Content uploaded by Tian Chan
Author content
All content in this area was uploaded by Tian Chan on Jan 08, 2023
Content may be subject to copyright.
When do teams generate valuable inventions? The
moderating role of invention integrality on the effects of
expertise similarity, network cohesion, and gender diversity
Tian Heong Chan (corresponding author)
Goizueta Business School, Emory University, Atlanta, Georgia 30322
tian.chan@emory.edu
Haibo Liu
School of Business, University of California, Riverside, California 92521
haibo.liu@ucr.edu
Steffen Keck
Faculty of Business, Economics, and Statistics, University of Vienna, 1090 Vienna, Austria
steffen.keck@univie.ac.at
Wenjie Tang
Faculty of Business, Economics, and Statistics, University of Vienna, 1090 Vienna, Austria
wenjie.tang@univie.ac.at
Abstract
Research has demonstrated that certain team composition factors—high expertise similarity, high network
cohesion, and mixed-gender teams—have predominantly negative effects on the teams’ invention
outcomes. Yet these factors have also been shown to improve team coordination, which should (in theory)
lead to better invention outcomes. We address this tension by highlighting the need to consider the
invention’s integrality, which increases task interdependencies among team members and thereby
strengthens the positive relationship between team coordination and invention value. We hypothesize that
(i) the main effects of these team composition factors reduce a team’s invention value but, more
importantly, that (ii) invention integrality positively moderates those effects. We support these claims
with evidence from utility patent data filed in the United States during the period 1983–2015.
Keywords: team composition, invention integrality, technical interdependencies, patents, team
coordination
Revision history: Received October 2021; accepted: December 2022 by Glen Schmidt after two revisions.
- 1 -
1. Introduction
Teams are central to inventive work (Hoegl and Gemuenden 2001, Wuchty et al. 2007), yet effective
teamwork does not come easily. Specifically, various aspects of a team’s composition affect the quality of
team processes and thereby invention outcomes. Building on an extensive literature on this topic, we
explore how three commonly studied aspects of team composition—expertise similarity, network
cohesion, and gender composition—affect invention outcomes (Baer et al. 2014, Huckman and Staats
2011, Nielsen et al. 2018, Sosa et al. 2015, Vakili and Kaplan 2020).
Broadly, the empirical literature has found that high expertise similarity, high network cohesion, and
mixed-gender teams are negatively associated with the value of teams’ inventions (Fleming et al. 2007;
Jensen et al. 2018; Singh and Fleming 2010).
1
However, common among these team dimensions is that
scholars have argued that all three encourage (in different ways) the effective team coordination necessary
for successful teamwork. More specifically: high expertise similarity facilitates mutual understanding and
idea integration; high network cohesion can engender stronger behavioral norms, which deter
uncooperative behaviors; and female team members on an otherwise all-male team are conducive to
effective team discussions regarding new information and potential problems (Lingo and Mahony 2010;
Sosa et al. 2015; Tortoriello et al. 2015; Woolley et al. 2010). It is therefore puzzling that these three
dimensions of team composition seem to hinder team invention performance even as they facilitate team
coordination—a key predictor of successful invention.
We address this puzzle by emphasizing the role an invention’s integrality plays, which affects the
extent of interdependence among team members’ tasks (Colfer and Baldwin 2016). The structure of a
technological invention can range from the highly decomposable (or modular) to the very difficult to
decompose (or integral ) (Baldwin and Clark 2000; Fixson and Park 2008; Schilling 2000; Simon 1962;
Ulrich 1995). Whereas modular inventions can be partitioned into small, discrete chunks (Pisano and
Verganti 2008), integral inventions exhibit strong technical interdependencies—which implies that
changes in one part will likely lead to cascading changes in related parts. A critical implication for
teamwork is that, relative to more modular ones, integral inventions require more effective coordination
among team members (Baldwin and Clark 2000; Chan et al. 2021; Kavadias and Sommer 2009; Schilling
2000). Building on this insight, we argue that an invention’s integrality positively moderates the effects of
team composition factors on its performance; hence, teams with high expertise similarity, high network
cohesion, and mixed-gender composition perform better when working on more integral inventions.
1
The evidence on the performance implications of these team composition factors—especially with regard to
mixed-gender teams (see e.g., Nielsen et al. 2018)—varies across contexts and is not fully conclusive. Yet when
focusing on invention and patents (as we do here), the empirical evidence suggests an overall negative correlation.
- 2 -
To test these propositions, we leverage data from the United States Patent and Trademark Office
(USPTO) for patents granted from 1983 to 2015. Patent data contains information on the team, which
allows us to construct measures based on the team’s expertise, social network, and demography. It also
contains a detailed textual description of the invention, which allows us to derive the invention’s structure
(Chan et al. 2021). We measure a patent’s value through market reactions to patent grant announcements
(Kogan et al. 2017). Following prior research on team innovation (e.g., Fleming et al. 2007; Singh and
Fleming 2010), we incorporate firm-level fixed effects and various team-level controls to capture possible
firm- and team-level confounders. We also employ a novel matching technique that matches on (and
places fixed effects on) a large subset of team members, which allows us to create estimates based on
comparisons of inventive performance across patents with marginal changes in team members.
We first provide evidence for the negative main effects of expertise similarity, network cohesion, and
mixed-gender team composition. Our results show that, when working on an invention with an average
level of integrality, a mixed-gender team with high expertise and network overlaps underperforms an all-
male team with little overlap in expertise and networks by 6.3% on average. More importantly, we
theorize and find strong evidence for the moderating effect of invention integrality on all three
dimensions of team composition. Specifically, our result shows that, when working on highly integral
inventions, a team that features all three coordination-facilitating factors outperforms a team that features
none of the three factors by 6.2% on average.
The present paper makes two principal contributions. First, prior results in the innovation literature
have suggested that aspects of team composition such as network cohesion (Fleming et al. 2007; Sosa et
al. 2015), expertise similarity (Huckman and Staats 2011; Vakili and Kaplan 2020), and mixed-gender
composition (Nielsen et al. 2017, Nielsen and Börjeson 2019) can have a negative effect on invention
outcomes. At the same time, however, robust evidence reveals that these factors facilitate within-team
coordination (Lingo and Mahony 2010, Sosa et al. 2015, Tortoriello et al. 2015, Williams and Polman
2015). In reconciling these seemingly contradictory results, the present research highlights the crucial
moderating role of invention structure: the effects associated with certain team composition dimensions
change markedly when teams are working on integral, rather than modular, inventions.
Second, although team composition (Fleming et al. 2007; Huckman and Staats 2011; Nielsen and
Börjeson 2019) and invention integrality (Baldwin and Clark 2000; Fixson and Park 2008; Ulrich 1995;
Yayavaram and Ahuja 2008) are both widely acknowledged—in the domains of operations management
and new product development—as fundamentally shaping innovation processes and invention outcomes,
the related literature has largely remained in separate research streams. Specifically, the literature on
teams and team composition is mostly interested in the type of team characteristics that would lead to
more effective team coordination (e.g., Huckman and Staats 2011) but typically pays less attention to the
- 3 -
nature of the task interdependencies. In contrast, the literature on modularity has argued for the
importance of better organizational communication and coordination when an invention is integral
(Gokpinar et al. 2010, Sosa et al. 2015), yet does not provide systematic insights into the team-level
characteristics that can engender effective coordination. Our research bridges both streams of literature.
We theorize and demonstrate empirically how team composition factors conducive to effective
coordination are especially valuable to teams that work on integral inventions. We remark on the
managerial implication of this result: organizations, within their constraints, can use the insights from our
results to assemble teams that better match the invention’s structure.
2. Hypotheses Development
We begin with an overview of previous research on the three team composition factors of interest. Next,
we introduce the concept of invention integrality and describe how it affects the need for coordination in
teams. For each of these team composition factors, we postulate (1) a baseline hypothesis for the most
likely direct effect on invention value and (2) how this effect could be moderated by invention integrality.
2.1. Team composition and invention integrality: An overview
All three of our focal team composition factors have been argued, based on prior literature, to have both
positive and negative effects. Specifically, overlaps in expertise across team members is strongly
associated with a smaller pool of ideas and knowledge on which teams could draw to develop inventions
(Hargadon and Sutton 1997). Yet such expertise overlaps also facilitate team coordination (Huckman and
Staats 2011). In a similar vein, overlaps in the collaborative network tend to make teams converge too
early or resort in groupthink (e.g., Burt 2004; Fleming et al. 2007; Tortoriello et al. 2015). But scholars
have argued as well that such network overlaps add paths to trust and effective communication thus
resulting in better coordination (Sosa et al. 2015). Finally, the presence of female group members in
typically male-dominated innovation settings could negatively affect inventive outcomes, because the
expertise and ideas female team members provide are more likely to be discounted or ignored than the
contributions of their male counterparts (Joshi 2014). And yet, mixed-gender teams have been shown to
induce more team-oriented interaction processes that can enable better team coordination (e.g., Woolley
et al. 2010). When taken together, this evidence suggests that the effects of team composition would
differ significantly depending on the coordination challenges a particular task posed.
One fundamental factor that determines the need for effective coordination during the innovation
process is the invention’s structure. Generally, technological inventions feature structures ranging from
the highly decomposable to the difficult to decompose (Baldwin and Clark 2000; Fixson and Park 2008;
Schilling 2000; Ulrich 1995; Yayavaram and Ahuja 2008). Whereas the former description characterizes
inventions that can be partitioned into small, discrete chunks (Pisano and Verganti 2008), inventions of
the latter type cannot be so partitioned and hence are essentially “integral” (Ulrich 1995).
- 4 -
Highly decomposable inventions thus comprise several (nearly) separable chunks, where each chunk
consists of tightly coupled components even though the chunks themselves are loosely coupled with each
other (Simon 1962). So when teams work on highly decomposable inventions, roles can be distributed
among team members to operate independently—an approach that minimizes interactions among team
members (Sanchez and Mahoney 1996, Simon 1962, Ulrich 1995, von Hippel 1990). In contrast, an
integral invention is tightly coupled; it cannot be partitioned easily into smaller chunks, and any change in
one component affects many others. As a consequence, working on an integral invention requires
engagement of the whole system (von Hippel 1998), and members must work in concert, sharing
information and perspectives, to ensure success (Sanchez and Mahoney 1996, Simon 1962, Thompson
1967, Ulrich 1995, Ulrich and Eppinger 1999, von Hippel 1990). With integral inventions, then, a change
in one of its parts can cascade through to other parts—requiring other team members to accommodate the
change by reworking their own input. At the same time, if information is not promptly updated or is too
sticky to be conveyed precisely (von Hippel 1998) then the team members responsible for different parts
of the invention may find themselves working with outdated assumptions (Mihm et al. 2003, Skilton and
Dooley 2010). Moreover, greater integrality also implies greater difficulty in assigning responsibilities to
team members; that is, failures become hard to trace, and success is ultimately a function of team effort
and not easily attributable to individuals (Bruns 2013). As a result, ambiguities about who should do what
and about when those tasks should be performed can lead to confusion and even withdrawal (Perry-Smith
and Mannucci 2017). In summary, the foregoing discussion suggests that, as an invention becomes more
integral, the need for effective coordination among team members substantially increases; for inventions
that are mostly modular, team coordination is less essential (Kamrad et al. 2017).
2.2. Average effects of team composition and the moderating role of invention integrality
Having introduced our central theoretical constructs of interest, we next turn to formulating our
hypotheses about the expected baseline effects of expertise similarity, network cohesion, and mixed-
gender composition, as well as the hypothesized moderating effect of invention structure.
Prior empirical evidence in research contexts similar to ours have generally suggested that high
expertise similarity is associated with poorer team inventive outcomes (Hargadon and Sutton 1997; Singh
and Fleming 2010; Uzzi and Spiro 2005). One explanation for such negative outcomes is that high
overlaps in expertise among team members can constrain teams from being able to tap into a greater pool
of diverse ideas—the essential raw materials of innovation upon which creative and novel solutions are
built (Girotra et al. 2010, Hargadon and Sutton 1997, Singh and Fleming 2010, Sosa 2011, Tortoriello
et al. 2015). Furthermore, high expertise similarity may also prevent team members from viewing a
problem from different perspectives, hampering the team’s ability to search for novel solutions (Sutton
and Hargadon 1996). Based on the prior evidence, we thus formulate our first baseline hypothesis:
- 5 -
Hypothesis 1 (H1). Team expertise similarity has a negative average effect on invention value.
Importantly, however, contrary to these negative effects, which arise from a lack of diverse ideas and
perspectives caused by high expertise similarity, prior research has also associated expertise similarity
with more effective team coordination. Because they share a similar background, members in teams with
high expertise similarity tend to employ similar vocabularies, thereby improving mutual understanding by
facilitating the interpretation of information (Gilson and Shalley 2004). Moreover, having similar
perspectives on the task at hand facilitates finding common ground and avoiding unproductive conflicts
(Gebert et al. 2007, Kurtzberg and Amabile 2001), and thus helps the team reach consensus on the final
solution (Ancona and Caldwell 1992, Bercovitz and Feldman 2011, Lovelace et al. 2001). As a result,
teams characterized by higher levels of expertise similarity usually find it easier to synthesize different
pieces of information and knowledge into a coherent solution (Hargadon and Bechky 2006; Harvey 2014;
Lingo and Mahony 2010; Liu et al. 2018). Such teams are thus better at coping with the need—in the case
of integral inventions—to continuously integrate and coordinate the work of individual team members.
As we have argued, the benefits that arise from improved coordination in a team with high expertise
overlaps are not expected to be the same for all types of inventions. More specifically, the ability to
coordinate activities effectively will not provide much benefit to teams working on modular inventions,
for which coordination requirements are minimal. Therefore, for modular inventions, the previously
predicted negative effects of high expertise similarity should thus prevail. However, when teams work on
highly integral inventions, and thus the quality of coordination efforts becomes a crucial determinant of
team effectiveness, we can expect the positive effect of expertise similarity to become more pronounced
and perhaps even to outweigh any potential drawbacks.
In short, we expect that invention integrality accentuates the benefits of (i.e., positively moderates the
effect of) team expertise similarity on invention value. Overall, this is consistent with insights that
Huckman and Staats (2011) suggested, as they studied the effect of team expertise similarity in the
context of software development. These authors showed that the frequency of task change (which they
used as a lever to generate increased coordination requirements; we use an invention’s integrality)
similarly accentuates the benefits of team expertise similarity on software development outcomes. These
arguments and prior findings thus lead us to our first moderation hypothesis:
Hypothesis 1m (H1m). Invention integrality positively moderates the effect of team expertise
similarity on invention value such that the effect of expertise similarity will be less negative, and
possibly positive, when teams work on highly integral inventions.
Turning to network cohesion, the empirical evidence suggests overall that high network cohesion will
predict poorer inventive outcomes (Burt 2004; Hargadon and Sutton 1997; Singh and Fleming 2010; Uzzi
- 6 -
and Spiro 2005). The negative effects are likely to occur for several reasons. First, high network cohesion
among collaborators—in the sense of team members having high overlaps (i.e., common third parties) in
their work connections—can hinder a team’s access to novel and creative ideas. This is because team
members with common outside connections tend to provide redundant information (Burt 2004; Fleming
et al. 2007). Second, teams with high network cohesion may also be more susceptible to groupthink. This
occurs because the existence of common third parties could exert social pressure on individual team
members to converge quickly on a solution that is deemed appropriate within the circle of common
collaborators rather than first engaging in a prolonged period of divergent thinking to develop novel ideas
(Perry-Smith and Shalley 2003). Hence, our second baseline hypothesis (H2) predicts negative average
effects of high team network cohesion on invention value.
Hypothesis 2 (H2). Team network cohesion has a negative average effect on invention value.
And yet, high network cohesion has likewise been shown to facilitate team coordination. Common
third-party connections can serve as added pathways for communication and can motivate members to
invest time, energy, and effort in sharing knowledge (Tortoriello et al. 2015). Such connections have
consistently been shown to facilitate the communication of technical information, especially when
individuals work on an interdependent task (Sosa et al. 2015). High network cohesion can also help
enforce behavioral norms in teams. When members of the current team have worked with the same set of
past collaborators, a clique tends to emerge, along which information about a person’s behaviors (or
misbehaviors) spreads rapidly. The presence of such relationships can be a strong deterrent to undesirable
actions, such as free-riding or exploitive behavior (Rahmani et al. 2017). However, our previous logic
implies that if a team is working on highly integral inventions and therefore effective communication and
coordination are essential for high performance, then the benefits of effective team coordination—and
thus of high network cohesion—should be amplified, possibly enough to outweigh existing drawbacks.
This reasoning leads to our second moderation hypothesis:
Hypothesis 2m (H2m). Invention integrality positively moderates the effect of network cohesion
on invention value such that the effect of network cohesion will be less negative, and perhaps
even positive, when teams work on highly integral inventions.
Finally, we address the role of team gender composition. Previous studies in the innovation literature
have yielded mixed results regarding the effect of gender diversity (for an overview, see e.g., Nielsen
et al. 2018). Yet some prior evidence also suggests that, in the domain we consider (patented inventions),
female inventors might experience specific obstacles that could negatively affect the performance of their
teams. Specifically, prior work has identified the extent to which a field is male-dominated as a contextual
- 7 -
factor to consider when assessing the overall effect of gender diversity (Fernandez-Mateo and Kaplan
2018). Often, women’s expertise in highly male-dominated fields is not fully utilized, which causes
mixed-gender teams to exhibit lower innovation performance (Joshi 2014). The science and technology
domain—the focus of the present paper—clearly constitutes such a male-dominated area, despite women
having gradually gained more representation. For example, the USPTO (2019) reported that only 12% of
inventors in 2016 were women and that inventor teams with at least one woman were responsible for only
21% of all team-generated patents (vs. 79% from all-male teams) in that year.
Note that the association between female inventors and lower performance has been shown in prior
work, notwithstanding the strong evidence that female inventors are no less qualified than their male
counterparts (Hoisl and Mariani 2017). It is certainly possible that, despite being equally qualified, female
inventors might face gender-specific barriers in innovation settings; if so, then innovation teams with
female members may well exhibit lower performance. Supporting this interpretation is that most large-
scale patent-based studies—which usually focus on the performance and productivity outcomes of
individual inventors—report that female inventors experience more negative outcomes in their work
(Ding et al. 2006). Moreover, this gender effect may extend to team settings. Jensen et al. (2018) used a
pooled sample to show that female inventors (either individually or in all-female teams) tended to face
more difficulties during patent prosecution. Jensen and colleagues (2018) documented that patent
examiners were more likely to narrow the claims of female inventors (vis-à-vis individual men’s or all-
male teams’ patent applications), thereby reducing their scope and value. The argument developed here
suggests an overall negative relationship between the presence of female members in teams and invention
outcomes. Thus, we have our third baseline hypothesis:
Hypothesis 3 (H3). Mixed-gender team composition has a negative average effect on invention
value.
In contrast to these negative effects, previous research frequently has shown that the addition of
women to an otherwise all-male team may result in team members making a mental shift toward more
team-oriented norms and toward a psychologically safer atmosphere, which greatly facilitates effective
coordination. Notably, a large body of social psychology research has found that women are more likely
than men to exhibit higher levels of interpersonal sensitivity; that is, they pay more attention to and show
more respect for other people’s feelings and thoughts (Fletcher 1998; Hall 1978), resulting in more
constructive team processes in teams with a higher proportion of female members (Woolley et al. 2010).
The presence of female team members can also alter the behavior of their male peers. For example,
having female team members can lead men to behave in a more caring, generous, and helpful manner to
other team members of both genders (Keck and Tang 2018; Williams and Polman 2015). Such a
- 8 -
supportive environment encourages team members to speak up when noticing problems (Gong et al.
2012, Siemsen et al. 2009), to offer and accept constructive criticism when there are disagreements
(Bradley et al. 2012), and to benefit more from feedback (Edmondson 1999). All of these qualities makes
it easier for team members to coordinate their activities. The logic of our two previous moderation
hypotheses implies that these positive effects on team processes should be especially valuable when the
need for effective coordination is high, as it is when the team works on an integral invention. This
reasoning is summarized in our third moderation hypothesis. Figure 1 summarizes our theoretical
discussion.
Hypothesis 3m (H3m). Invention integrality positively moderates the effect of mixed-gender
team composition on invention value such that the effect of mixed-gender team composition will
be less negative, and possibly even positive, when teams work on highly integral inventions.
Figure 1: Overview of Theoretical Framework
3. Empirical Context
Testing our theory requires detailed information at the level of expertise and network cohesion as well as
demographic information on invention teams. We must also be able to identify and compare the structure
and value of the inventions. Following prior work on invention teams, we leverage the richness of patent
data to deliver this insight. Specifically, we combine analyses that focused on team composition (see e.g.,
Fleming et al. 2007; Hoisl and Mariani 2017; Singh and Fleming 2010), measuring invention integrality
(Chan et al. 2021, Fixson et al. 2017), and the market valuation of patents (Kogan et al. 2017).
Our data set is based on utility patent information published online by the USPTO. Utility patents
grant intellectual property (IP) rights to the creators of an invention of a new and useful process, machine,
manufacture, or composition of matter (USPTO 2010). Ever since patent rights were strengthened in 1982
when the Court of Appeals for the Federal Circuit (CAFC) was established, patents have been
indispensable for inventors seeking to protect their creative ideas (Bessen 2008). Importantly, unlike
many countries where “utility” refers to IP rights conferred to minor inventions, utility patents in the US
Expertise similarity
Network cohesion
Gender composition
Invention value
_
Invention Integrality
+ Hypotheses 1m, 2m, 3m
Hypotheses 1, 2, 3
- 9 -
require inventions to clear the dual bars of novelty and non-obviousness, making a US patent equivalent
to what is known as an “invention patent” in international settings (USPTO 2010).
We have electronic patent data for all patents granted from 1976 through 2018. Whereas we use the
entire data set to construct backward and forward measures, we focus on those patents filed and granted
between 1983 and 2015. Our sample is therefore unaffected by legal changes to the US patent system
stemming from establishing the CAFC in 1982, although it still ensures a sufficiently long period over
which to calculate backward and forward measures.
Among the 5.2 million patents the USPTO granted during this period, we start by focusing on a
subset of 3.1 million patents granted to teams (two or more members). To improve comparability, when
constructing team-based measures we further limit our data to patents from teams whose members (1) are
all located in the United States and (2) have previous patenting experience; these restrictions reduce the
number of patents to 1.1 million. Our final sample consists of 468,023 patents granted to US public
corporations listed on a major US stock exchange, from which we can derive each patent’s market value
and company information. To obtain the relevant company information, we match the patent data to firm
identifiers in Compustat and the Center for Research in Security Prices using curated matching results
from the National Bureau of Economic Research Patent Data Project (Hall et al. 2005), which covers
patents from 1976 through 2006. For the 2007–2015 data, we use Bessen’s (2008) name-matching
algorithm to match names between the two databases. To ensure accuracy, this procedure is followed by a
close visual inspection of all non-exact name matches.
In sum, using utility patents uniquely allows researchers to create insights by connecting team
composition, invention integrality, and invention value into a single dataset. We leverage established
methodologies to construct our measures. Yet, using utility patents also requires care and carries some
drawbacks. For example, not all firms across all industries patent their inventions. By focusing on patents,
our insights apply particularly to large firms, especially those in the manufacturing and information
technology sectors, where patents closely reflect inventive activities (Moore 2005). More studies may be
needed to assure generalizability to other sectors. We discuss additional possible limitations of patent data
in the Discussion (see Section 6), and where possible, perform analyses to assuage them (see Section 5
and Online Appendix III).
3.1. Dependent variable: A patent’s value
Our measure of a patent’s value is an estimate of the amount of financial value added to the firm. Kogan
et al. (2017) showed how one can estimate the value that an invention brings to a firm via the stock
market reaction to its patent being granted. This approach exploits the unique characteristics of the
USPTO patent-granting process. In particular, patents are granted every Tuesday and, on that same day,
the USPTO publishes the Official Gazette—its official journal that lists all issued patents and presents
- 10 -
their details. Given that the USPTO controls the granting process and that the public does not know in
advance whether (or when) a patent will be issued for an invention, the patent grant event is an
unanticipated “information shock” to the public. From a theoretical perspective, therefore, the market
movements in the days following a patent grant can be used to infer the value of the granted patent.
Kogan et al. (2017) leveraged the patent information event to develop a method for deriving a
patent’s value based on the cumulative abnormal return (the return of a company’s stock minus the return
of the market index) after a patent is granted. The choice of a three-day window is based on the number of
days during which a value signal due to the patent grant can be observed empirically. Kogan et al. (2017)
documented a significant increase in market reaction—measured in terms of both trading volume and
stock price movements—in the three days after the focal firm is granted a patent. The value signal
dissipates after three days; that is, the stock price movements after three days become too noisy due to the
influence of other events. Nonetheless, this transitory value signal can be exploited effectively to derive
the value of a patent; with few exceptions, studies that derive individual patent values from market signals
focus on windows of just a few days (Boscaljon et al. 2006, Desyllas and Sako 2013, Kline et al. 2019).
Likewise, our results are robust to using either a two- or four-day window.
Our dependent variable, LogValue, is the log of the patent’s value (in thousands 1983 US dollars).
2
This method—which we describe in detail in Online Appendix I—yields a direct and continuous estimate
of an invention’s value to the firm. In Online Appendix III, we proxy an invention’s value using another
widely adopted measure—forward citations—that may better reflect an invention’s value over the long
term (i.e., in years). Our results are robust to using this alternative dependent variable.
3.2. Independent variables: Team composition
We now turn to discussing the construction of our independent variables on team composition:
Expertise similarity. Our first team-level variable, ExpertiseSim, is the degree to which team members
share similar expertise. We first construct a team member i’s expertise profile Pi based on that member’s
past patenting experience. Thus Pi is a vector reflecting the counts of patents in each Cooperative Patent
Classification (CPC) class that team member i has previously patented. We then measure the expertise
similarity between team members i and j using the cosine similarity index (Pi Pj / |Pi | |Pj |) (see e.g., Van
Alstyne and Brynjolfsson 2005). If the team has more than two members, we assess the team’s overall
expertise similarity by averaging the pairwise similarity indices.
Network cohesion. Our second team-level variable, NetworkCohesion, is the team members’ number
of common third parties. We measure this by first identifying persons who are not part of the current team
2
We use a log formulation (or a “log + 1” when a variable’s lower bound could reach zero) to adjust for skewness in
the distribution of variables, and the name of each log-transformed variable has the prefix Log. The exact form used
for each variable is given in Table 1 (see Section 3.5).
- 11 -
but who have previously collaborated with any pair of team members. Then, NetworkCohesion is (the log
of ) the number of common third parties. We double-count such individuals if they appear as a common
third party within multiple pairs of team members—that is, to account for the possibility that such a
person strengthens the cohesiveness across multiple pairs of team members.
Gender composition. Our last team-level variable, MixedGender, is an indicator set to 1 if the patent
is granted to a team consisting of both male and female inventors (and set to 0 otherwise). Following prior
research (Ding et al. 2010), we identify the gender of inventors by their first names. Specifically, we rely
on gender-api.com (an online platform that infers gender from first names) to identify female inventors.
In line with observations the USPTO reported, most patents (77%) are granted to all-male teams; mixed-
gender teams account for 23% (include a negligible 0.4% of all female teams) of all patents in our data.
Given the rarity of all-female teams in this data set (and the focus of our theorizing), MixedGender
effectively identifies mixed gender teams against all-male teams. We derive consistent findings when we
analyze, in Online Appendix III, how the number of female team members affect team performance.
3.3. Moderator: Invention’s integrality
An invention is more integral if it cannot be partitioned easily into “chunks”, where each chunk consists
of tightly coupled components, yet the chunks themselves are loosely coupled (Simon 1962). Our
approach follows Chan et al. (2021) in evaluating an invention’s number of chunks based on the formal
claims made in the patent documentation. In a utility patent, such claims are central because they identify
the invention’s novel aspects. Each claim identifies a single idea of the invention that defines the extent of
the intellectual property to be protected (Lanjouw and Schankerman 2004). For this reason, the language
of a patent’s claim section is highly structured and mechanical (Faber 2001). Specifically, each claim
must be made in exactly one sentence, is uniquely numbered, and must begin with a “subject matter” (the
chunk)—which can be a process, machine, manufacture, or composition of matter (USPTO 2010).
A claim’s subject matter points to the novel aspect of a specific part of the invention independently of
its other parts (Faber 2001). Hence, the number of distinct subject matters can proxy for the invention’s
number of (loosely) coupled chunks. Consider the following two inventions of a playground toy structure
(an example borrowed from Chan et al. 2021), the patents for which each make four different claims.
Patent 5387165 was awarded for the invention of a structure of interconnected tubes through which
children climb or crawl. In the patent, claims 1 and 2 discuss the alternative set-ups of the “connective
junction box;” claim 3 addresses the “connecting means;” and claim 4 describes a “simulated play control
mechanism”—that is, a plug-in driving wheel for simulated driving play (Chan et al. 2021, p.1010).
Claims 1 and 2 are ideas related to the same subject matter (and hence fall into a single chunk), whereas
claims 3 and 4 cover separate subject matters. Thus, the patent highlights three different chunks of the
invention, each of which can exist relatively independently; therefore, the invention is relatively more
- 12 -
decomposable (or less integral). Patent 4629182, for the invention of an inflatable toy tunnel, also makes
four claims. Although this invention similarly contains a structure, sections, and supports, it is less
decomposable (or more integral) in that all four claims discuss ideas related to the same subject matter:
the “inflatable toy tunnel.” Hence, this invention consists of just a single chunk.
The rigid structure of a patent’s claim section allows for a straightforward textual approach to
identifying the different subject matters covered by each patent’s claims (see Online Appendix I for
details). Next, we count the number of distinct subject matters across all claims. Finally, we invert the
measure (by taking the negative of its logarithm) so that it reflects the extent of an invention’s integrality;
the result is our Integrality variable.
We remark that our results are robust to alternative operationalizations of integrality (see Section 5.2).
In addition, we acknowledge that the team could influence an invention structure, so that invention
integrality and team characteristics could be correlated. Some such correlation would not be an issue,
although very high correlations could lead to empirical problems. There are many constraints that prevent
teams from freely changing an invention’s structure (Colfer and Baldwin 2016, Kamrad et al. 2017), and
in Section 5.3 we show that our results are not affected by this concern.
3.4. Control variables
We control for variables at the invention, team, and firm levels to model potential confounders that could
affect patent value. At the invention level, we are concerned primarily with constructs that could be
confounded with the invention’s integrality and its effect on the invention’s value. That is, we are
concerned with the possibility that the observed effects are driven by other aspects of the invention—for
example, the invention’s scope or the breadth of the knowledge base from which the invention draws. To
address this concern, following Lanjouw and Schankerman (2004), we first count the number of claims in
the patent (as a proxy for the patent’s scope) and control for LogPatentClaims (log of the number of
claims). Second, we consider the number of past ideas from which the patent draws; in this way, we also
measure and control for LogPatentCites, or the log of the number of backward citations of the focal patent
(Lanjouw and Schankerman 2004).
At the team level, our main models account for team capabilities by way of several variables used in
previous research. In particular, the models incorporate TeamSize, or the team’s number of inventors,
because larger teams tend to garner more resources (Singh and Fleming 2010). Teams may also be
heterogeneous in terms of their network resources, and innovative capability may be correlated with a
team’s ability to assemble a collaborative network (Lovejoy and Sinha 2010). We account for such
differences by controlling for LogNetworkSize, measured as the focal team members’ total number of
distinct past collaborators. Note that we also account for the focal team’s experience and past successes
- 13 -
by using, respectively, the (log of the) total number of distinct patents previously granted to any member,
LogExperience, and the total value of those patents, LogPastPatentValue.
3
Because a team’s success also depends on firm backing, we account for a variety of firm-level factors
that might drive the team’s success (Alan et al. 2014). We include, first, factors that account for the firm’s
scale: the amount of assets held by the firm, LogAssets, and its working capital, Wcap. We also include
the firm’s operating characteristics. These include: the firm’s research and development expenditures,
LogR&D; sales and general administrative expenditures, LogSGA; capital expenditures, LogCapEx; total
inventory level, LogInvt; and total receivables, LogRect. Finally, we consider the firm’s profitability in
terms of its operating margin, or OIADP/Sale. We also apply firm-level fixed effects in all our models to
account for unobserved, time-invariant differences across firms that might affect patent value. It follows
that the identification of our model is driven largely by within-firm differences (Wooldridge 2010).
Our main models therefore assume that the team’s innovative capabilities are sufficiently modeled by
these team- and firm-level variables. In Section 5.1, we consider more stringent fixed effects and
matching models to capture finer unobserved differences across teams that could confound our results.
However, all our findings are robust to these specifications.
3.5. Summary statistics
Because our analyses focus on variations in teams and inventions, in Table 1 we present descriptions and
summary statistics of the patent-level variables. (Descriptions and summary statistics of firm-level
variables are given in Online Appendix II.) Table 1 reveals that the average team is small, comprising
3.47 members on average. This relatively small team size allows us to maintain consistency with (and
extend the insights of) prior literature, which has similarly focused on small teams (Keck and Tang 2018;
Keck and Tang 2021; Steffens et al. 2011; Woolley et al. 2010; Wu et al. 2019). Whereas large teams
have more complex dynamics, by focusing on small teams, we can better summarize team characteristics
along a few dimensions (viz., how the team members’ expertise and networks overlap, as well as its
gender composition; we discuss this point in more detail in Section 6).
Online Appendix II presents the full correlation matrix across all variables. We note here some
observations from the matrix. First, there is a high correlation between Integrality (a key independent
variable) and LogPatentClaims (a control variable); there are also high correlations between the control
variables LogNetworkSize, LogExperience, and LogPastPatentValue. With respect to the first-named
3
Individuals’ experiences and successes affect the likelihood of collaboration and are used therefore to help correct
for possible selection biases. Creating such historical measures requires that we disambiguate inventor data—that is,
whether the inventors listed on two different patents with the same or similar names are the same individual. Our
main analysis uses the output from Balsmeier et al. (2018), noting that our results are robust to using an alternate
identification (due to USPTO 2019); see Online Appendix III.
- 14 -
correlation, we show in Section 5.2 that our results are robust to using two alternative measures of
integrality: a finer concentration-based measure and a broader network-based measure (using Newman’s
modularity Q; Fixson et al. 2017, Newman 2006). Both approaches reduce the magnitude of correlation
between integrality and the logged number of claims from 0.8 to 0.6, and our results remain robust. On
the high correlation between the control variables: because each of our controls could, in theory, affect an
invention’s value, their exclusion could yield a biased model (Greene 2008); hence, they are retained in
the regressions. Yet their inclusion could inflate our estimated standard errors of the model, which means
that some effects that are actually present may seem statistically weaker (Greene 2008). We establish in
Section 4 (see Table 2 Model 5) that our findings are robust to excluding all control variables.
Table 1: Summary statistics (N = 468,023)
Variable
Description
Mean (SD)
LogValue
Log of patent value inferred from 3-day cumulative abnormal returns
(thousands of 1983 US dollars)
9.00 (1.27)
Integrality
Negative log of the number of chunks (distinct subject matters) in the
patent claim
–2.04 (0.59)
ExpertiseSim
Average cosine similarity of the experience set (i.e., technology
classes) across inventors on the team
0.65 (0.32)
NetworkCohesion
Log of the number of persons who are not part of the current team
but who have previously collaborated with at least two of the team
members
1.60 (1.45)
MixedGender
Indicator variable set to 1 only if the team is a mixed-gender team
0.24 (0.42)
TeamSize
Number of inventors listed on the patent
3.47 (1.78)
LogNetworkSize
Log of the total number of the team’s distinct past collaborators
3.20 (1.13)
LogExperience
Log of the total number of distinct patents any of the team's inventors
has filed
3.26 (1.25)
LogPastPatentValue
Log of the average value of past patents any of the team's inventors
has filed
4.70 (1.66)
LogPatentClaims
Log of the number of claims in the patent
2.74 (0.72)
LogPatentCites
Log of the number of backward citations of the patent
2.62 (1.04)
Note: SD = standard deviation.
4. Empirical Results
Besides controlling for several observable time-varying confounders as described in Section 3.4 (control
variables), all our models incorporate fixed effects for the firm, the patent technology class (given by the
CPC), and the patent filing year. In this way, we control for unobserved heterogeneity in the value of
patents across firms, technologies, and time. All variables with interactions are demeaned to aid
interpretation of the coefficients for non-interacted variables (i.e., the effect of a unit change of the non-
interacted variable when all other variables are held at the population mean). All coefficients are
generalized least-squares (GLS) estimates reported with standard errors clustered by firm. Table 2 reports
the results of our analyses. In Model 1, we consider a model with no interaction variables; this model
allows us to consider the average main effects of various measures of team composition on an invention’s
- 15 -
Table 2: Team composition, invention integrality, and patent value—
GLS model predicting LogValue (N = 468,023)
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
ExpertiseSim(dm)
–0.025***
–0.025***
–0.025***
–0.025***
–0.014
–0.025***
(0.007)
(0.007)
(0.007)
(0.007)
(0.012)
(0.007)
NetworkCohesion(dm)
–0.005*
–0.005*
–0.005*
–0.005*
–0.007*
–0.005*
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
(0.003)
MixedGender(dm)
–0.017***
–0.017***
–0.017***
–0.017***
–0.001
–0.017***
(0.005)
(0.004)
(0.005)
(0.005)
(0.006)
(0.005)
ExpertiseSim(dm) ×
Integrality(dm)
0.035***
0.022*
0.022*
(0.010)
(0.011)
(0.010)
NetworkCohesion(dm) ×
Integrality(dm)
0.011***
0.008+
0.009**
(0.003)
(0.004)
(0.003)
MixedGender(dm) ×
Integrality(dm)
0.018**
0.020**
0.013*
(0.006)
(0.006)
(0.005)
Integrality(dm)
0.011
0.011
0.011
0.011
0.002
0.011
(0.010)
(0.010)
(0.010)
(0.010)
(0.006)
(0.010)
Control variables
TeamSize
0.009***
0.009***
0.009***
0.009***
0.009***
(0.001)
(0.001)
(0.001)
(0.001)
(0.001)
LogNetworkSize
0.020+
0.020+
0.020+
0.020+
0.020+
(0.012)
(0.012)
(0.012)
(0.012)
(0.012)
LogExperience
–0.233***
–0.233***
–0.233***
–0.233***
–0.233***
(0.021)
(0.021)
(0.021)
(0.021)
(0.021)
LogPastPatentValue
0.185***
0.185***
0.185***
0.185***
0.185***
(0.015)
(0.015)
(0.015)
(0.015)
(0.015)
LogPatentClaims
0.008
0.008
0.008
0.008
0.008
(0.012)
(0.012)
(0.012)
(0.012)
(0.012)
LogPatentCites
–0.005
–0.005
–0.005
–0.005
–0.005
(0.006)
(0.006)
(0.006)
(0.006)
(0.006)
Firm-level controls
Yes
Yes
Yes
Yes
No
Yes
Firm FE
Yes
Yes
Yes
Yes
Yes
Yes
Filing Year FE
Yes
Yes
Yes
Yes
Yes
Yes
Product-class FE
Yes
Yes
Yes
Yes
Yes
Yes
Within-R2
0.12
0.12
0.12
0.12
0.00048
0.12
F
32.4***
31.2***
30.6***
30.4***
3.77***
28.1***
Notes: Standard errors (in parentheses) are clustered by firm. FE = fixed effects. Firm-level controls include firm assets,
working capital, R&D expenditures, sales and general administrative expenditures, capital expenditures, total inventory
level, total receivables, and operating margin.
+p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001
value. We find a significantly negative coefficient for ExpertiseSim ( β = −0.025, p < 0.001). Given that
the standard deviation of ExpertiseSim is 0.32, this coefficient implies that a single standard-deviation (1-
SD) increase in the team’s expertise similarity would result in 0.025 × 0.32 = 0.8% decline in the
invention’s value. There is also a significant negative coefficient for NetworkCohesion ( β = −0.005, p <
0.05). Because the standard deviation of NetworkCohesion is 1.45, the reported coefficient implies that a
1-SD increase in the team’s expertise similarity would result in 0.005 × 1.45 = 0.7% decline in the
- 16 -
invention’s value. Finally, we find that the coefficient for MixedGender is significantly negative ( β =
−0.017, p < 0.001). This result indicates that mixed-gender teams produce inventions that are 1.7% lower
in value, on average, than those of all-male teams.
These main effects replicate extant findings on patenting teams (e.g., Jensen et al. 2018, Singh and
Fleming 2010) documenting that team composition factors, which theoretically could improve team-level
coordination, can lead to poorer team outcomes on average. This is because average inventive work could
be mostly modular, and hence do not require much coordination in the first place (Fleming and Sorenson
2001). As a result, other factors beyond coordination dominate team performance (see Section 2.2).
Our key arguments—that the effects of team composition on invention value depend strongly on the
invention’s modularity—are tested in Models 2–6 of Table 2. Models 2–4 test these moderation
hypotheses separately; Model 5 (excluding all control variables) and Model 6 (including all control
variables) test them jointly. All the models provide strong support for all our hypotheses. Focusing on the
most complete model (Model 6), we first find that the coefficient for ExpertiseSim × Integrality is
positive and significant ( β = 0.022, p < 0.05), providing empirical support for H1m: teams with greater
expertise similarity benefit more when working on more integral inventions. Moreover, the coefficient for
Integrality × NetworkCohesion is positive at β = 0.009 ( p < 0.01), supporting our argument in H2m that a
cohesive past collaboration network offers information conduits that increase performance when an
invention is integral. Finally, the coefficient for MixedGender × Integrality is also positive and significant
( β = 0.013, p < 0.05). This finding indicates that, in line with H3m, mixed-gender teams perform
relatively better than do all-male teams when working on more integral inventions.
We can understand the nature of these moderation effects better by examining Figure 2. Panels A, B,
and C of the figure plot (respectively) the marginal effects between: (1) a team with “low” expertise
similarity (at 1 SD below the mean) and a team with “high” expertise similarity (at 1 SD above the mean);
(2) a team with “low” and “high” network cohesion (similarly defined at 1 SD below and above the mean,
respectively.); and (3) an all-male team and a mixed-gender team. We plot these marginal effects against
inventions that are “modular” (i.e., inventions consisting of 23 chunks, which is at the 99th percentile of
all inventions), those that are of “average” integrality (inventions consisting of eight chunks), and that are
“integral” (inventions consisting of a single chunk). Across all these models we see that, whereas teams
featuring high expertise overlap, cohesive networks and/or mixed gender tend to underperform both on
modular inventions and on inventions of average integrality, but that disadvantage is substantially erased
when such teams work on integral inventions.
The analyses in this section show that the effects of each team composition factor on performance
varies with the invention’s integrality. Even though teams with increased expertise similarity, increased
network cohesion, and a mixed-gender composition tend to exhibit lower performance when working on
- 17 -
an invention of average integrality, the coordination challenges imposed by integral inventions accentuates
the benefits of team coordination—to the extent that the negative effects just cited disappear when a team
works on more integral inventions. And because our hypotheses are not mutually exclusive, combining
these separate advantages results in a substantial overall advantage. We plot the total marginal effects
based on our full model (Model 6) in Panel D of Figure 2. As the graph shows, teams that feature all three
coordination-facilitating factors suffer from a total net loss of 6.3% ( p < 0.001) versus teams with none of
them—when working on an invention of average integrality (8 chunks). However, they enjoy a total net
gain of 6.2% ( p < 0.05) when working on integral inventions (1 chunk). In other words, when working on
integral inventions, a mixed-gender team with high expertise and network overlaps should significantly
outperform (by an average of 6.2%) an all-male team with little overlap in expertise and network. That
difference corresponds to an increase of $560,000 (in 1983 US dollars) in patent value.
Figure 2: Marginal effects of team expertise similarity (Panel A), network cohesion (Panel B), a
mixed-gender team (Panel C), and these effects combined (Panel D) on the log of patent value. Error
bars represent 95% confidence intervals.
-0.05
-0.03
-0.01
0.01
0.03
0.05
Modular
(23 chunks) Average
(8 chunks) Integral
(1 chunk)
The invention's integrality
Panel A: Marginal effects of expertise
similarity
-0.09
-0.06
-0.03
0.00
0.03
0.06
0.09
Modular
(23 chunks) Average
(8 chunks) Integral
(1 chunk)
The invention's integrality
Panel B: Marginal effects of network
cohesion
-0.05
-0.03
-0.01
0.02
0.04
Modular
(23 chunks) Average
(8 chunks) Integral
(1 chunk)
The invention's integrality
Panel C: Marginal effects of mixed-
gender teams over all-male teams
-0.18
-0.12
-0.06
0.00
0.06
0.12
Modular
(23 chunks) Average
(8 chunks) Integral
(1 chunk)
The invention's integrality
Panel D: Total marginal effects
- 18 -
5. Additional Analyses
The additional analyses aim to address possible endogeneity threats to our main analysis. We consider
possibly unmodelled variables at the team level and apply more stringent fixed effects on the team in
Section 5.1. We show alternative measures of integrality in Section 5.2. We examine multicollinearity
issues (because teams might affect an invention’s integrality) in Section 5.3. Finally, we provide analyses
and discussion to other potential measurement issues in Online Appendix III. All our analyses are robust.
5.1. Partial team-member fixed effects
Our main models control for a variety of team-level heterogeneities and also include firm-level fixed
effects, which account for unobserved time-invariant differences across firms. Yet as discussed in
Section 3.4 (control variables), a central concern of team-level observational studies is the presence of
unmeasured heterogeneities in team members’ capabilities. Such unmeasured factors could cast doubt on
our conclusions if they correlate with our independent variables of interest and can themselves predict
team success. For example, our results might be confounded by selection issues if an unobserved
component of team member capabilities correlated with either the team’s likelihood of including female
members or its expertise similarity (and/or its network cohesion). Such an unobserved component could
arise from selection by the inventors themselves—for example, if members are concerned about accepting
work that is highly rewarding but takes significant time away from the family (Hoisl and Mariani 2017).
In this section, we examine whether our results hold up across more stringent models: those that consider
the unobserved time-invariant differences across teams.
If one assumes that the unobserved components of team-level capabilities—such as team members’
backgrounds and general innovation capabilities, which are often difficult to observe—are largely innate
or fixed over time, then a way to model unobserved characteristics in heterogeneous teams is to apply
team-level fixed effects. Adopting this approach poses two challenges in our context. First, invention
teams are often “fluid” in the sense of member turnover (Huckman and Staats 2011). Consider two
inventor teams: the first with members {𝐴, 𝐵, 𝐶} and the other with members {𝐴, 𝐵, 𝐷}. A strict team-level
fixed-effects model would consider their respective inventions as being produced by different teams and
thus not comparable, despite common members 𝐴 and 𝐵. A second challenge is that such an approach
would absorb variations fully for some variables (namely, gender mix) that are fixed at the team level.
The identification strategy we adopt in this section seeks to approximate such fixed effects by keeping
only parts of the team fixed. That is, we treat the partial team as fixed and examine performance
implications when changes occur to the remaining team membership. In this spirit, our first robustness
check will apply fixed effects to a random team member listed on each patent. We implement this model
by first randomly picking a team member from the list of inventors on each patent; then we apply fixed
effects to the identified team member. Doing so allows us to examine performance implications by
- 19 -
comparing the performance of the chosen inventor when working with different team members and in
subsequent invention contexts. In the example described previously, if inventor 𝐴 is picked, then the
model will infer performance implications based on comparing patents authored by {𝐴, 𝐵, 𝐶}, by
{𝐴, 𝐵, 𝐷}, and all other patents in which inventor 𝐴 participates. The results of this identification strategy
are presented in Model 7 of Table 3. Observe that the standard errors of the coefficients in this model are
estimated via bootstrapping alternative scenarios of “random picking of team member.” All our insights
remain robust under this model.
The approach just described works better if a larger part of the team can be “fixed”—that is, if we can
compare situations where multiple team members are constant (e.g., both inventors {𝐴, 𝐵} remain
constant) —when there are changes across other team members and invention contexts. We could then
account for a larger part of unobserved team heterogeneities: we would compare the performance
of {𝐴, 𝐵, 𝐶}, {𝐴, 𝐵, 𝐷}, and all other patents in which inventors 𝐴 and 𝐵 played a part. In that case, we can
see whether any unobserved team capabilities are confounding our results.
Because our data set is dominated by men and because an important variation in our comparisons is
whether a group of men work with female inventor(s) as part of the group, we undertake a matching
approach that first identifies the male members of all mixed-gender teams. For example: if {𝐴, 𝐵, 𝐶} is a
mixed-gender team in which only 𝐶 is female, then we first identify the set {𝐴, 𝐵}. Next we identify all
inventions produced by {𝐴, 𝐵} jointly. This matching approach would thus include not only inventions
due to {𝐴, 𝐵} alone but also inventions involving other new members. Our method yields groups of
patented inventions for which, within each group, a subset of inventors appears across all patents.
To ensure maximal overlap for the team members that are fixed, we perform this matching procedure
starting with inventions patented by the largest number of team members. The matching proceeds without
replacement, therefore, we continue matching until no more feasible matches are found. This procedure
enables us to identify 257,859 patents authored by 55,607 groups, where the average team size in this
sample of patents is 3.63 members. There is considerable overlap, as the number of inventors identified as
fixed in each group is (on average) 3.08. We thus construct a more stringent model that controls for hard-
to-observe but time-invariant features by fixing a large part of each team, enabling us to derive estimates
based on changes to team composition. The result of this matching approach is presented in Model 8 of
Table 3. All our insights remain robust to this significantly more stringent comparison.
5.2. Alternative measures of integrality
Our measure of integrality—based on a simple count of the number of chunks—helps us capture
invention integrality and remain parsimonious. Yet, we can imagine more detailed measures of integrality
that would alternatively consider that (1) chunks are of different sizes, and (2) there may be links
(interdependencies) across chunks.
- 20 -
To the issue that chunks can be of different sizes, we can replace our main measure of integrality with
a concentration-based alternative. We look at whether claims tend to be concentrated on a single chunk
(indicating an integral invention) or instead tend to be distributed across multiple chunks (indicating a
more decomposable invention). Then we use the Herfindahl–Hirschman index (HHI) to create a
concentration measure whose value is small if claims are highly dispersed across many chunks, increases
as claims concentrate on fewer chunks, and reaches 1 if all claims relate to a single chunk. It is intuitive
that this measure accounts for not just the number, but also any size differences of the chunks (in terms of
how many claims are associated with a given chunk; see Panel B Figure 3 for an illustration). Model 9 in
Table 3, which employs the HHI integrality measure, is consistent with our main results.
Our second model (Model 10 in Table 3) considers the scenarios where chunks may themselves be
linked (i.e., not fully dependent). Here, we leverage on a measure of integrality based on the Q measure of
modularity (Newman 2006; we thank reviewers for proposing this measure). Fixson et al. (2017) used Q
to capture the knowledge modularity at the industry level; we draw inspiration from their approach and
apply the same measure at the invention level. Q considers the possibility that the chunks may be less
independent than assumed in our main approach; instead, there might be links among those chunks that
render them interdependent. Panel C in Figure 3 illustrates an example. As it shows, Q captures the
degree to which the links between chunks fall within subgroups of chunks rather than across them.
A higher value of Q therefore indicates a higher level of modularity and hence a lower level of integrality.
Panel A (count)
Panel B (HHI)
Panel C (Q measure)
Figure 3: Three measures of integrality. Each node represents a distinct chunk identified in the invention. Panel A
illustrates our main approach of counting the number of subject matters. Panel B (Table 3, Model 9) illustrates an
integrality measure using the Herfindahl–Hirschman index; it captures the distribution of sizes of each chunk (as
measured by the number of claims). Panel C (Table 3, Model 10) illustrates integrality calculated by Q measure,
which considers the relative strengths of the connection between chunks.
Note that Q depends on how the subgroups of chunks are defined. Classic work on the subject
references not only Newman’s (2006) creation of Q but also the spectral partitioning algorithm that uses
this measure as an objective function to find the optimal grouping that produces the highest Q (see also
Chan et al. 2018). Put differently: in the absence of pre-determined subgroups, the Q measure is better
- 21 -
described as the highest Q (denoted as Q*) resulting from an optimal partitioning of chunks into
subgroups. We describe the technical details of the partitioning approach in Online Appendix I. Model 10
of Table 3 indicates that our results remain robust to measuring an invention’s integrality by the negative
of Q* (recall that a higher Q* corresponds to greater modularity).
Table 3: Fixed effects on part of the team and alternative measures of integrality
Model 7
Individual
inventor FE
Model 8
Set of fixed
inventors
Model 9
HHI measure of
integrality
Model 10
Q measure of
integrality
ExpertiseSim(dm)
–0.013*** (0.003)
–0.007 (0.010)
–0.024***(0.007)
–0.024***(0.007)
NetworkCohesion(dm)
–0.007*** (0.001)
–0.005 (0.003)
–0.005* (0.003)
–0.005* (0.003)
MixedGender(dm)
–0.009*** (0.003)
–0.018*** (0.005)
–0.017***(0.005)
–0.017***(0.005)
ExpertiseSim(dm) × Integrality(dm)
0.020*** (0.004)
0.021* (0.011)
0.092** (0.030)
0.099** (0.033)
NetworkCohesion(dm) × Integrality(dm)
0.008*** (0.001)
0.007** (0.002)
0.028** (0.010)
0.025* (0.011)
MixedGender(dm) × Integrality(dm)
0.011*** (0.003)
0.016* (0.006)
0.049** (0.017)
0.045* (0.018)
Integrality(dm)
–0.004+ (0.002)
–0.005 (0.006)
0.023 (0.023)
0.030 (0.037)
Control variables
TeamSize
0.011*** (0.001)
0.016*** (0.002)
0.009***(0.001)
0.009***(0.001)
LogNetworkSize
0.024*** (0.003)
0.029* (0.012)
0.020+ (0.012)
0.020+ (0.012)
TeamExperience
–0.225*** (0.004)
–0.285*** (0.014)
–0.233***(0.021)
–0.233***(0.021)
LogPastPatentValue
0.155*** (0.002)
0.189*** (0.007)
0.185***(0.015)
0.185***(0.015)
LogPatentClaims
0.002 (0.002)
–0.001 (0.005)
0.004 (0.009)
0.004 (0.009)
LogPatentCites
–0.012*** (0.001)
–0.012*** (0.003)
–0.005 (0.006)
–0.005 (0.006)
Firm controls
Yes
Yes
Yes
Yes
Filing Year FE
Yes
Yes
Yes
Yes
Product-class FE
Yes
Yes
Yes
Yes
FE level
Random inventor
Set of fixed
inventors
Firm
Firm
Within-R2
0.17
0.17
0.12
0.12
F
–
303.5***
30.5***
27.9***
N
468,023
257,465
468,023
468,023
Notes: FE = fixed effects. Standard errors (in parentheses) are clustered on the “FE level”: that is, on a randomly identified
inventor on a team in Model 7, set of fixed inventors on a team in Model 8, and on the firm in Models 9 and 10. The set of firm-
level controls are the same as those used in our main analysis (see notes in Table 2).
*p < 0.05, **p < 0.01, ***p < 0.001
5.3. To what extent does the invention team determine an invention’s modularity?
In line with the literature on modularity, which typically assumes that an invention’s modularity is mostly
beyond the invention team’s control (Colfer and Baldwin 2016), our analysis has assumed that the team
does not meaningfully shape the invention’s modularity. Specifically, estimation from our moderation
analyses will be robust only if the independent variables (i.e., team composition factors) have little or no
influence on the moderating variable (here, the invention’s integrality). Yet if such influence is very large,
then results from our moderation analyses would be affected by multicollinearity issues (Frazier et al.
2004). Moreover, there must be sufficient variation in invention integrality that is independent of the
- 22 -
team’s characteristics—that is, beyond the team’s control—because such variation drives our
identification of the moderating role of invention integrality in the team’s effects on invention outcomes.
To address these issues, we perform a partial correlation analysis to estimate how much of an
invention’s integrality can be explained by our independent variables. Such an analysis estimates the
proportion of the variance in an invention’s integrality that can be explained by each independent variable
while accounting for the effects of all other independent variables. Thus, the squared semi-partial
correlation of an independent variable reflects the increment in R2 that occurs when the variable is added
to the model of an invention’s integrality (Greene 2008; StataCorp LLC 2021). This procedure
corresponds to Models 11–13 of Table 4, where we examine the squared semi-partial correlation of each
variable on integrality. Model 11 considers our main measure of integrality, the negative of (the log of)
the number of chunks; Model 12 is for the HHI measure; Model 13, the Q measure.
4
Model 11 shows the squared semi-partial correlations of the three team composition factors. The
reported numbers are quite low: the squared semi-partial correlation of MixedGender is 0.00%, of
ExpertiseSim is 0.18%, and of NetworkCohesion is 0.00%. In other words, our team composition factors
(team gender mix, expertise similarity, and network cohesion) together explain only 0.18% of the total
variation in the invention’s integrality. As a result, our analyses presented here are not compromised by
multicollinearity issues (Frazier et al. 2004).
Furthermore, the sum of squared semi-partial coefficient for additional team-level controls—the
team’s number of inventors, network size, patenting experience, and past invention values—is only
0.10%, adding a little more to the variation in the invention’s modularity. Together, team-level
characteristics only contribute to 0.28% (0.18% + 0.10%) of the variance of integrality. By contrast,
summing the values under firm-level controls and the contribution from firm-level fixed effects reveals
that the total contribution of firm-level factors is 5.56% (0.11% + 5.45%). Technology (as captured by
product-class fixed effects) similarly constrains the invention’s integrality by explaining 2.17% of its
variation. Thus, technology- and firm-level factors in total account for about 7.73% of the variation in
integrality, which is a much larger proportion than the 0.28% from team-level factors. This pattern of
results is similar when the Herfindahl–Hirschman index (Model 12) or the Q measure (Model 13) is used
as our basis for measuring an invention’s integrality. Because we expect managers (e.g., R&D managers
in charge of assembling R&D teams) to have more information about customers’ preferences and the
technological or firm-level constraints that drive an invention’s integrality than the coarse measures we
4
We focus on how team- and firm-level characteristics account (or not) for the variation in invention integrality.
Other invention-level measures—such as scope of the invention and breadth of relevant knowledge base—are
viewed as outcomes of the invention, not predictors of integrality; therefore, we do not discuss them.
- 23 -
employed in our analysis, the 7.73% figure that we obtained earlier likely amounts to a lower bound on
what managers would know in advance about the invention’s level of integrality.
Hence, we can see that team-level factors (including team composition factors and team-level
controls) explain only a negligible portion of the invention’s integrality. This is consistent with
considerable empirical support for the prevailing theoretical stance that team-level factors exert little
influence on an invention’s integrality (see e.g., Colfer and Baldwin 2016). In fact, integrality (of a
product) is driven largely by the heterogeneity of customer preferences and the frequency with which
certain components of a product are updated (Kamrad et al. 2017). To the extent that firm fixed effects
capture time-invariant product differentiation, this account is consistent with the larger contributions of
firm-level factors in determining integrality (of an invention).
Table 4: Squared semi-partial correlation explaining variation in invention integrality
Model 11
Negative of log
number of chunks
Model 12
HHI measure
Model 13
Negative Q
Team composition factors
MixedGender
0.00%
0.00%
0.00%
ExpertiseSim
0.18%
0.11%
0.11%
NetworkCohesion
0.00%
0.00%
0.00%
Control variables
Team-level controls
0.10%
0.02%
0.02%
Firm-level controls
0.11%
0.12%
0.11%
Firm FE
5.45%
3.86%
3.80%
Filing Year FE
1.03%
0.71%
0.88%
Product-class FE
2.17%
1.37%
1.16%
Notes: Squared semi-partial correlations are reported in percentages. Team-level controls include TeamSize,
LogNetworkSize, LogExperience, and LogPastPatentValue. The set of firm-level controls are the same as
those used in our main analysis (see notes in Table 2).
6. Discussion
The results we derive from a large data set of patented inventions establish that—when working on
integral inventions—teams with mixed-gender membership, more similar expertise, and higher network
cohesion can generate inventions of substantially higher value than all-male teams with small expertise
overlap and little network cohesion. Our estimates suggest that, taken together, the total net increase in
market value that arises from these three effects can be as great as 6.2%.
The present study makes several contributions. First, although past research has examined how
innovation processes and invention outcomes are affected both by the integrality of inventions (Baldwin
and Clark 2000; Fixson and Park 2008; Ulrich 1995; Yayavaram and Ahuja 2008) and by team
composition (Fleming et al. 2007; Huckman and Staats 2011; Nielsen and Börjeson 2019), scholarly
treatment of these avenues of influence has remained in more or less separate silos; hence, little is known
- 24 -
about the interplay between these two factors. Our work examines this relationship and demonstrates the
need to consider an invention’s structure and team composition jointly rather than in isolation. The results
reported here indicate that certain team composition factors—namely, a high degree of expertise
similarity (Tortoriello et al. 2015), network cohesion (Fleming et al. 2007; Girotra et al. 2010), and
mixed-gender composition (Nielsen et al. 2017, Nielsen and Börjeson 2019)—that appear to be
innovation liabilities, on average, are actually beneficial when teams work on integral inventions.
Our paper also advances the discussion of team composition and performance. Previous studies have
suggested that expertise similarity and network cohesion have generally negative effects on invention
outcomes even as these same factors facilitate team coordination and thereby the innovation process. By
bringing task interdependencies into the picture, our findings yield novel insights regarding team
composition and its impact on performance. In a similar vein, demographic diversity has often been
associated with a greater risk of dysfunctional conflict in teams (van Knippenberg and Schippers 2007,
Mannix and Neale 2005). Nonetheless, and consistent with prior evidence from laboratory experiments
(Keck and Tang 2018, Kennedy 2003, Woolley et al. 2010) demonstrating a heightened quality of team
interactions associated with the presence of female members, our results establish that mixed-gender
teams outperform all-male teams with respect to integral inventions.
We remark that our work replicates results reported previously that, on average, mixed-gender teams
underperform all-male teams (Ding et al. 2006; Jensen et al. 2018). However, we caution against
interpreting these results as suggesting that firms should prefer all-male over mixed-gender teams when
the invention is characterized by average (or lower) levels of integrality. In fact, such underperformance
is likely less indicative of any innate male–female differences in ability (Hoisl and Mariani 2017) than of
the organization’s failure to fully leverage the female team members’ expertise (Fernandez-Mateo and
Kaplan 2018; Joshi 2014). Hence, our results should be viewed as advocating for gender equality in
organizations: we show that, despite facing more organizational challenges, mixed-gender teams can still
outperform all-male teams when the focal invention is integral. At the risk of unduly extrapolating our
results, we believe that the gains we document could be larger if those challenges were mitigated.
Lastly, our work identifies team characteristics that boost performance when facing the coordination
requirements integral inventions impose (Baldwin and Clark 2000; Chan et al. 2021; Kavadias and
Sommer 2009; Schilling 2000). We thus add a new perspective to the prior literature, which has focused
on coordination within supply chains or between business units, firms, or teams (Colfer and Baldwin
2016, Gokpinar et al. 2013, Lakhani et al. 2013, Schilling and Steensma 2001, Sosa et al. 2015, Ülkü and
Schmidt 2011, von Hippel 1994). Our findings suggest that there is also value in considering how within-
team factors alter a team’s ability to coordinate—thereby shifting the locus of discussing coordination
from a larger unit of analysis, such as coordination between firms, to the inventors in teams.
- 25 -
Besides contributing to the literature, our findings bear implications for practitioners responsible for
managing new product development. Organizations are often well aware of the desired architecture of
inventions in their domains, an awareness based on demand-side factors—such as customer preferences
for the upgradability and flexibility afforded by modular products versus the smoother user experience
afforded by integral products (Kamrad et al. 2017). They might also have learned, in the context of an
integral invention, to prioritize improving organizational communication and coordination (Gokpinar
et al. 2010, Sosa et al. 2015). By highlighting the interplay between team composition and invention
structure, our results provide insights that go beyond this existing knowledge. Our findings can help guide
firms in designing their teams to optimally leverage the advantages of different types of team
compositions depending on the structure of the invention on which a team is working. Which of the
various aspects of team composition firms might then want to focus will likely depend on their specific
resource constraints. For example, some firms may incur higher costs building teams with more network
cohesion, whereas others may find it costlier to adjust the level of expertise similarity.
Limitations and future research
Our work has several limitations that point to fruitful avenues for future research. We first note that,
although patents make it possible to study our research question in a real-world setting, not all useful
ideas are patented. Nevertheless, patenting is relatively more reflective of invention activities in our
sample’s large public firms, because they tend to patent as much as possible. This means that our data set
of patents is less likely to miss out on useful but unpatented ideas yet more likely to include ideas that,
though not entirely useless, may have no immediate utility (Moore 2005). Another problem common to
all research based on archival data (such as patents) is that establishing causality is complicated by the
possibility of unobserved characteristics driving both the independent and dependent variables. However,
all our results hold even when exploiting only marginal changes in team composition (via matching and
fixed effects on a partial set of team members). This provides evidence that our findings should not be
perceived as spurious. A third issue is that we only consider utility patents in our dataset—it is thus
unclear whether all our insights will carry over to innovations less well represented in utility patents.
Extending our results to such settings might thus provide a fruitful avenue for future research.
Moreover, we studied the three team composition dimensions suggested by previous research as
being the most relevant to explore in the context of integral inventions. We did so because of their likely
effects on the quality of team coordination processes. However, these three dimensions are not the only
team factors that are important for the success of teams working on integral inventions. Future research
should therefore explore other aspects of team composition that have been shown to affect team
processes; examples of such aspects include cultural diversity (Watson et al. 1993), geographic distance
- 26 -
(Bardhan et al. 2013; Malhotra et al. 2001), organizational boundaries in the context of open innovation
(Baldwin and von Hippel 2011), heightened levels of team autonomy (Chandrasekaran and Mishra 2012),
and teams with both star and non-star innovators (Liu et al. 2018).
Finally, we considered teams of relatively small size (the average team size in our patent data was
3.47); therefore, we refrain from extrapolating our findings to much larger teams in which cooperation
processes might be very different (see e.g., Devine et al. 1999). Nonetheless, we believe that a focus on
smaller teams is also advantageous due to their prevalence in companies (Steffens et al. 2011; Wu et al.
2019), as well as in the literature on innovation and group dynamics (e.g., Woolley et al. 2010). By
contrast, the literature on modularity tends to consider large teams or firm alliances (Fixson et al. 2017;
Yayavaram and Ahuja 2008; Yayavaram and Chen 2015); hence, there has been limited understanding
about the role of invention integrality in small teams. Our focus on small teams allows us to fill this gap
by answering different sorts of questions and providing novel insights to team management.
Acknowledgement
The authors are grateful for constructive feedback from two anonymous reviewers, the anonymous senior
editor, and department editor Glen Schmidt. They also thank Subramanian Balachander, Diwas KC,
Elodie Adida Goodman, Jurgen Mihm, Jill Perry-Smith, Manuel Sosa, Puay Khoon Toh, and audience
members at seminars at Durham University, National University of Singapore, Singapore Management
University, and the University of Minnesota for helpful comments.
References
Alan, Y., G. P. Gao, V. Gaur. 2014. Does inventory productivity predict future stock returns? A retailing
industry perspective. Management Science 60(10): 2416–2434.
Ancona, D. G., D. F. Caldwell. 1992. Demography and design: predictors of new product team
performance. Organization Science 3(3): 321–341.
Baer, M., K. Abhijeet, R. Vadera, T. A. J. Leenders, G. R. Oldham. 2014. Intergroup competition as a
double-edged sword: How sex composition regulates the effects of competition on group creativity.
Organization Science 25(3): 892–908.
Baldwin, C. Y., K. B. Clark. 2000. Design rules: The power of modularity. MIT Press.
Baldwin, C. Y., E. von Hippel. 2011. Modeling a paradigm shift: From producer innovation to user and
open collaborative innovation. Organization Science 22(6): 1399–1417.
Balsmeier, B., M. Assaf, T. Chesebro, G. Fierro, K. Johnson, S. Johnson, L. Guan-Cheng, S. Luck, D.
O’Reagan, B. Yeh, G. Zang, L. Fleming, G. C. Li, S. Lück, D. O’Reagan, B. Yeh, G. Zang, L.
Fleming. 2018. Machine learning and natural language processing on the patent corpus: Data, tools,
and new measures. Journal of Economics and Management Strategy 27: 535–553.
Bardhan, I., V. V. Krishnan, S. Lin. 2013. Team dispersion, information technology, and project
performance. Production and Operations Management 22(6): 1478–1493.
Bercovitz, J., M. Feldman. 2011. The mechanisms of collaboration in inventive teams: Composition,
- 27 -
social networks, and geography. Research Policy 40(1): 81–93.
Bessen, J. 2008. The value of U.S. patents by owner and patent characteristics. Research Policy 37(5):
932–945.
Boscaljon, B., G. Filbeck, T. Smaby. 2006. Information content of business methods patents. The
Financial Review 41(3): 387–404.
Bradley, B. H., B. E. Postlethwaite, A. C. Klotz, M. R. Hamdani, K. G. Brown. 2012. Reaping the
benefits of task conflict in teams: The critical role of team psychological safety climate. Journal of
Applied Psychology 97(1): 151–158.
Bruns, H. C. 2013. Working alone together: Coordination in collaboration across domains of expertise.
Academy of Management Journal 56(1): 62–83.
Burt, R. S. 2004. Structural holes and good ideas. American Journal of Sociology 110(2): 349–399.
Chan, T. H., J. Mihm, M. Sosa. 2021. Revisiting the role of collaboration in creating breakthrough
inventions. Manufacturing & Service Operations Management 23(5): 1005–1024.
Chan, T. H., J. Mihm, M. E. Sosa. 2018. On styles in product design: An analysis of U.S. Design patents.
Management Science 64(3): 1230–1249.
Chandrasekaran, A., A. Mishra. 2012. Task design, team context, and psychological safety: An empirical
analysis of R&D projects in high technology organizations. Production and Operations Management
21(6): 977–996.
Colfer, L. J., C. Y. Baldwin. 2016. The mirroring hypothesis: Theory, evidence, and exceptions.
Industrial and Corporate Change 25(5): 709–738.
Desyllas, P., M. Sako. 2013. Profiting from business model innovation: Evidence from Pay-As-You-
Drive auto insurance. Research Policy 42(1): 101–116.
Devine, D. J., L. D. Clayton, J. L. Philips, B. B. Dunford, S. B. Melner. 1999. Teams in organizations:
Prevalence, characteristics, and effectiveness. Small Group Research 30(6): 678–711.
Ding, W. W., S. G. Levin, P. E. Stephan, A. E. Winkler. 2010. The impact of information technology on
academic scientists’ productivity and collaboration patterns. Management Science 56(9): 1439–1461.
Ding, W. W., F. Murray, T. E. Stuart. 2006. Gender differences in patenting in the academic life sciences.
Science 313(5787): 665–667.
Edmondson, A. C. 1999. Psychological safety and learning behavior in work teams. Administrative
Science Quarterly 44(2): 350–383.
Faber, R. C. 2001. Landis on mechanics of patent claim drafting (5th ed.). New York: Practising Law
Institute.
Fernandez-Mateo, I., S. Kaplan. 2018. Gender and organization science: Introduction to a virtual special
issue. Organization Science 29(6): 1229–1236.
Fixson, S. K., D. Khachatryan, W. Lee. 2017. Technological uncertainty and firm boundaries: The
moderating effect of knowledge modularity. IEEE Transactions on Engineering Management 64(1):
16–28.
Fixson, S. K., J. K. Park. 2008. The power of integrality: Linkages between product architecture,
innovation, and industry structure. Research Policy 37(8): 1296–1316.
Fleming, L., S. Mingo, D. Chen. 2007. Collaborative brokerage, generative creativity, and creative
success. Administrative Science Quarterly 52(3): 443–475.
- 28 -
Fleming, L., O. Sorenson. 2001. Technology as a complex adaptive system: Evidence from patent data.
Research Policy 30(7): 1019–1039.
Fletcher, J. K. 1998. Relational practice: A feminist reconstruction of work. Journal of Management
Inquiry 7(2): 163–186.
Frazier, P. A., A. P. Tix, K. E. Barron. 2004. Testing moderator and mediator effects in counseling
psychology research. Journal of Counseling Psychology 51(1): 115–134.
Gebert, D., S. Boerner, E. Kearney. 2007. Cross-functionality and innovation in new product
development teams: A dilemmatic structure and its consequences for the management of diversity.
European Journal of Work and Organizational Psychology 15(4): 431–458.
Gilson, L. L., C. E. Shalley. 2004. A little creativity goes a long way: An examination of teams’
engagement in creative processes. Journal of Management 30(4): 453–470.
Girotra, K., C. Terwiesch, K. T. Ulrich. 2010. Idea generation and the quality of the best idea.
Management Science 56(4): 591–605.
Gokpinar, B., W. J. Hopp, S. M. Iravani. 2010. The impact of misalignment of organizational structure
and product architecture on quality in complex product development. Management Science 56(3):
468–484.
Gokpinar, B., W. J. Hopp, S. M. Iravani. 2013. In-house globalization: The role of globally distributed
design and product architecture on product development performance. Production and Operations
Management 22(6): 1509–1523.
Gong, Y., S. Y. Cheung, M. Wang, J. C. Huang. 2012. Unfolding the proactive process for creativity:
Integration of the employee proactivity, information exchange, and psychological safety perspectives.
Journal of Management 38(5): 1611–1633.
Greene, W. H. 2008. Econometric analysis (6th ed.). Upper Saddle River, New Jersey, Pearson
Education.
Hall, B. H., A. B. Jaffe, M. Trajtenberg. 2005. Market value and patent citations. RAND Journal of
Economics 36(1): 16–38.
Hall, J. A. 1978. Gender effects in decoding nonverbal cues. Psychological Bulletin 85(4): 845–857.
Hargadon, A. B., B. A. Bechky. 2006, August 1. When collections of creatives become creative
collectives: A field study of problem solving at work. Organization Science. INFORMS, 484–500.
Hargadon, A., R. I. Sutton. 1997. Technology brokering and innovation in a product development firm.
Administrative Science Quarterly 42(4): 716–749.
Harvey, S. 2014. Creative synthesis: Exploring the process of extraordinary group creativity. Academy of
Management Review 39(3): 324–343.
Hoegl, M., H. G. Gemuenden. 2001. Teamwork quality and the success of innovative projects: A
theoretical concept and empirical evidence. Organization Science 12(4): 435–449.
Hoisl, K., M. Mariani. 2017. It’s a man’s job: Income and the gender gap in industrial research.
Management Science 63(3): 766–790.
Huckman, R. S., B. R. Staats. 2011. Fluid tasks and fluid teams: The impact of diversity in experience
and team familiarity on team performance. Manufacturing & Service Operations Management 13(3):
310–328.
Jensen, K., B. Kovács, O. Sorenson. 2018. Gender differences in obtaining and maintaining patent rights.
Nature Biotechnology 36(4): 307–309.
- 29 -
Joshi, A. 2014. By whom and when Is women’s expertise recognized? The interactive effects of gender
and education in science and engineering teams. Administrative Science Quarterly 59(2): 202–239.
Kamrad, B., G. Schmidt, S. Ülkü. 2017. In pursuit of product modularity: Impediments and stimulants.
California Management Review 59(4): 97–113.
Kavadias, S., S. C. Sommer. 2009. The effects of problem structure and team diversity on brainstorming
effectiveness. Management Science 55(12): 1899–1913.
Keck, S., W. Tang. 2018. Gender composition and group confidence judgment: The perils of all-male
groups. Management Science 64(12): 5877–5898.
Keck, S., W. Tang. 2021. Elaborating or aggregating? The joint effects of group decision-making
structure and systematic errors on the value of group interactions. Management Science 67(7): 4287–
4309.
Kennedy, C. 2003. Gender differences in committee decision-making: Process and outputs in an
experimental setting. Women and Politics 25(3): 27–45.
Kline, P., N. Petkova, H. Williams, O. Zidar. 2019. Who profits from patents? Rent-sharing at innovative
firms. The Quarterly Journal of Economics 134(3): 1343–1404.
Kogan, L., D. Papanikolaou, A. Seru, N. Stoffman. 2017. Technological innovation, resource allocation,
and growth. The Quarterly Journal of Economics 132(2): 665–712.
Kurtzberg, T. R., T. M. Amabile. 2001. From guilford to creative synergy: Opening the black box of
team-level creativity. Creativity Research Journal 13(3–4): 285–294.
Lakhani, K. R., H. Lifshitz-Assaf, M. L. Tushman. 2013. Open innovation and organizational boundaries:
Task decomposition, knowledge distribution and the locus of innovation. A. Grandori, ed. Handbook
of Economic Organization. Northampton, MA, Edward Elgar Publishing, 355–382.
Lanjouw, J. O., M. Schankerman. 2004. Patent quality and research productivity: Measuring innovation
with multiple indicators. The Economic Journal 114(495): 441–465.
Lingo, E. L., S. O. Mahony. 2010. Nexus work : Brokerage on creative projects. Administrative Science
Quarterly 55(1): 47–81.
Liu, H., J. Mihm, M. E. Sosa. 2018. Where do stars come from? The role of star vs. nonstar collaborators
in creative settings. Organization Science 29(6): 1149–1169.
Lovejoy, W. S., A. Sinha. 2010. Efficient structures for innovative social networks. Management Science
56(7): 1127–1145.
Lovelace, K., D. L. Shapiro, L. R. Weingart. 2001. Maximizing cross-functional new product teams’
innovativeness and constraint adherence: A conflict communications perspective. Academy of
Management Journal 44(4): 779–793.
Malhotra, A., A. Majchrzak, R. Carman, V. Lott. 2001. Radical innovation without collocation: A case
study at Boeing-Rocketdyne. MIS Quarterly 25(2): 229–249.
Mannix, E., M. A. Neale. 2005. What differences make a difference? The promise and reality of diverse
teams in organizations. Psychological Science in the Public Interest 6(2): 31–55.
Mihm, J., C. Loch, A. Huchzermeier. 2003. Problem–solving oscillations in complex engineering
projects. Management Science 49(6): 733–750.
Moore, K. A. 2005. Worthless patents. Berkeley Tech. LJ 20: 1521.
Newman, M. E. J. 2006. Modularity and community structure in networks. Proceedings of the National
Academy of Sciences of the United States of America 103(23): 8577–82.
- 30 -
Nielsen, M. W., S. Alegria, L. Börjeson, H. J. Falk-krzesinski, A. Joshi, E. Leahey, L. Smith-doerr, A. W.
Woolley. 2017. Gender diversity leads to better science. Proceedings of the National Academy of
Sciences of the United States of America 114(13): E2796.
Nielsen, M. W., C. W. Bloch, L. Schiebinger. 2018. Making gender diversity work for scientific
discovery and innovation. Nature Human Behaviour 2(10): 726–734.
Nielsen, M. W., L. Börjeson. 2019. Gender diversity in the management field: Does it matter for research
outcomes? Research Policy 48(7): 1617–1632.
Perry-Smith, J. E., P. V. Mannucci. 2017. From creativity to innovation: The social network drivers of the
four phases of the idea journey. Academy of Management Review 42(1): 53–79.
Perry-Smith, J. E., C. E. Shalley. 2003. The social side of creativity: A static and dynamic social network
perspective. Academy of Management Review 28(1): 89–106.
Pisano, G. P., R. Verganti. 2008. Which kind of collaboration is right for you? Harvard Business Review
86(12): 1–7.
Rahmani, M., G. Roels, U. S. Karmarkar. 2017. Collaborative work dynamics in projects with co-
production. Production and Operations Management 26(4): 686–703.
Sanchez, R., J. Mahoney. 1996. Modularity, flexibility, and knowledge management in product and
organization design. Strategic Management Journal 17(S2): 63--76.
Schilling, M. A. 2000. Toward a general modular systems theory and its application to interfirm product
modularity. Academy of Management Review 25(2): 312–334.
Schilling, M. A., H. K. Steensma. 2001. The use of modular organizational forms: An industry-level
analysis. Academy of Management Journal 44(6): 1149–1168.
Siemsen, E., A. V. Roth, S. Balasubramanian, G. Anand. 2009. The influence of psychological safety and
confidence in knowledge on employee knowledge sharing. Manufacturing and Service Operations
Management 11(3): 429–447.
Simon, H. A. 1962. The architecture of complexity. Proceedings of the American Philosophical Society
106(6): 467–482.
Singh, J., L. Fleming. 2010. Lone inventors as sources of breakthroughs: Myth or reality? Management
Science 56(1): 41–56.
Skilton, P. F., K. J. Dooley. 2010. The effects of repeat collaboration on creative abrasion. Academy of
Management Review 35(1): 118–134.
Sosa, M. E. 2011. Where do creative interactions come from? The role of tie content and social networks.
Organization Science 22(1): 1–21.
Sosa, M. E., M. Gargiulo, C. Rowles. 2015. Can informal communication networks disrupt coordination
in new product development projects? Organization Science 26(4): 1059–1078.
StataCorp LLC. 2021. Partial and semipartial correlation coefficients. Stata Reference Manual (13th ed.).
Stata Press, 1–3.
Steffens, P., S. Terjesen, P. Davidsson. 2011. Birds of a feather get lost together: New venture team
composition and performance. Small Business Economics 2011 39:3 39(3): 727–743.
Sutton, R. I., A. Hargadon. 1996. Brainstorming groups in context: Effectiveness in a product design firm.
Administrative Science Quarterly 41(4): 685–718.
Thompson, J. 1967. Organizations in action: Social science bases of administrative theory. New York:
McGraw-Hill.
- 31 -
Tortoriello, M., B. McEvily, D. Krackhardt. 2015. Being a catalyst of innovation: The role of knowledge
diversity and network closure. Organization Science 26(2): 423–438.
Ülkü, S., G. M. Schmidt. 2011. Matching product architecture and supply chain configuration. Production
and Operations Management 20(1): 16–31.
Ulrich, K. 1995. The role of product architecture in the manufacturing firm. Research Policy 24(3): 419–
440.
Ulrich, K., S. Eppinger. 1999. Product design and development. New York, McGraw Hill.
USPTO. 2010. Manual of patent examination procedures (8th ed.).
USPTO. 2019. Progress and potential: A profile of women inventors on US patents. Technical Report.
Uzzi, B., J. Spiro. 2005. Collaboration and creativity: The small world problem. American Journal of
Sociology 111(2): 447–504.
Vakili, K., S. Kaplan. 2020. Organizing for innovation: A contingency view on innovative team
configuration. Strategic Management Journal.
Van Alstyne, M., E. Brynjolfsson. 2005. Global village or cyber-Balkans? Modeling and measuring the
integration of electronic communities. Management Science 51(6): 851–868.
van Knippenberg, D., M. Schippers. 2007. Work group diversity. Annual Review of Psychology 58(1):
515–541.
von Hippel, E. 1990. Task partitioning: An innovation process variable. Research Policy 19(5): 407–418.
von Hippel, E. 1994. “Sticky information” and the locus of problem solving: Implications for innovation.
Management Science 40(4): 429–439.
von Hippel, E. 1998. Economics of product development by users: The impact of “sticky” local
information. Management Science 44(5): 629–644.
Watson, W. E., K. Kumar, L. K. Michaelsen. 1993. Cultural diversity’s impact on interaction process and
performance: Comparing homogeneous and diverse task groups. Academy of Management Journal
36(3): 590–602.
Williams, M., E. Polman. 2015. Is it me or her? How gender composition evokes interpersonally sensitive
behavior on collaborative cross-boundary projects. Organization Science 26(2): 334–355.
Wooldridge, J. 2010. Econometric analysis of cross section and panel data. Cambridge, MA, MIT Press.
Woolley, A. W., C. F. Chabris, A. Pentland, N. Hashmi, T. W. Malone. 2010. Evidence for a collective
intelligence factor in the performance of human groups. Science 330(6004): 686–688.
Wu, L., D. Wang, J. A. Evans. 2019. Large teams develop and small teams disrupt science and
technology. Nature 566(7744): 378–382.
Wuchty, S., B. F. Jones, B. Uzzi. 2007. The increasing dominance of teams in production of knowledge.
Science 316(5827): 1036–1039.
Yayavaram, S., G. Ahuja. 2008. Decomposability in knowledge structures and its impact on the
usefulness of inventions and knowledge-base malleability. Administrative Science Quarterly 53(2):
333–362.
Yayavaram, S., W.-R. Chen. 2015. Changes in firm knowledge couplings and firm innovation
performance: The moderating role of technology complexity. Strategic Management Journal 36(3):
377–396.
