1983 The effects of departmental demography on turnover: The case of a university.
Academy of Management Journal, 26: 626-641.
OReilly, Charles A. III, David F. Caldwell, and William P. Barnett
1989 Work group demography, social integration, and turnover. AdministrativeScience Quarterly, 34: 21-37.
Pelled, Lisa Hope, Kathleen M. Eisenhardt, and Katherine R. Xin
1999 Exploring the black box: An analysis of work group diversity, conflict, andperformance. Administrative Science Quarterly, 44: 1-28.
Pfeffer, Jeffrey
1982 Organizational demography. In L. L. Cummings and Barry M. Staw (eds.),Research in Organizational Behavior, 5: 299-357. Greenwich, CT: JAI Press.
1983 Organizational demography: Implications for management. California
Management Review, 28: 67-81.
Pfeffer, Jeffrey and Gerald R. Salancik.
1978 The external control of organizations. New York: Harper and Row.
Portes, Alejandro.
1998 Social capital: Its origins and applications in modern sociology. Annual Reviewof Sociology, 24: 1-24.
Portes, Alejandro and Julia Sensenbrenner.
1993 Embeddedness and immigration: Notes on the social determinants of economic
action. American Journal of Sociology, 98: 1320-1350.
8/8/2019 Networks, Diversity and Performance
27/33
24
Riordan, Christine M. and Lynn McFarlane Shore
1997 Demographic diversity and employee attitudes: an empirical examination of
relational demography within work units. Journal of Applied Psychology, 82:
342-358.
Robinson, W.S.
1950 Ecological correlations and the behavior of individuals. American Sociological
Review, 15: 351-357.
Ryder, Norman B.
1965 The cohort as a concept in the study of social change. American Sociological
Review, 30: 843-861.
Shenhav, Yehouda A.
1992 Expected managerial careers with growing and declining R&D establishments.
Work and Occupations, 18: 46-71.
Simmel, Georg.
1955 Conflict and The Web of Group Affiliations. Trans. Kurt H. Wolff and Reinhard
Bendix. New York: Free Press.
Staw, Barry M.
1980 The consequences of turnover. Journal of Occupational Behavior, 1: 253-273.
Tajfel, Henri and John Turner.
1986 The social identity of intergroup behavior. In Stephen Worchel and William G.
Austin eds., Psychology and intergroup relations, pp. 7-24. Chicago: Nelson-
Hall.
Tushman, Michael L.
1977 Technical communication in r&d laboratories: The Impact of Project WorkCharacteristics. Academy of Management Journal 20, 624-645.
Wagner, Gary W., Jeffrey Pfeffer, and Charles A. OReilly, III.
1983 Organizational demography and turnover in top-management Groups.
Administrative Science Quarterly, 29: 74-92.
Williams, Katherine Y. and Charles A. OReilly, III
1998 Demography and diversity in organizations: a review of 40 years of research.
Research in Organizational Research, 20:77-140.
Zenger, Todd R., and Barbara S. Lawrence
1989 Organizational demography: The differential effects of age and tenure
distributions on technical communication. Academy of Management Journal, 32:353-376.
8/8/2019 Networks, Diversity and Performance
28/33
25
Endnotes
1 It is worth noting that, while Pfeffer has been taken as asserting that diversity lowers performance, he has
also suggested that diversity promotes learning. For instance, he describes the benefits of employee
turnover in terms of the different perspective that new blood often brings to an organization (Pfeffer1983,
pp. 325-330; cf., Staw 1980). He also counsels managers to defy the homophily principle by selecting
assistants from different cohorts, thereby improving the managers access to parts of the organization to
which he might otherwise be cut off (Pfeffer1985, pp. 75-76)2 The seeming tension between brokerage and social closure is actually illusory (see Gabbay and
Zuckerman 1998, p.195). By contrast, the information benefits associated with boundary-spanning ties do
indeed come at the cost of a loss of social cohesion, and vice versa (Granovetter1973).3 While a focus on firm tenure has clear advantages, a more complete account of the relationship between
networks, diversity, and performance would involve a consideration of other demographic characteristics
and particularly those that are relevant to major social divisions, such as race and gender. Unfortunately,
the R&D units under study are too homogeneous with respect to these variables to lend themselves to
useful study. Thus, as we emphasize in the discussion below, future research would do well to analyze
populations in which such demographic characteristics, and their interaction with networks and
performance, may be examined.4 In particular, managers were asked to whether the unit had produced 0, 1, 2-5, or more than 5 of each of
the eleven items.5 In the original formulation, the denominator is N2. We use N(N-1) here as the latter is the true number of
dyads in a group.6 In analyses not presented here, we find no difference in our results when the CMD is replaced by the
standard deviation of firm tenurethe numerator of the coefficient of variation.7 It should be noted that our measurement of demographic diversity and network homogeneity capture
cohorts in a rather crude fashion. In particular, the cohort concept is properly considered not as a function
of calendar time but of social time; it groups together individuals who have moved through social structure
in parallel (e.g., Ryder1965; Elder1974). In the present context, identifying such cohorts would require
isolating those tenure ranges that are socially similar. Unfortunately, while an effort was made to follow
the method of Burt (1991) in isolating such cohorts through the network of inter-year relations, the data
under study do not lend themselves to such a strategy. In particular, each firm included in the study
provides insufficient network and tenure data to render reliable cohort categories.8
Results based on friendship networks generated substantially the same results as those we report for thecommunication networks.9 One might think that, rather than introduce an additional variable, hypotheses 1a and 1b may be tested by
examining for interaction effects between diversity and density. That is, if relations across cohorts lower
(raise) performance then there should be a negative (positive) interaction effect between density and
diversity. However, while such an interaction effect is broadly consistent with the hypotheses, it suffers
from the ecological fallacy (Robinson 1950). For instance, while a positive interaction effect would imply
that more dense networks increased the diversity effect, the relations responsible for this effect could be
intercohort or intracohort ties. Thus, in order to distinguish between these possibilities, one must directly
analyze the pattern of ties on either side of demographic boundaries.10 Non-response poses a potential problem for the measurement of tenure diversity, density, and network
homogeneity because these variables require information on all members of a team. In particular, each
non-response produces N-1 missing relations in a network. Fortunately, the response rate for the present
study was excellent: we have full information on 83.8% of all possible network relations, and at least one
response on another14.6%. Following Gabbay and Zuckerman (1997, p.201), we conducted a series of
sensitivity analyses to assess the reliability of the results presented below in the face of different
assumptions about pattern of missing response both to the network items and firm tenure. We also
compared the results using raw network data with networks that were scaled using a loglinear model (see
Gabbay and Zuckerman 1997, p. 200). We found our results to be highly robust to all such transformations
of the data. Details of these tests are available from the authors.11 Alternative measures that do not control for this opportunity set correlate very highly with
tenurediversity, as one would expect.
8/8/2019 Networks, Diversity and Performance
29/33
26
12 The inclusion of several different types of R&D in the present data set might seem to lend itself to an
analysis of how the effects for network density and network homogeneity vary by task type. Some tasks
are more dependent upon the exchange of information and others are more dependent upon diverse
information. We added slope adjustments for network density and network homogeneity for each task
type. None of the adjustments were significant. Indeed, it would seem that, when considered in light of the
full range of possible work tasks, that all types of R&D involve roughly similar issues.
8/8/2019 Networks, Diversity and Performance
30/33
Item Mean SD
Positions Papers. 1.5 1.2
Project proposals. 2.0 1.1
Published scientific/ technical articles. 1.4 1.3
Patents or patent applications. 1.5 1.2
Books (including editorship). .13 .47
Algorithms, blueprints, drawings, etc. 1.4 1.4
Reports which remained within the unit. 2.0 1.3
Reports which circulated outside the unit. 2.5 .93
Experimental prototypes of devices, instruments, components of devices, etc. 1.4 1.3
Experimental materials, e.g., fibers glass, plastics, metals, drugs, chemicals, etc. 1.2 1.3
Prototype computer programs. 1.1 1.3
Table 1: Team Performance
8/8/2019 Networks, Diversity and Performance
31/33
Table 2: Descriptive Statistics
Va r i a b l e
P e r f o r ma n c eB a s i c Re s e a r c hA p p l i e d Re s e a r c hP r o d u c t De v e l o p me n tP r o d u c t I mp r o v e me n t
P r o c e s s I mp r o v e me n tMa r k e t Co mp e t i t i o nA v e r a g e T e a m T e n u r eT e a m Di v e r s i t yT e a m Si z eNe t w o r k De n s i t yNe t wo r k Ho mo g e n e i t y
Mean
0.17.11.42.06
.242.1
10. 68.2
10. 2.541.9
S D
.86
.37
.3149.23
.431.14.93.84.9.181.7
Mi n
-2.00000
00
1.913
.180
Ma x
1. 51111
14
24. 918341
9. 4
8/8/2019 Networks, Diversity and Performance
32/33
1 Performance
2 Basic Research
3 Applied Research
4 Product Development5 Product Improvement
6 Process Improvement
7 Market Competition
8 Average Team Tenure
9 Team Diversity
10 Team Size
11 Network Density
12 Network Homogeneity
These are zero-order correlations. * = p < .10, ** = p < .05, *** = p < .001.
Table 3: Correlations
1
1
-.03
-.07
.004-.09
.12*
.23
.16**
.07
.33***
.001
-.37***
2
1
-.15**
-.38***-.11*
-.26***
-.09
-.08
-.13**
-.14**
.16**
.09
3
1
-.30***-.09
-.20**
-.04
-.02
.02
.05
-.09
-.09
4
1-.21**
-.49***
.07
.03
.04
-.005
.03
.08
5
1
-.14**
.02
.07
.12*
-.09
-.09
.05
6
1
.02
-.002
-.02
-.13**
-.05
-.14**
7
1
.02
.07
.10
-.05
-.09
8
1
.72***
.02
-.08
-.06
9
1
-.01
.004
.07
10
1
-.51***
-.63***
8/8/2019 Networks, Diversity and Performance
33/33
Predictors
Constant
Basic Research
Applied Research
Product Development
Product Improvement
Market Competition
Team Size
Team DiversityAverage Team Tenure
Network Density
Network Homogeneity
Model Fit
N
R-squaredAdj R-squared
I
-.55
-.06
-.38**
.02
-.28
.10**
.04**
211
.41
.29
II
-.88
-.04
-.40**
.02
-.20
.09**
.04**
-.001.03*
211
.42
.30
III
-1.6
-.02
-.34*
.04
-.09
.10**
.06***
-.004.04**
.88**
211
.44
.31
Table 4: Networks and Performance
These are ordinary least squares for variables predicting team performance. The units are clustered
within firms and models are estimated using the AREG procedure in STATA. AREG controls for
mean differences across firms, for predictors and the dependent variable.
* = p < .10, ** = p< .05, *** = p < .001
IV
-.97
-.01
-.33*
.07
-.24
.10**
.03*
-.001.03
.84**
-.14**
211
.47
.35