This article was downloaded by: [185.216.92.21] On: 05 March 2019, At: 07:18Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Organization Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Where Do Stars Come From? The Role of Star vs. NonstarCollaborators in Creative SettingsHaibo Liu, Jürgen Mihm, Manuel E. Sosa

To cite this article:Haibo Liu, Jürgen Mihm, Manuel E. Sosa (2018) Where Do Stars Come From? The Role of Star vs. Nonstar Collaborators inCreative Settings. Organization Science 29(6):1149-1169. https://doi.org/10.1287/orsc.2018.1223

Full terms and conditions of use: https://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact permissions@informs.org.

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2018, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

ORGANIZATION SCIENCEVol. 29, No. 6, November–December 2018, pp. 1149–1169

http://pubsonline.informs.org/journal/orsc/ ISSN 1047-7039 (print), ISSN 1526-5455 (online)

Where Do Stars Come From? The Role of Star vs. NonstarCollaborators in Creative SettingsHaibo Liu,a Jürgen Mihm,b Manuel E. Sosac

a School of Business, University of California, Riverside, Riverside, California 92521; bTechnology and Operations Management, INSEAD,77300 Fontainebleau, France; cTechnology and Operations Management, INSEAD, Singapore 138676Contact: haibo.liu@ucr.edu, https://orcid.org/0000-0002-5999-1385 (HL); jurgen.mihm@insead.edu,

https://orcid.org/0000-0002-7197-1598 (JM); manuel.sosa@insead.edu, https://orcid.org/0000-0002-2675-0518 (MES)

Received: February 21, 2014Revised: January 31, 2016; August 20, 2017;January 19, 2018Accepted: April 24, 2018Published Online in Articles in Advance:October 29, 2018

https://doi.org/10.1287/orsc.2018.1223

Copyright: © 2018 INFORMS

Abstract. Creative stars make disproportionately influential contributions to their fields.Yet we know little about how an innovator’s creative performance is affected by col-laborating with stars. This paper studies the creative aspects of interpersonal collaborationfrom a distinct perspective: the quality of the collaborator. Both star and nonstar col-laborators provide different benefits to a focal innovator. The innovator benefits fromcollaborating with nonstars because they may provide access to diverse informationimproving the outcome of the creative task at hand. In contrast, the focal innovator benefitsfrom collaborating with stars because the focal innovator can also experience and learnfrom the star’s superior set of creative synthesis skills (which integrate diverse, sometimescontradictory ideas into new coherent and holistic solutions) and, thus, build lastingcreative capabilities. Building on theoretical arguments about those two different col-laboration purposes, we first examine how a star collaboration (versus a nonstar col-laboration) affects a comprehensive measure of an innovator’s creativity: the likelihood ofemerging as a star. Second, we examine how the different creative benefits of engagingwith a star versus a nonstar collaborator affect the effect of two widely studied aspectsof interpersonal collaboration on star emergence: social network cohesion and expertisesimilarity. In contrast to collaborations with nonstars, for which social cohesion and ex-pertise similarity limit access to diverse information, negatively affecting star emergence,social network cohesion and expertise similarity have a decidedly positive effect on starcollaborations by improving the transfer of the star’s set of creative skills. Our empiricalsetting consists of designers who have been granted design patents in the United Statesfrom 1975 through 2010.

Keywords: stars • interpersonal collaboration networks • emergence of stars • design patents • knowledge transfer • creative synthesis

1. IntroductionPrevious research has recognized the existence ofcreative stars—individuals who, by virtue of their ex-traordinary creative talent, generate disproportionatelyinfluential output and, thus, make outstanding con-tributions to their fields (Zucker et al. 1998, Ernst 2001,Oettl 2012, Godart et al. 2015). Scholars have shownthat stars may enjoy higher social attention, some ofwhich they can bestow on their collaborators (Azoulayet al. 2010, Simcoe and Waguespack 2011); however,we know little about how collaborating with a staraffects the creative approaches experienced by a focalinnovator and what consequences ensue. In fact,throughout the literature on interpersonal collabora-tion in creative settings, there is an implicit assumption:that the creative approach associated with a star col-laboration is not fundamentally different from theone associated with a nonstar collaboration (Burt 2004,Obstfeld 2005, Uzzi and Spiro 2005, Perry-Smith 2006,Fleming et al. 2007, Tortoriello and Krackhardt 2010,Sosa 2011, Tortoriello et al. 2015). Our paper challenges

this view. We provide evidence that collaborators ofdifferent quality (i.e., stars versus nonstars) inducedifferent ways of innovating, each of which has distinctbenefits for the focal innovator. These various benefitsinduce a distinct outlook for the focal innovator’s ex-traordinary creative performance and thus the focalinnovator’s propensity to also rise to stardom.Many studies that explore the effects of interpersonal

collaboration on creative output have focused on ex-amining the idea-generation phase of innovation (e.g.,Hargadon and Sutton 1997, Burt 2004, Fleming et al.2007, Girotra et al. 2010, Sosa 2011, Tortoriello et al.2015). This focus on idea generation makes sense giventhat constructing a wide-ranging pool of potentiallyinnovative ideas increases the odds that one such ideabecomes a breakthrough innovation (Schumpeter 1934,Ahuja and Lampert 2001, Dahan and Mendelson 2001,Girotra et al. 2010, Singh and Fleming 2010). Accordingto this research stream, the main role played by col-laborators is that of providing new information to a focalinnovator—information that the focal innovator can

1149

recombine and reconfigure into a stream of new ideas.Thus, new information is the raw material, the “in-gredients” that innovators combine to form potentiallyinnovative ideas (Ahuja 2000, Fleming 2001, Perry-Smithand Shalley 2003, Burt 2004, Rodan and Galunic 2004,Perry-Smith 2006, Fleming et al. 2007, Tortoriello andKrackhardt 2010, Sosa 2011, Tortoriello et al. 2015). Themore diverse the collaborators’ backgrounds and,hence, the new information they can contribute, thebetter the focal innovator’s creative performance andthe higher the likelihood of creating a breakthrough(Girotra et al. 2010, Singh and Fleming 2010, Dahlanderand Piezunka 2014). From this perspective, however,no explicit distinctions among levels of collaboratorquality (stars versus nonstars) have been necessarypresumably because that literature has distinguishedcollaborators mainly by the diversity of ideas theyprovide and by those aspects of the collaboration thatmaximize access to diverse ideas.

However, this literature has not addressed how theraw information is recombined and assembled intocoherent innovations; rather, the recombination ofideas and information has been treated as automatic(i.e., a black box). Yet more recent work exploringcreativity has acknowledged how important is theprocess of synthesizing diverse, sometimes contra-dictory ideas and pieces of information into a newcoherent and holistic solution (Hargadon and Bechky2006, Long-Lingo and O’Mahony 2010, Harvey 2014,Chen and Adamson 2015). From that standpoint,innovators must also be able to integrate informationeffectively; theymust have both synthesis knowledgeand skills (Sutton and Hargadon 1996, Ahuja 2000,Harvey 2014). In this context, collaborator quality isdecisive; creative stars become central.

We argue (and provide empirical evidence) that starsare much more likely to possess the synthesis knowl-edge and associated skills that enable effective in-tegration of diverse inputs. This knowledge is highlytacit and difficult to acquire—akin to learning how tosculpt a complex shape or how to create vanguardcuisine (Kogut and Zander 1992, Svejenova et al. 2007,Harvey 2014). Therefore, collaborating with those few“masters” who possess synthesis knowledge givesa focal innovator access to such knowledge and achance to apply it under the guidance of the master.The benefits of such guided learning by doing in acollaboration with a star provide a lasting transfor-mation of the innovator’s creative skills (von Hippeland Tyre 1995, Ainley and Rainbird 2013); the innovatoris then far more likely to create breakthrough innova-tions on a consistent basis and, ultimately, to also becomea creative star.

Hence, we formulate our paper’s central premise asfollows: creative star collaborators have a fundamentallydifferent effect on the focal innovator than do nonstar

collaborators. Nonstar collaborators can support aninnovator in the creation of outstanding outcomes byproviding access to diverse sources of ideas. Such newinformation is typically context-specific (Kogut andZander 1992) and, even when reused in other contexts(Hargadon and Sutton 1997), does not improve thefocal innovator’s ability to create breakthroughs per se.Star collaborators, in contrast, provide unique oppor-tunities to the focal innovator to observe, learn, andpractice synthesis skills that transform and better equipthe innovator to develop extraordinary creativity. Theresult is a much-elevated likelihood of ultimately be-coming a star.If the main role of a star collaborator is to provide

a salient opportunity for the learning of synthesisknowledge and if a nonstar collaborator’s main pur-pose is to provide new (and potentially diverse) in-formation, then the conditions that increase or reducethe benefits of such collaborations should be fun-damentally different. We study how two particularfactors, which are prominent in the literature oninterpersonal collaboration in creative settings, relateto an innovator’s star emergence as a function of thequality of collaborators (Reagans and McEvily 2003,Gargiulo et al. 2009, Reagans et al. 2015, Tortorielloet al. 2015): network cohesion, or the extent to which thefocal innovator and the focal innovator’s collaboratorshare common third parties, and expertise similarity, orthe extent to which the focal innovator and the focalinnovator’s collaborator share similar past work-relatedexperience. Examining these two contingencies is im-portant because they would further validate the distinctlearning mechanisms that underlie effective star andnonstar collaborations.We establish that, in the absence of a creative star,

high levels of network cohesion and expertise simi-larity actually reduce the focal innovator’s likelihoodof becoming a creative star. High network cohesioncan cause social pressure that steers the collaborationtoward convergent groupthink, and high expertisesimilarity implies that there will be far fewer recombi-nant perspectives available to the focal innovator (Janis1972, Perry-Smith and Shalley 2003, Burt 2004, Fleminget al. 2007). As a result, any innovation effort involvingonly nonstar collaborators will likely be adversely af-fected by increased levels of network cohesion andexpertise similarity.In contrast, an effective collaboration with a star that

helps the focal innovator become a creative star is likelyto emphasize the practice and transfer of synthesisknowledge. A cohesive network with a star not onlyfacilitates carrying out the synthesis of potentially creativeideas but also enforces cooperative behavior that em-powers the effective transfer of tacit synthesis knowledge(Reagans and McEvily 2003, Obstfeld 2005, Hargadonand Bechky 2006, Tortoriello and Krackhardt 2010,

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1150 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

Tortoriello et al. 2015). Similarly, shared expertisewith the creative star establishes a common knowl-edge base and, thus, increased absorptive capacity thatalso facilitates the transfer of creative skills to the focalinnovator (Cohen and Levinthal 1990, Reagans andMcEvily 2003, Tortoriello 2014). Therefore, that in-novator thrives in a star collaboration—perhaps to theextent of also becoming a star—because network co-hesion and expertise similarity facilitate the learning ofcomplex synthesis knowledge, which transforms thefocal innovator’s creative capability.

We find support for our theoretical arguments in theempirical setting of industrial design, a discipline ofincreasing relevance to building innovative and com-petitive capabilities in various industries (Brown 2008,Brunner et al. 2008). Industrial design efforts, whichfocus on creating new form, not new functionality,can seldom be definitively formulated because theproblems are ambiguous and the solutions uncertain.Hence, the context of industrial design providesan opportunity to test our arguments related to thedistinct nature of interpersonal collaborations withcreative stars (Ulrich 2011). Our data on this industrycome from the design patent database of the U.S. Patentand Trademark Office (USPTO 2015). We processeddesign patent data for the period 1975–2010 to examinethe creative performance of individual designers ina pool of 144,288 designers who had been granted atleast one design patent.

2. Theory and HypothesesIn what follows, we build an argument that star andnonstar collaborators have distinct effects on the focalinnovator’s extraordinary creative outcomes—that is,the focal innovator’s emergence as a star. We shallproceed in three steps. First, in line with emergingliterature, we argue that there are two different ways tocreate breakthrough innovations; of these, perhaps themore reliable is the one requiring a specific set ofsynthesis skills. We then argue that the exceptionalcreative talent of stars is an example of such creativeskills. Finally, we posit that the focal innovator learnsskills of this nature from a star collaborator, whichsignificantly enhances the focal innovator’s ability toproduce breakthroughs and possibly even emerge asa star in the focal innovator’s own right.

2.1. Two Ways to Create a Breakthrough InnovationThe extant literature has established that collaboratingwith others (rather than working as a lone inventor)increases the chances that the innovator creates abreakthrough (Ahuja and Lampert 2001, Wuchty et al.2007, Singh and Fleming 2010, Uzzi et al. 2013). In-deed, there is broad literature informing us abouthow interpersonal collaboration advances the pur-suit of creative endeavors (Burt 2004, Obstfeld 2005,

Perry-Smith 2006, Fleming et al. 2007, Tortoriello andKrackhardt 2010, Sosa 2011, Tortoriello et al. 2012).This stream of research has, although emphasizingthe idea-generation phase of development, implicitlyadopted an evolutionary view of the creative process.A focal actor exposed to diverse sources of inspiration(e.g., different ideas, different personal backgrounds,different ways of framing problems) is well positionedto identify many new and potentially useful relation-ships among the diverse sources of inspiration and,thus, to generate, in turn, many distinct new ideas(Schumpeter 1934, Csikszentmihalyi 1996, Simonton1999). If a large enough number of diverse ideas aregenerated, then at least some of them are bound(statistically speaking) to be exceptionally good: theseare the breakthroughs (Dahan and Mendelson 2001,Girotra et al. 2010, Singh and Fleming 2010). Fol-lowing exposure to a pool of diverse ideas, individualinnovators who come up with breakthroughs maywell become stars.There have been some recent challenges, however, to

conceptualizing the creation of breakthroughs from thatevolutionary perspective (Hargadon and Bechky 2006,Long-Lingo and O’Mahony 2010, Harvey 2014, Chenand Adamson 2015). Emerging literature emphasizesa lack of deep understanding of the process that allowsinnovators to synthesize different ideas into a new ho-listic solution. For example, Harvey (2014) developedan alternative view, dubbed “creative synthesis,” basedon her observations of consistently innovative organi-zations. According to Harvey (2014), the innovationprocess of many creative organizations starts with athorough understanding of the existing paradigm.Then friction with that paradigm emerges as differentactors expose inconsistencies in their understandingof it. This friction is the beginning of a dialecticalprocess—the creative synthesis that develops a novelunderstanding of the situation by integrating variouspoints of friction into the existing paradigm. Theresulting new framework serves to identify innovationopportunities and, hence, as a map for innovation.A prominent instance of adopting a new innovation mapwas IBM’s move from being a provider of “computingmachines” to a company engaged in “the business ofinformation” (Harvey 2014, p. 330). From a new in-novation map, innovators derive guidance on con-structing exemplars: instantiations of a newunderstandingin the form of a practical product. Multiple exemplarscan be derived from the new framework, yet those in-stantiations may themselves trigger new insights and,hence, (via another round of the dialectic process) evolveand refine the very framework that first generated theexemplars. Thus, innovators can iteratively produce a setof high-potential ideas, refine them, and increase thelikelihood of producing not just one breakthrough buta series of them. Of course, an innovatorwho consistently

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1151

produces breakthrough inventions is far more likely thanothers to become a star.

Although one may challenge the completeness ofHarvey’s (2014) model (Chen and Adamson 2015), itclearly identified a neglected aspect of creative in-teractions and, in doing so, introduced a fundamentallynew perspective. The model highlighted that anycreative process requires an extensive set of skills andknowledge so that creative input can be synthesizedinto coherent innovative outcomes (Hargadon andBechky 2006, Long-Lingo and O’Mahony 2010). Forexample, Harvey (2014) argued that each iteration ofcreative synthesis is aided by three sets of process fa-cilitators: collective attention, enacting ideas, andbuilding on similarities. Collective attention helps tobuild an intellectual understanding of the existingparadigm or to steer the identification of areas ofdisillusionment with that paradigm. Enacting ideasbrings ideas closer to realization; this process consistsof posing difficult questions, building tangible repre-sentations of ideas (e.g., mock-ups, prototypes, ex-periments), and gathering and interpreting insightsfrom those tangible representations. Enacting ideasexposes underlying assumptions and unforeseenproblems (Bechky 2003, Seidel and O’Mahony 2014)and also builds a deeper and shared understanding ofa problem and/or its solution (Nicolini et al. 2012).Finally, building on similarities allows collaborators tosee the connections between disparate perspectives.This process helps frame ideas in terms of the otherperson’s frame of reference, thereby, thereby facilitatingmeaningful communication and, thus, creative synthesis(Reagans and McEvily 2003).

These facilitators make clear that intricate skills arerequired to master a process such as creative synthesis,which raises the question of how an innovator canacquire such skills. We argue that creative stars aremore likely than others to have developed creativesynthesis skills and that innovators who work closelywith stars are more likely (than innovators who havenot) to experience and internalize those skills.

2.2. Creative Stars and Creative Synthesis SkillsCreative stars possess skills that make them especiallyadept at performing and guiding creative synthesis.Stars are better than nonstars at focusing collectiveattention: first on “what is” to understand the statusquo and then on “what could be” to trigger innovation.Stars are consummate questioners with a passion forchallenging standard assumptions; not only do starinnovators askmore questions than nonstar innovators,they also ask more provocative ones. Hargadon andBechky (2006, p. 492) described moments of collectivecreativity as involving “not only the original question,but also [considering] whether there is a better questionto be asked.” Stars are more likely to “fundamentally

reassess the purpose, function or use of a product”(Cross 2003, p. 6). Such questioning may open up newand/or unexpected possibilities. Stars also steer col-lective attention by facilitating (and leading) cognitivegroup processes. They attend to ideas in a way thateach member can understand (Vera and Crossan 2005),which helps their collaborators make meaningfulconnections to others’ ideas. Stars, thus, can help bridgegaps in understanding between collaborators and fa-cilitate communication that pushes the innovation ef-fort forward (Bartunek 1984).In addition, creative stars are better than nonstars at

enacting ideas. For many potentially fruitful problems,both the problem statement and the solution spaceare open ended; hence making progress requires athorough understanding of the context to even definethe problem properly. Stars are much more likely thanothers to decompose complex and poorly definedproblems so that they become clear and well defined(Ho 2001). Stars often reformulate the problem to addstructure (Eckert and Stacey 2003); in this, they focuson what rules to break, whereas nonstars—more oftenthan not—conform to those rules (Weisberg 1999). Starshave a particular talent for identifying those aspects ina solution proposal that are most in need of questioningand verification. Stars are typically obsessed with build-ing the actual enactment and encouraging the searchfor insights from the evaluation of prototypes. Eckert andStacey (2003) found that highly skilled designers, in theirquest for novelty, use prototypes more creatively thando less skilled designers.Creative stars are better than nonstars at building on

similarities. Put simply, innovative thinkers are able toconnect fields, problems, and ideas that others perceiveas being unrelated (Mednick 1962). Not only do theyhave deeper knowledge, but that knowledge is cog-nitively organized in ways that make it more accessibleand functional and, thus, a more efficient facilitator offuture recombination (Bédard and Chi 1992). Also starstend to store knowledge based on meaning, concepts,and principles—an approach that allows for a morethorough and useful cross-referencing of the knowl-edge they possess (Chi et al. 1981, Bordage and Zacks1984). As a result, stars can see more profound con-nections between ideas and are better prepared to in-tegrate opposing dialectic opinions.Finally, stars are better at pacing the entire iterative

creative process (Lloyd and Scott 1994). They intuitwhen to initiate what aspect in the creative journey,when to continue evaluating ideas, and when to startenacting ideas (Ahmed et al. 2003). Stars are not averseto triggering rework for others, which restarts theentire enacting-ideas cycle (Sosa 2014). Most impor-tantly, stars are often more persistent than nonstars atworking with a principal concept; thus, a star willpursue the star’s principal solution concept for as long

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1152 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

as possible and even in the face of unexpected diffi-culties (Cross 2004).

2.3. Learning Creative Synthesis fromCreative Stars

The aforementioned skills make stars exceptionallygood at creative synthesis. This skill set could, inprinciple, be discovered and mastered by any innovatorthrough individual trial and error or via collaborationwith nonstars. Yet, as implied by the preceding dis-cussion, creative synthesis skills are both tacit andintricately linked. It follows that learning them is un-likely to occur through either self-discovery or codifiedmethods (Kogut andZander 1992). Instead, it is by close-up observation and emulation that a focal innovatorlearns such a special set of skills. Interpersonal collab-oration is an ideal environment for learning by doingunder a star’s guidance (Dougherty 1992, von Hippeland Tyre 1995, Ainley and Rainbird 2013).

Suppose, for example, an innovator is learning theskill of focusing collective attention on what is andwhat could be. If this innovator collaborates with a starthen the innovator can observe how the star asksquestions to foster a common understanding of what is;the innovator can also observe how the star thinksabout potential improvements to the existing paradigmand experience how the star steers discussions withcollaborators that both bridge the gaps between dif-ferent viewpoints and allow those collaborators toassimilate behaviors that are difficult to explain inwords. Now consider an innovator who is learning theskills and behaviors that support enacting ideas.Observing creative stars approach an ill-defined prob-lem, decide on what to focus when building initialprototypes, and collect feedback from prototypes pro-vides a learning experience that can hardly be matchedby collaboration with nonstars. Finally, consider learn-ing the stars’ approach to building on similarities. Seeingcreative stars reconcile and connect ideas, seeing thequestions they ask, and observing their frame of mindwhen they connect ideas enables collaborators to con-struct templates for future behavior. Overall, the futureof an innovator is enhanced by learning how to engagein creative synthesis, how to derive exemplars from thenew understanding that results, and how to identifycircumstances under which iteration in search of newoptions is (or is not) advisable.

2.4. Putting It All TogetherA focal innovator benefits from collaborating withother nonstars mainly because such collaborations canbe catalysts of divergent idea generation from whichcreative sparks can lead to a breakthrough idea. Incontrast, the innovator benefits from collaboratingwith a star not only because it will likely improveidea generation but also because the innovator can

experience and learn creative synthesis skills. In thissense, the potential benefits of collaborating with a stargo beyond the specific context of that collaboration; inparticular, such collaborating equips the focal innovatorwith a skill set that allows the focal innovator to generatea series of breakthroughs and also rise to star status. Asa result, the innovator’s long-term creative performance—and therefore the innovator’s likelihood of becominga star—depends on the quality of the innovator’s col-laborators (i.e., star versus nonstars).1 We, thus, for-mulate our first hypothesis.

Hypothesis 1 (H1). A focal innovator is more likely tobecome a star after collaborating with a creative star thanafter collaborating with a nonstar.

From the perspective of knowledge transfer, nonstarcollaborators support a focal innovator’s star emergenceprimarily by opening up access to diverse sources ofknowledge, whereas star collaborators—in addition topossibly providing diverse information—support staremergence primarily by furthering the transfer of creativesynthesis skills; hence, the focal innovator benefits fromthese two types of interpersonal collaboration (starversus nonstar) for distinct reasons. As a result, thecontextual factors often invoked when examining thedeterminants of successful knowledge transfer mayhave diametrically opposed effects. Toward the end ofan improved understanding of such differences, westudy two of the most salient contextual factors presentin interpersonal collaborations: social network cohesionand expertise similarity (Argote et al. 2003, Reagans andMcEvily 2003, Reagans 2005, Sosa 2011, Tortorielloet al. 2012).We focus on these particular factors for two im-

portant reasons. First, previous research has shownthat, despite a high correlation between the social andknowledge structures of interpersonal relations, thosestructures are conceptually different and have com-plementary effects on dyadic knowledge transfer(Reagans and McEvily 2003, Rodan and Galunic 2004,Sosa 2011, McEvily et al. 2012, Tortoriello et al. 2012,Reagans et al. 2015). Second, previous studies haveshown that social network and knowledge structureseach play both supporting and hindering roles as afunction of the interpersonal relation’s nature or thecharacteristics of the knowledge being transferred(Ahuja 2000, Burt 2005, Fleming et al. 2007, Tortorielloet al. 2015). Therefore, we have grounds to believe thatthese two contextual factors will affect the focal in-novator’s chances of becoming a star differently fornonstar versus star collaborations.

2.5. Social Network CohesionSocial network cohesion characterizes the extent towhich two individuals are connected through commonthird parties (Coleman 1988, Gargiulo et al. 2009,

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1153

Tortoriello et al. 2012). If network cohesion is high, thena large proportion of the focal innovator’s collaboratorsalso collaborate with each other. If network cohesion islow, then the focal innovator and the focal innovator’scollaborators work mostly with different people whoare not connected among themselves; in this case, thenetwork has many structural holes around the focalinnovator (Burt 1992).

Network cohesion is known to have negative andpositive effects on creative performance (Burt 2005,Fleming et al. 2007, Sosa 2011, Tortoriello et al. 2015).Low levels of social network cohesion have been as-sociated with the creation of potentially creative ideasbecause the focal innovator can connect ideas fromnonredundant idea sources (Burt 2004, Fleming et al.2007). Yet low network cohesion levels also weakenthe establishment of a supporting environment, whichis needed to nurture creative ideas (Obstfeld 2005,Uzzi and Spiro 2005, Tortoriello and Krackhardt 2010,Tortoriello et al. 2015). In what follows, we describethe contrasting effects of social network cohesion onthe focal innovator’s emergence as a star depending onwhether the focal innovator collaborated with a star orwith nonstars.

When a focal innovator collaborates with a nonstar,social conditions that increase the chances of the focalinnovator becoming a star are those that foster thegeneration of divergent ideas as when a collaborator’sideas stimulate the focal innovator’s divergent thinking.The aim is to increase variance in the ideas generated.In a dense network, however, most collaborators sharea mind-set that reflects common beliefs and experi-ences. So, in a highly cohesive network, collaboratorsbegin to resemble one another in terms of their thoughts,actions, and/or knowledge. Such a network around thefocal innovator and the focal innovator’s nonstar col-laboratormay, therefore, lead to groupthink (Janis 1972).Because network cohesion thereby reduces the potentialfor divergent idea generation, it could limit the potentialfor breakthrough innovations and, thus, hinder thefocal innovator from emerging as a star (Singh andFleming 2010).

Star collaborators can incite divergent thinking aswell, but this aspect of a star collaboration becomessecondary in the presence of its benefits resulting frompracticing and learning synthesis knowledge. If a focalinnovator collaborates with a star, then the benefitsof collaboration primarily depend on the application ofcreative synthesis as well as on the effective transfer ofknowledge and skills associated with creative syn-thesis. In that sense, high levels of social network co-hesion with a star have two benefits. First, they providesupportive conditions for implementing creative syn-thesis. If a focal innovator collaborates with a star andif that pair is embedded within a cohesive networkof collaborators, then it is easier to build a common

understanding of the current paradigm and to agree onwhich of its aspects should be changed; thus, cohesionenables and enhances collective attention, which is thefirst enabling factor of creative synthesis. A close andcohesive collaborative relationship creates a supportiveenvironment for giving and responding to feedback,which allows collaborators to recognize when furtherconcept iteration is necessary (Reagans and McEvily2003, Sosa 2014). Candid feedback is, therefore, crucialfor enacting ideas—the second enabling factor of cre-ative synthesis. Finally, a cohesive environment sup-ported by a dense network of common collaboratorssurrounding the focal innovator and the star encour-ages collaborators to seek similarities among theirdifferent perspectives (Reagans and McEvily 2003,Tortoriello et al. 2015); this dynamic furthers thenotion of building on similarities, which is the thirdtenet of creative synthesis.The second way in which social network cohesion

benefits creative synthesis is through its role in thetransfer of tacit knowledge and skills from the star tothe focal innovator. For such a transfer to be effective,there must be both active cooperation and mutual trustbetween the collaborators, especially when one con-siders the complex and tacit nature of creative synthesisknowledge and skills (Hansen 1999). An increasedpresence of common third parties encourages trustbetween collaborators (Coleman 1988) and increasesthe willingness to share creative synthesis knowledgeon the star’s side (Tortoriello et al. 2012). All this easesthe transfer of knowledge from star to focal innovator(Reagans and McEvily 2003).Note that our argument does not deny the fact that

a star collaboration can also provide access to diverseinformation. Rather, we argue that, in a star collabo-ration, experiencing and learning creative synthesisis by far more consequential than gaining access tosources of diverse information for two reasons. First,the effects of accessing diverse inputs are typicallylimited to the immediate task at hand and less portableto future work, whereas learning synthesis knowledgeand skills permanently transforms the focal inventor’screative capabilities. Second, both stars and nonstarscan act as sources of diverse ideas, but only stars area source of creative synthesis knowledge. As a result,nonstars could replace stars as far as accessing diverseideas is concerned. In sum, although a cohesive net-work with a star collaborator could potentially hinderthe access to diverse information, such a cohesivenetwork facilitates the practice and transfer of synthesisknowledge, which should result in a net positive effectof social cohesion on star emergence. We hence hy-pothesize the following:

Hypothesis 2 (H2). The focal innovator’s emergence as a staris affected in opposite ways by the focal innovator’s social

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1154 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

network cohesion with star versus nonstar collaborators.(i) The focal innovator’s likelihood of becoming a star isnegatively associated with increasing network cohesionwith nonstar collaborators. (ii) The focal innovator’s likelihoodof becoming a star is positively associated with increasingnetwork cohesion with star collaborators.

2.6. Expertise SimilarityA similar line of argument as for social network cohesionapplies to the effect of expertise similarity. As innovatorsaccumulate expertise in various aspects of their field(e.g., techniques, industries, types of products), theymayencounter opportunities to collaborate with colleagues ofsimilar or different backgrounds. We therefore defineexpertise similarity as the extent to which the focal in-novator’s expertise is in areas similar to the focal in-novator’s collaborator’s areas of expertise (Reagans andMcEvily 2003, Reagans 2005, Singh and Fleming 2010).

When a focal innovator collaborates with a nonstarcollaborator, the collaboration thrives on access to di-vergent sources of information; such access increasesthe odds of the focal innovator establishing novel anduseful connections between previously unconnectedideas. In such settings, high levels of expertise simi-larity are likely to be detrimental. A lack of divergentexperiences and backgrounds leads to fewer alternativeperspectives and, thus, fewer opportunities to findnovel combinations of knowledge (Dougherty 1992,Hargadon and Sutton 1997, Ahuja 2000, Obstfeld 2005,Perry-Smith 2006, Tortoriello et al. 2015). Expertisesimilarity, thus, hinders the generation of potentiallycreative ideas and so limits a focal innovator’s chancesto achieve stardom.

When the focal innovator collaborates with a star,engaging in creative synthesis and learning the skillsassociated with creative synthesis become the mostprominent mechanisms by which an innovator emergesas a star. In such collaborations, higher levels of ex-pertise similarity are likely to be beneficial. They supportall three creative synthesis’ facilitators (collective at-tention, enacting ideas, building on similarities) becausea common “idea space” furthers meaningful commu-nication and, thus, the reconciliation of seeminglycontradictory perspectives (Katila and Ahuja 2002,Reagans 2005). Expertise similarity also expeditesthe transfer of knowledge and skills about creativesynthesis from the star to the focal innovator. A focalinnovator must appraise, adapt, and ultimately trans-form the focal innovator’s observations of a star engagedin creative synthesis if the focal innovator seeks to applycreative synthesis skills in the focal innovator’s owncontext. It is easier to adapt new knowledge when it isassociated with familiar ideas and concepts (Cohen andLevinthal 1990, Ahuja and Katila 2001, Tortoriello 2014,Tortoriello et al. 2015). Common frameworks, heuristics,and coding schemes make knowledge transfer more

effective and efficient (Dougherty 1992, Bechky 2003,Carlile 2004). As a consequence, learning creative syn-thesis in the context of well-understood fields of ex-pertise is much easier than learning while also engagingin decoding and building contextual knowledge. Thisgeneralization holds particularly for learning creativesynthesis because most such knowledge is tacit and/orcomplex (Kogut and Zander 1992, Hansen 1999).As was the case for network cohesion, we do not

deny the fact that a star can provide access to diverseideas; however, we believe it to be a secondary effectfor the same two reasons discussed when arguing forH2(ii). That is, synthesis knowledge has a long-lastingeffect on the focal designer, whereas the effect ofaccessing diverse information is likely to be moretransient, and star collaborations are hard to replacebecause they are the primary source to learn creativesynthesis knowledge. Thus, any potential negativeeffect resulting from any increased expertise similarityassociatedwith a star collaboration is likely to be by farexceeded by its positive effects. Given these consider-ations, our theory postulates that expertise similarity hasa positive (negative) effect on star (nonstar) collabora-tions of the focal innovator.

Hypothesis 3 (H3). The focal innovator’s emergence asa star is affected in opposite ways by the focal innovator’sexpertise similarity with star versus nonstar collaborators.(i) The focal innovator’s likelihood of becoming a star isnegatively associated with increasing expertise similaritywith nonstar collaborators. (ii) The focal innovator’s likeli-hood of becoming a star is positively associated with in-creasing expertise similarity with star collaborators.

3. Data, Methods, and AnalysesTo test our hypotheses, we need a longitudinal data setin a creative setting with a large number of repeatedobservations at the individual level. The data shouldenable us to identify each designer uniquely, to trackeach designer’s work over time, to identify those whocollaborated on that work, and—crucially—to evaluateoutput objectively. Design patent data fulfill these re-quirements. Such data contain detailed information onpatent designers’ names and locations as well as oneach patent’s application date, content classification,assignee organization (i.e., the entity to which thepatent is granted), and citations to other patents. Therich information embedded in design patent data isone of the few publicly available sources that hasdocumented creative design output en masse (Chanet al. 2017).

3.1. Design PatentsIn the United States, a design patent can be granted fora “new, original, and ornamental design for an articleof manufacture” (USPTO 2015). In general terms,

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1155

a design patent protects the form of an item, whereasa utility patent protects an item’s functionality. Be-cause design patents focus on innovation in form, notfunction, their scope is limited to the “overall orna-mental visual impression” (USPTO 2015). Design pat-ents provide the ideal context for testing our theorybecause of design’s paradoxical nature: it seeks to findhigher-order solutions that accommodate seeminglyopposed forces. Buchanan (1992) referred to design asdialectic because it occurs at the intersection of con-straint, contingency, and possibility. It follows thatcreative synthesis skills are a fundamental prerequisiteto excelling in the field of industrial design.

Although our study is one of the first ones in usingdesign patent data, we can draw on and extendmethods developed in the context of utility patents.First, the data’s longitudinal nature provides richhistorical information at both the personal and networklevels. Patent applications allow us to track a designer’scollaborators, which means that we can constructa comprehensive collaborative history for each de-signer in our database.2 Second, there is a tradition ofusing patent data to analyze creativity because suchdatamake it possible to quantify creative output and sosubstantially reduce the difficulty of measuring in-novation (Ahuja 2000). We use patent citations as aproxy for the influence of an invention and, hence, of itsinventor(s). So the more citations a patent receives, themore it is viewed as an inspiration for subsequentcreative endeavors (Audia and Goncalo 2007). Thecitations that a patent receives are, therefore, widelyviewed as reliable and systematic indicators of an in-vention’s economic, social, and technological success(Jaffe et al. 2002, Singh and Fleming 2010).

3.2. Identifying the Emergence of Star DesignersWe define the popularity index of a designer at anymoment in time by counting the citations received,excluding self-citations, by that designer’s patents inthe preceding three-year rolling window (Ahuja 2000).An inventor whose popularity index is in the top 2% ofall inventors is considered, at the time of measurement,to be a star designer. Operationalizing “star” in this wayaccords with related literature. Ahuja and Lampert(2001) use the top 1% (of the distribution of patent ci-tations) as their thresholdwhen defining a breakthroughinvention, and Singh and Fleming (2010) use a 5% cutoff.As for the relevant time span, scholars who study ar-chival data have used windows ranging from threeyears (Fleming et al. 2007) to five years (McFayden andCannella 2004) when assessing an inventor’s perfor-mance. In the results section, we discuss the robustnessof our results to various distribution cutoffs and severalrolling time windows.3

We define the emergence of a star designer—or thetransition from designer to star—as an event in the

designer’s career. This event corresponds to the daywhen the designer’s popularity index first attains the2% threshold. Although we processed all the designpatents granted in the United States during the1975–2010 period, our observation window rangesfrom year 1985 to year 2004 because we need to allowfor some time before and after that window for de-signers to accumulate enough citations to establisha stable popularity index. In our observation window,144,288 designers were granted at least one designpatent. Of these, we identified 9,971 star designers(about 7% of the sample).4 This proportion is in linewithprevious research, which has pegged the prevalence ofextreme performers at values ranging from 0.75% to 10%(Zucker et al. 1998, Ernst et al. 2000, Groysberg et al.2011). Following prior research that has examined stars(Groysberg et al. 2011, Grigoriou and Rothaermel 2014),we assume that a star designerwill remain a star throughoutthe study’s observation window.

3.3. Collaborating with a Star DesignerBefore we can test our hypotheses in Section 3.4,we need to discuss a coarsened exact matching (CEM)sampling method that allows us to mitigate potentialconcerns about the selection bias inherent in collabo-rative settings.Our baseline hypothesis (H1) argues that a focal

designer’s chances of emerging as a star are sub-stantially greater after collaborating with a star designerthan with a nonstar designer. Perhaps the foremostchallenge in testing such a hypothesis is the possibilitythat a designer’s inherent capabilities (quality) drivenot only the focal designer’s chance of working witha star but also the likelihood of also emerging as a star.We address this endogeneity concern by employinga CEMprocedure to identify a sample inwhich selectionissues are significantly mitigated (Azoulay et al. 2010,Singh and Agrawal 2010, Iacus et al. 2012, Oettl 2012).This procedure helps us balance the (pretreatment) focaland control subsamples by constructing matched pairsthat are “statistical twins” in the period prior to the focaldesigner’s first star collaboration (the treatment event).Because collaborating with a star is the treatment eventin our sample, the two members of each matched de-signer pair are similar in terms of certain observablepretreatment variables; they differ only in that one de-signer in each pair undergoes the treatment—that is,collaborates with a star.Constructing the CEM sample requires us to identify

key confounders that might be correlated with en-gaging in a star collaboration and also with the focaldesigner becoming a star. We, therefore, consider sixindividual observable characteristics: (i) year of thefocal designer’s first patent application, (ii) the de-signer’s career age, (iii) number of the focal designer’s(nonstar) collaborators, (iv) number of patents granted

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1156 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

to the focal designer, (v) number of inventors that citethe focal designer’s patent(s), and (vi) major patent class(es)to which the designer’s patents have been assigned—a categorical variable that captures 33 major designclasses (USPTO 2015). We follow Bode et al. (2015) inperforming a logit regression to test whether thesefactors are indeed associated with the probability ofcollaborating with a star designer. Table 1 reports theresults, in which the dependent variable is “entering intocollaboration with a star designer.” In this table, model(1) excludes themajor class dummies, whereasmodel (2)includes them. The reported coefficient estimates indi-cate that our factors are indeed significant determinantsof a focal designer’s collaborating with a star designer.Hence, when building our CEM sample, we are justifiedin using these six variables.

To build the CEM sample, we identify pairs of de-signers who were, in terms of the six key pretreatmentobservables, statistically similar at the time of thetreatment event (star collaboration). We match the twogroups—that is, those who do or do not undergo thetreatment event—using exact matches for the first twovariables and using four discrete buckets for each of thelast four variables. As in the extant literature, we matcheach treated designer with exactly one control designerso that we need not assign weights to potentiallymultiple control designers (Azoulay et al. 2010, Singhand Agrawal 2010). Table 2 displays the outcome ofour CEM procedure, which splits 14,250 designers intotreatment and control groups of equal size. As expected,our matching variables are now almost identically

distributed between the treated designers (who end upcollaborating with star designers) and the control group(who do not) at the time of that treatment event: theirrespective covariates are well balanced, and their meanvalues do not differ significantly (at the 5% level).

3.4. Testing H1 via Event History AnalysisIn this section, we test H1 directly by way of an eventhistory analysis that allows us to model precisely therelative likelihood of an event (here, star emergence)over a specific time span while accounting for thedifference between censored and uncensored cases(Blossfeld andRohwer 1995). Our data are right-censoredbecause information about the event of interest mayarrive in the future—as when a designer has not becomea star by the end of our observation window but mightbecome one later.As is customary in event history analysis, we use

maximum likelihood techniques to estimate Cox pro-portional hazardmodels on survival-time data (Cox 1972).To control for the nonindependence of multiple obser-vations of the same designer and to account for hetero-scedasticity,models were estimatedwith robust standarderrors clustered by individual designer. For estimationpurposes, we use the stcox command in Stata 15.0.We test H1 using three samples selected via our CEM

analysis. These samples reflect different criteria re-garding collaboration patterns before the first starcollaboration (i.e., prior to the treatment event). In thefirstsample, each designer had at least one nonstar collabo-rator before the treatment event; this is the CEM sample,

Table 1. Logit Regressions on the Antecedents of Collaborating with a Star Designer

Variable (1) (2)

Year of first patent application −0.014* (0.007) −0.01 (0.007)Career age 0.112*** (0.008) 0.115*** (0.008)Number of nonstar collaborators −0.130*** (0.017) −0.145*** (0.018)Number of patents 0.053*** (0.013) 0.049*** (0.014)Number of citers 0.001*** (0.000) 0.001*** (0.000)Major patent class dummies No YesLog-likelihood −5,285 −5,169

Notes. Robust standard errors clustered by focal designers shown in parentheses. N = 145,200designer–patent observations.

*p < 0.10; ***p < 0.01.

Table 2. Summary Statistics for Matched Samples

Treated designers Control designers

Mean Standard deviation Mean Standard deviation

Year of first patent application 1,997.42 5.14 1,997.42 5.14Career age 1.25 2.91 1.25 2.84Number of nonstar collaborators 3.23 2.91 3.17 3.17Number of patents 3.49 1.76 3.52 1.87Number of citers 14.51 46.72 14.06 56.79

Note. N = 14,250 designers.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1157

described previously, that contains 14,250 designers.Each of the 4,486 designers in the second sample hadno previous collaboration, and so the treatment eventconstituted the first one. Finally, the third samplesimply comprises the previous two and, therefore,contains 18,736 designers. We use the three samples tocompare the effect of star collaboration against twodifferent baselines. Thus, the first sample is used tocompare the effect of star collaboration against nonstarcollaboration and the second to compare the effect ofstar collaboration against no collaboration. We use thethird sample to compare the effect of star collaborationagainst any type of nonstar collaboration (includingno collaboration).

Formally, we create an indicator variable set to onefor treated designers (who collaborate with a star de-signer) or set to zero for control designers (who do not).We control for other possible sources of heterogeneityin the creative abilities of all the sample’s designersby constructing six additional variables described inTable 3.

Table 4 presents the descriptive statistics for andcorrelations between our variables, and Table 5 pres-ents the results for the Cox proportional hazard modelestimations. The effect of our Treated designer dummyvariable in allmodels is positive and significant (p< 0.01),

supporting H1. That effect is also significant in modelsthat exclude all controls. In model (1), in which we usethe sample in which the treated designer collaboratesat least once with a nonstar before the treatmentevent, the dummy indicator has a positive and highlysignificant coefficient (0.438, p < 0.01). It followsfrom this model that our treated designers are 55%(e0.438 − 1 = 0.550) more likely to emerge as a star thanour control designers. Model (2) reports correspondingvalues for when the focal designer worked alone beforethe focal designer’s first collaboration with a star de-signer; these values confirm the notion that a star col-laboration is more beneficial to star emergence than is notcollaborating at all. Hypothesis 1 continues to be sup-ported when instead we use the most general CEM sam-ple, which includes all treated designers; see model (3).

3.5. Complementary Empirical Evidence for H1:Creative Synthesis and Star Designers

We have shown empirically that collaborating witha star increases the likelihood of a focal designer alsobecoming a star. In this section, we provide some ev-idence that the main mechanism behind this effect is, asclaimed in H1, the transfer of creative synthesis skills.Our argument for H1 rested on three claims: (i) starshave mastered creative synthesis, (ii) focal designers

Table 3. Definition of Control Variables for Testing H1

Assignee’s past patents. The assignee is the organization owning the patent and usually the organization with which the designer is associated.The organization’s cumulative number of patents until time t is a good proxy for the scale of its innovation activities (Audia andGoncalo 2007)and strongly correlated with its innovativeness (Trajtenberg 1990). More innovative organizations tend to attract and retain the field’s bestapplicants, and hence, an organization’s patent stock is associated with the quality of its staff.

Class diversity.Audia andGoncalo (2007) report thatmore highly skilled patenting inventors aremore likely to venture into different innovationareas, making the diversity of classes within which a designer has secured patents a proxy for the designer’s capability to think and act ina creative way. Our Class diversity variable is a count of the number of unique subclasses in which a focal designer has been awarded a patentup to time t.

Mobility. The number of firms with which an individual has been associated is strongly correlated with that individual’s experience (Fujiwara-Greveand Greve 2000). The Mobility variable counts the number of unique assignees associated with the focal designer until time t.

Location diversity. There is sufficient abundant evidence suggesting that living in diverse locations lets an individual absorb different cultures,gain access to novel ideas, or experience “conceptual expansion”—all of which increases individual creativity (Maddux andGalinsky 2009). It,thus, makes a focal designer more capable. For each designer, we count the number of different states (in the United States) or number ofdifferent countries (outside the United States) in which the designer has worked until time t.

Patent stock year. Because the number of patents applied for (and granted) has increased over the years (Trajtenberg 1990), we count, for eachyear, the total number of patents granted in eachmajor patent class. Then, for each designer we include the count that corresponds to themajorclass of the designer’s patents (at time t).

Cohort. This is a set of dummy variables used to control for when a designer begins to file patents. It indicateswhether the designer first patentedin one of the five-year intervals during the 1985–2004 period.

Table 4. Descriptive Statistics and Correlation of Variables

Variable Mean Standard deviation 1 2 3 4 5

1 Treated designer dummy 0.56 0.502 Assignee’s past patents 5.53 4.18 0.003 Class diversity 2.65 2.57 0.12 0.004 Mobility 1.30 0.81 0.08 0.00 0.465 Location diversity 1.03 0.18 0.06 0.00 0.17 0.156 Patent stock year 6.72 0.68 0.03 0.03 0.13 0.07 0.03

Notes. N = 44,320. Correlations greater than |0.003| are significant at p < 0.05.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1158 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

learn tacit knowledge about creative synthesis whencollaborating with stars, and so (iii) focal designers aremore likely to become stars themselves after collab-orating with stars. Evidence in favor of claim (iii)followed from the testing described in Section 3.4.Here we provide some empirical evidence for claims (i)and (ii).

It is challenging to derive direct measures of creativesynthesis or the transference of that skill set. Creativesynthesis is amicrolevel process, and it presupposes thatcollaborators have developed a deep understanding ofthe existing innovation paradigm and can create a newone by reconciling distant and often seemingly contra-dictory perspectives; such synthesis is produced byinnovators who excel at focusing collective attention,enacting ideas, and building on similarities. Our data setdoes not allow for direct measurement of these micro-processes. Yet we can measure a reliable indicator of anindispensable corollary of creative synthesis. That is,whatmakes creative synthesis unique is less the creationof breakthroughs through new-to-the-world combina-tions of knowledge fields and more how designersfollow up after identifying a new-to-the-world combi-nation. In a traditional recombinatorial approach toinnovating, the innovators go on to devise new com-binations with the expectation that one of them willbecome a breakthrough (Fleming 2001). In contrast,collaborators engaged in creative synthesis have iden-tified a new paradigm based on the new combination;they are focused on producing exemplars of this newcreative insight, and they continue to iterate and refinethe new paradigm building upon its basic premise.Thus, these collaborators will use the new combinationrepeatedly. It is most fortunate that our data set is richenough to enable measurement of such reuse by thefocal designer of new knowledge combinations that thefocal designer created.

To operationalize the concepts of new combinationsand reuse of new combinations, we exploit the U.S.Patent Office’s sorting of all design patents into some33 classes and hundreds of subclasses based on theirsubject matter. Many patents fall into more than oneclass or subclass. (The USPTO updates its classificationperiodically; we use the 2010 scheme.) We followFleming et al. (2007) and define as a new combinationa pair of subclasses in a patent application that appearsin the USPTO design patent database for the first time.Any further uses of this combination are treated asconventional combinations. To establish a baseline, wetreat all subclass pairs through the end of 1984 asconventional combinations and start identifying newpairs in 1985 (which is the starting year of our empiricalanalysis). We say that a focal designer reuses a newcombination if the designer has filed a patent witha new combination and then subsequently files otherpatents with the same subclass pair.

3.5.1. Do Stars Engage More than Nonstars in CreativeSynthesis? We estimate a logistic regression model tofind whether stars are more likely than nonstars toreuse their new idea combinations. Because we aretesting the likelihood that a designer reuses one of thedesigner’s own new combinations, the unit of analysisis the designer who has created a new combination andwho is observed each time the designer is granteda patent thereafter. Our dependent variable is a Reusedummy set to one if the patent contains at least onereuse of a new combination previously created by thedesigner (and set to zero otherwise). Our main in-dependent variable is the indicator Star, which is set toone (to zero) for individuals who are (are not) a star.As controls, we use all variables defined for the

previous matching analysis and for the tests of H1.However, our Number of collaborators variable now

Table 5. Proportional Hazard Models Predicting Star Emergence to Test H1

Variable

(1) Patents without a starcollaborator (2) Lone innovator patents (3) All patents

Coefficient Standard error Coefficient Standard error Coefficient Standard error

Treated designer dummy 0.438*** (0.05) 1.243*** (0.095) 0.638*** (0.044)ControlsAssignee’s past patents 0.003 (0.006) 0.013 (0.01) 0.005 (0.005)Class diversity 0.306*** (0.028) 0.343*** (0.043) 0.318*** (0.023)Mobility 0.231*** (0.034) 0.307*** (0.076) 0.226*** (0.031)Location diversity 0.751*** (0.121) 0.664*** (0.157) 0.714*** (0.099)Patent stock year 0.652*** (0.058) 0.539*** (0.080) 0.612*** (0.047)

Cohort Yes Yes YesNumber of subjects 14,250 4,486 18,736Number of observations 33,861 10,459 44,320Log-likelihood −15,210 −4,795 −20,397

Note. Robust standard errors clustered by focal designer are reported in parentheses.***p < 0.01.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1159

includes all collaborators (i.e., stars and nonstars)because in this analysis we must control for the focaldesigner’s engagement in collaboration with othersmore generally. Correctly predicting the probabilityof new combinations being reused requires that wecontrol also for the designers’ number of past patentsand the number of past new combinations.

Our regression results are given in Table 6. Model (1)limits the sample to patents without any collaborators(lone patents) and provides strong evidence that starsexhibit creative synthesis skills when working alone.Model (2) is based on a sample of patents for which thefocal designer does not collaborate with a star. Model (3)contains all the patents considered for this analysis. For allthree models, the Star dummy is both positive and sig-nificant (p < 0.01); this result indicates that stars are morelikely than other innovators to reuse their own previouslycreated new combinations. These models, thus, provideevidence that stars engage more than do nonstars inbehaviors that are consistent with creative synthesis.

3.5.2. Do Focal Designers Learn Creative Synthesisfrom Star Collaborators? Next we address the ques-tion of whether focal designers learn creative synthesisbehavior by collaborating with stars. For this purpose,we assess the extent (if any) to which a focal designer ismore likely to reuse new combinations after collabo-rating with a star.

For this analysis, we define Reuse probability as theratio of a focal designer’s patents that contain a reusedcombination to the total number of patents that thedesigner has been granted.We test for reuse probabilityby designers—after their collaboration with a star—viaa difference-in-differences (DID) analysis based on aCEM sample (Angrist and Pischke 2009). We employ

the same CEM procedure described in Section 3.3 butwith one modification: rather than creating matchedpairs based on number of patents granted before thetreatment, we match designers based on their respectivereuse probabilities.To construct a sample for the DID analysis, we ob-

serve all of the treated designers (those who collabo-rated with a star) and control designers for as manyas 12 years (see Singh and Agrawal 2010). Derivingmeaningful comparisons of pretreatment and post-treatment performance requires that both the pre-treatment and posttreatment periods for all designersof interest be at least two years long.Our DID sample for this analysis consists of

2,444 individuals, 1,222 each in the treatment andcontrol groups. The values of our matching variables(not shown here for brevity) are not statistically dis-tinguishable between treatment and control groups.Note that we include only those patents of the focaldesigner that are not made in collaboration with a star(except, of course, for the treatment event patent, whichinvolves a star by definition). This approach makes iteasier to isolate the learning effect from other effects ofcollaboration with a star, which is the treatment event.As shown in the first row of Table 7, the reuse

probabilities of these two groups are statistically in-distinguishable before the treatment event of a starcollaboration. However, there is a positive and sig-nificant boost (at the 0.05 level) in reuse probabilityamong the treated designers following their star col-laboration. Even as the reuse probability remained prac-tically unchanged for the control group, it increased bya factor of nearly eight for the treatment group.We also perform these CEM and DID procedures for

subsamples that contain (i) only patents held by lone

Table 6. Logit Regression on the Likelihood of Combination Reuse

Variable

(1) Lone patents (2) No star collaborative patents (3) All patents

Coefficient Standard error Coefficient Standard error Coefficient Standard error

ControlsStar dummy 0.425*** (0.128) 0.300*** (0.109) 0.337*** (0.091)Number of new combinations 0.011** (0.004) 0.015*** (0.004) 0.015*** (0.003)Number of patents 0.004* (0.002) 0.006*** (0.002) 0.002* (0.001)Number of collaborators −0.004 (0.013) 0.019*** (0.006) 0.011** (0.005)Number of citers −0.001 (0.001) −0.002*** (0.001) −0.001** (0.001)Assignee’s past patents 0.021 (0.024) 0.012 (0.017) 0.007 (0.016)Class diversity 0.019 (0.002) −0.004 (0.002) −0.004 (0.017)Mobility −0.070** (0.028) −0.060** (0.023) −0.022 (0.016)Location diversity 0.020 (0.041) −0.003 (0.035) −0.046 (0.033)Patent stock year −0.008 (0.005) −0.044 (0.036) −0.027 (0.033)Cohort Yes Yes YesDesigners 32,732 60,285 64,049Observations 48,430 91,525 103,955Log-likelihood −8,262 −16,500 −19,287

Note. Robust standard errors clustered by focal designers shown in parentheses.*p < 0.1; **p < 0.05; ***p < 0.01.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1160 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

inventors and (ii) all patents, including those resultingfrom star collaborations. Each subsample yields ro-bust evidence that treated designers engage in morereuse combinations after collaborating with a staralthough for case (i) the results are only marginallysignificant (at the 0.1 level). This outcome is not surprisingin light of the very stringent selection of lone designerpatents.

In sum, our analyses offer evidence that stars engagein more creative synthesis than do nonstars and thatdesigners can learn creative synthesis from their starcollaborators. Despite some limitations of the measuresemployed, our analyses confirm that star collabora-tions increase the focal designer’s likelihood of be-coming a star by activating and transferring creativesynthesis skills—a dynamic that echoes our main the-oretical argument behind H1.

3.6. Testing the Effects of Network Cohesion andExpertise Similarity

In this section, we test H2 and H3 and show howcollaboratingwith a star differs from collaboratingwitha nonstar as evidenced by the contrasting effects ofnetwork cohesion and expertise similarity. Because weposit distinct effects of network cohesion or expertisesimilarity on star versus nonstar collaborations, wemust use a sample that differs from the one used to testH1; in particular, wemust focus on designers who havea nonstar and a star collaborator in their network so thatthe network variables of interest can be measured forboth star and nonstar collaborations. This criterion selects27,278 designer–patent observations in our 1985–2004observation window.

The sample so selected includes designers who havecollaborated with at least one star and one nonstarcollaborator, so the results from testing H2 andH3mustbe interpreted as being conditional on designers collab-orating with both stars and nonstars. Much as in theempirical challenge faced by Bode et al. (2015), thisapproach is a less aggressively causal interpretation—inline with a “treatment on the treated” view of theseeffects (Angrist and Pischke 2009).

A modest amount of notation helps us define theindependent variables. Because the relative order ofevents will matter for our proposed event historyanalysis, we index each patent application date with

t ∈ [1,T]; here t is an integer and T is the last instanceof a patent application in our data. In what follows,we use H it, Lit, and Fit to denote the sets of (re-spectively) stars, nonstars, and all collaborators withwhom focal designer i has worked until time t.Similarly,Gjt is the set of past collaborators of star j attime t.

3.6.1. Direct Ties to Collaborators. We retest thebaseline H1 by measuring focal designer i’s direct tiesto stars and to nonstars as a count of the number offocal designer i’s unique star collaborators until time t(denoted |H it |) and unique nonstar collaborators untiltime t (|Lit |), respectively.

3.6.2. Network Cohesion. Testing H2 requires that wemeasure the (social) network cohesion of the focaldesigner i at time t with respect to focal designer i’s star(H it) and nonstar (Lit) collaborators. Our cohesionmeasure captures the average fraction of common thirdparties within the focal designer’s network. This mea-sure is equivalent to that actor’s local network density,which is defined as the ratio of existing ties in the focaldesigner’s network to all possible ties (Obstfeld 2005,Fleming et al. 2007, Gargiulo et al. 2009). For our study,network cohesion must be defined with respect to twotypes of collaborators: stars and nonstars. Based on thegeneral definition, we, therefore, assess focal designer i’snetwork cohesion with respect to focal designer i’s starcollaborators by calculating the (average) proportion offocal designer i’s past collaborators who also collabo-rated with focal designer i’s star collaborators ( j∈H it).Formally, we have

Network cohesion with starsit �∑

j∈H it(|Gjt ∩ Fit |/|Fit |)|H it |

,

which gives the average fraction, in the focal designer’snetwork, of common third parties with a star collab-orator. The resulting continuous measure is equal tozero if there are no third parties in common with anystar in the focal designer’s network and is equal toone if all the focal designer’s past collaborators arealso past collaborators of the focal designer’s starcollaborators. We define Network cohesion with nonstarsanalogously.

Table 7. Difference-in-Difference Analysis of Reuse Probability

Treated group Control groupDifference

(treated – control) LO95% HI95%

Pre–star-collaboration (n = 8,874) 0.016 0.035 −0.019 −0.060 0.022Post–star-collaboration (n = 18,010) 0.126 0.039 0.087 0.010 0.164

Notes. N = 26,884 designer–year observations. To facilitate interpretations of the results, all variableshave been multiplied by 100.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1161

3.6.3. Expertise Similarity. To identify—for any mo-ment in time—the areas of expertise of the designers inour database, we rely on the classification of designpatents made by the USPTO. Designs in the USPTOpatent classes (and their subclasses) pertain to a specifictype of industrial design and require substantiallydifferent skill sets to produce. It is, therefore, reason-able to view each subclass as a distinct area of expertise.

Following previous work that measures knowledgeoverlaps between actors (Reagans and McEvily 2003,Reagans 2005, Sosa 2011), we measure expertise sim-ilarity with stars as the (average) fraction of areas ofexpertise that are common to both focal designer i anda star collaborator j, divided by the number of the focaldesigner’s areas of expertise. Let Kit (Kjt) be the set ofsubclasses in which the focal designer i (the star j,j∈Hit) has been granted patents until time t. Then

Expertise similarity with starsit

�∑

j∈Hit(|Kit ∩Kjt |/|Kit |)|Hit |

.

This variable, too, ranges from zero to one; here zeroindicates no expertise similarity between the focaldesigner and all the focal designer’s star collaborators,and at the other extreme, one indicates a completeoverlap between the parties’ areas of expertise. Weanalogously define expertise similarity with nonstars.

To identify the different effects that star and nonstarcollaborators have on network cohesion and expertisesimilarity, we must control not only for the focal de-signer’s characteristics, but also for those of the focaldesigner’s star and nonstar collaborators. So, in additionto the control variables used to test H1, we include threesets of control variables defined with respect to both starand nonstar collaborators; see Table 8.

Table 9 gives summary statistics and correlations forall our defined variables. Table 10 reports our resultsfrom the event history analysis used to test H2 and H3.Model (1) includes all the control variables. Model (2) is

a partial model that includes the effects of the focaldesigner’s number of nonstar and star collaborators.Model (3) includes our network cohesion variables,and Model (4) includes the expertise similarity vari-ables. Model (5) is the full model, which we use to testall our hypotheses and then to interpret the results.Before testing our hypotheses, the effects of one set

of controls are noteworthy. We specifically control forcollaborators’ prominence in the field bymeasuring thenumber of other designers who have cited their work.The effects of these variables (for both star and nonstarcollaborations) are positive and highly significant,which suggests that the attention from other designers(the collaborators’ citers) that the focal designer’s thatcollaborators attract do have a positive impact on the staremergence of the focal designer. Because stars have (onaverage) an order of magnitude more citers, focal de-signers benefit more when collaborating with stars. Inaddition, because our Stars’ citers and Nonstars’ citers areproxies of the attention pool that a focal designer getsfrom the focal designer’s collaborators (and by defini-tion from their common third parties), these controlseffectively account for the contribution to star emer-gence of the possible increased attention that workingwith a prominent star might bring to a focal designer.Before testing H2 and H3, we retest H1 with the

sample used to test H2 andH3. In the full model (5), thecoefficients the coefficients for Direct ties to nonstars(0.426, p < 0.01) and for Direct ties to stars (0.673, p <0.01) are positive and also significant. More impor-tantly, the difference between these two collaborationeffects is statistically significant (p < 0.017). It is clearthat these sample data also support our baseline hy-pothesis that a star collaboration is significantly morebeneficial than a nonstar collaboration—that is, even forthe sample of designers who have collaborated at leastonce with both a star and a nonstar.The results from testing H2 confirm that increased

network cohesion has a negative (positive) effect on the

Table 8. Definition of Control Variables for Testing H2 and H3

Stars’ citers.As a control for the social attention that collaboratingwith a star may bring to the focal designer’s work, we define the Stars’ citers asthe number of individuals who have, up to time t, cited the patents of the stars with who a focal designer i has worked up to time t. We maysuppose that a star whose patents are cited bymore designers has accumulatedmore social attention. A starwith a larger following can bestowmore social attention on collaborators than can a star with a smaller following. Nonstars’ citers is defined analogously.

Stars’ direct ties. This variable captures the network size of all of the stars with which a focal designer i has worked up to time t. Wedefine Stars’ direct tiesit � | U

j∈Hit

Gjt |. Because collaboration consumes time and energy, stars with more collaborators might offer fewer benefits

(on average) to the focal designer. We define Nonstars’ direct ties similarly.Repeated collaborations with stars. To control for the closeness between the focal designer and the focal designer’s collaborators, we adopt

a strength-of-tie measure based on observations of repeated collaborations (Hansen 1999, McFayden and Cannella 2004, Fleming et al. 2007).Our variable measures a focal designer’s tendency to collaborate repeatedly with the same star(s). Formally, with cijt0 � 1 if designers i and j

collaborate at time t0, and with cijt0 � 0 otherwise. We define Repeated collaborations with starsit � ( ∑t−1t0�1

∑

j∈Hit

cijt0)/|H it |. The value of this measure

is one when the focal designer collaborates with a given star exactly once; the value increases when that focal designer works with the samestar 3 on subsequent projects. We measure Repeated collaborations with nonstars equivalently.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1162 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

outcomes of collaborations with nonstar (star) de-signers. Model (5) strongly supports both H2(i) andH2(ii). The negative and significant coefficient forNetwork cohesion with nonstars (−1.650, p < 0.01) in-dicates that having a more cohesive social network witha nonstar collaborator reduces, on average, the likelihoodof becoming a star—in line with H2(i). Thus, an increaseof one standard deviation in the value ofNetwork cohesionwith nonstars makes the focal designer’s likelihood ofemerging as a star 33% less likely (e(−1.650×0.24)− 1 =−0.33).In contrast, the positive and significant coefficient forNetwork cohesion with stars (0.371, p < 0.05) indicates thathaving a more cohesive social network with a star col-laborator increases, on average, the focal designer’slikelihood of becoming a star; this result accords withH2(ii). Indeed, an increase of one standard deviation inthe value ofNetwork cohesion with stars is associated witha 10% increase (e(0.371×0.26) − 1 = 0.10) in the likelihood ofthe focal designer becoming a star.

In the hypothesis section, we argued that a star col-laboration can provide access to diverse knowledge aswell as access to synthesis knowledge. As a consequence,it is important to emphasize that the positive effect ofNetwork cohesion with stars is the net result of two coun-tervailing forces: High levels ofNetwork cohesion with starsmay hinder a focal designer’s star emergence by limitingaccess to sources of diverse knowledge, but they benefitthe focal designer much more by supporting the ac-quisition of synthesis knowledge and skills, resulting ina positive overall effect (in line with H2(ii)). In contrast,high levels of Network cohesion with nonstars onlyhinder the access to diverse knowledge, resulting ina negative effect on star emergence (in line with H2(i)).

Hypothesis 3 posits that greater expertise similaritywith a nonstar (star) collaborator decreases (increases)the focal designer’s odds of emerging as a star. Our re-gression results strongly support H3. The negative andsignificant coefficient for expertise similarity with non-stars (−1.032, p < 0.01) indicates that designers who havegreater expertise similarity with such nonstar col-laborators thereby reduce significantly their likelihoodof becoming a star, in line with H3(i). Here an increaseof one standard deviation in expertise similarity withnonstars reduces the likelihood of emergence as astar by 27% (e(−1.032×0.31) − 1 = −0.27). In contrast, thepositive and significant coefficient for expertisesimilarity with stars (0.469, p < 0.01) indicates thatdesigners who collaborate with stars increase theirlikelihood of becoming a star by increasing theirexpertise similarity with star collaborators—in linewith H3(ii). An increase of one standard devia-tion in expertise similarity with stars increases theprobability of becoming a star by 15% (e(0.469×0.29) −1 = 0.15).As was true for the effects of network cohesion, the

positive effect of expertise similarity with stars on staremergence is the net result of two countervailing forcesin which an increase in expertise similarity limits accessto diverse sources of knowledge but advances thepractice and transfer of synthesis knowledge, over-all resulting in a higher likelihood of star emergence(in line with H3(ii)). The negative effect of expertisesimilarity with nonstars, in contrast, captures exclu-sively the negative effect of limited access to diverseknowledge in nonstar collaborations that high levels ofexpertise similarity induce (in line with H3(i)).

Table 9. Summary Statistics and Correlations for All Variables

Variable 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

1 Direct ties to nonstars (ln) 1.002 Direct ties to stars (ln) 0.27 1.003 Network cohesion with nonstars −0.06 −0.06 1.004 Network cohesion with stars 0.03 0.02 0.83 1.005 Expertise similarity with nonstars −0.42 −0.26 0.65 0.51 1.006 Expertise similarity with stars −0.37 −0.19 0.57 0.58 0.77 1.007 Nonstars’ citers (ln) 0.46 0.35 −0.17 −0.07 −0.34 −0.24 1.008 Stars’ citers (ln) 0.24 0.55 −0.04 0.00 −0.22 −0.17 0.26 1.009 Nonstars’ direct ties (ln) 0.66 0.39 −0.03 0.06 −0.30 −0.28 0.69 0.38 1.00

10 Stars’ direct ties (ln) 0.43 0.52 0.15 0.24 −0.11 −0.08 0.28 0.69 0.56 1.0011 Repeated collaboration with

nonstars0.04 0.07 0.08 0.08 0.05 0.04 0.10 0.02 0.02 −0.04 1.00

12 Repeated collaboration with stars 0.08 0.12 0.01 0.13 −0.11 0.10 0.12 0.09 0.04 0.01 0.71 1.0013 Assignee’s past patents (ln) −0.01 0.21 −0.08 −0.05 −0.08 −0.04 0.16 0.26 0.13 0.16 0.03 0.04 1.0014 Class diversity 0.38 0.29 −0.45 −0.35 −0.71 −0.56 0.35 0.19 0.29 0.09 0.23 0.34 0.07 1.0015 Mobility 0.36 0.12 −0.33 −0.29 −0.45 −0.46 0.22 0.11 0.27 0.11 −0.01 0.03 −0.23 0.47 1.0016 Location diversity 0.10 0.05 −0.14 −0.12 −0.18 −0.17 0.07 0.05 0.07 0.03 −0.01 0.01 0.02 0.14 0.14 1.0017 Patent stock year (ln) 0.10 0.10 0.01 0.02 −0.08 −0.05 0.13 0.27 0.17 0.20 0.01 0.03 0.08 0.08 0.00 0.02 1.00

Mean 1.59 0.91 0.47 0.51 0.63 0.73 2.05 4.81 2.19 2.89 2.05 2.42 3.95 3.47 1.5 1.05 6.67Standard deviation 0.66 0.33 0.24 0.26 0.31 0.29 1.96 1.11 0.9 0.86 2.18 2.88 1.95 2.99 1.1 0.25 0.72

Notes. N = 27,278. Correlations greater than |0.012| are significant at p < 0.05.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1163

The contrasting effects (for star versus nonstar col-laborations) associated with both network cohesionand expertise similarity provide additional evidencethat star and nonstar collaboration are different and arecharacterized by different mechanisms. More specifi-cally, in nonstar collaborations, the focal designerbenefits much more from open social structures andaccess to diverse sources of information, which canspark the generation of divergent ideas; in star col-laborations, more cohesive social structures and theopportunity for greater absorptive capacity with thestar collaborator provide the supporting conditions to

experience and learn creative synthesis. Moreover, wealso estimated an additional model that tests the effectsof both network cohesion and expertise similaritywithout distinguishing the quality of the collabora-tors (stars versus nonstars). Such a model yieldedsignificant negative coefficients for both overall net-work cohesion (−1.265, p < 0.01) and overall expertisesimilarity (−0.677, p < 0.01). This is both theoreticallyand empirically consistent with the fact that had wenot differentiated between different collaborators wewould have failed to recognize the transfer of the star’screative synthesis skills as a significant determinant of

Table 10. Proportional Hazard Model Predicting Star Emergence

Variable (1) (2) (3) (4) (5)

Direct ties to nonstars (H1) 0.496*** 0.506*** 0.339*** 0.426***(0.049) (0.05) (0.048) (0.051)

Direct ties to stars (H1) 0.502*** 0.654*** 0.594*** 0.673***(0.108) (0.092) (0.094) (0.089)

Network cohesion with nonstars (H2(i)) −2.145*** −1.650***(0.213) (0.205)

Network cohesion with stars (H2(ii)) 0.555*** 0.371**(0.17) (0.168)

Expertise similarity with nonstars (H3(i)) −1.715*** −1.032***(0.166) (0.161)

Expertise similarity with stars (H3(ii)) 0.394*** 0.469***(0.123) (0.131)

ControlsNonstars’ citers 0.370*** 0.349*** 0.324*** 0.319*** 0.311***

(0.017) (0.016) (0.017) (0.017) (0.017)

Stars’ citers 0.269*** 0.262*** 0.175*** 0.165*** 0.153***(0.035) (0.04) (0.038) (0.037) (0.038)

Nonstars’ direct ties −0.243*** −0.423*** −0.405*** −0.386*** −0.377***(0.041) (0.043) (0.044) (0.043) (0.044)

Stars’ direct ties −0.288*** −0.448*** −0.343*** −0.359*** −0.333***(0.045) (0.045) (0.046) (0.044) (0.045)

Repeated collaboration with nonstars 0.082*** 0.091*** 0.090*** 0.124*** 0.114***(0.02) (0.02) (0.017) (0.02) (0.018)

Repeated collaboration with stars −0.019 −0.024 −0.008 −0.036* −0.025(0.02) (0.02) (0.016) (0.019) (0.016)

Assignee’s past patents 0.115*** 0.130*** 0.114*** 0.127*** 0.116***(0.013) (0.013) (0.013) (0.013) (0.013)

Class diversity 0.198*** 0.174*** 0.143*** 0.128*** 0.127***(0.01) (0.01) (0.01) (0.012) (0.011)

Mobility 0.072** 0.056** 0.019 0.048** 0.028(0.029) (0.028) (0.025) (0.024) (0.024)

Location diversity 0.119 0.084 0.029 0.006 0.007(0.092) (0.092) (0.089) (0.086) (0.087)

Patent stock year 0.153*** 0.164*** 0.160*** 0.135*** 0.145***(0.037) (0.036) (0.037) (0.036) (0.037)

Cohort Yes Yes Yes Yes YesLog-likelihood −20,786 −20,701 −20,538 −20,579 −20,509

Notes. N = 27,278 designer–patent observations; 11,573 designers. Robust standard errors clustered byfocal designer are reported in parentheses. All models control for cohort dummy variables.

*p < 0.10; **p < 0.05; ***p < 0.01.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1164 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

star emergence; we would have rather only accountedfor the benefits of accessing diverse inputs as themain determinant to star emergence, which is themain mechanism by which the larger set of nonstarcollaborators affect star emergence.

Finally, we test our results for robustness againstalternative definitions of our dependent variable, theindicator of who is a star designer. In particular, we setthe cutoffs in the star definition to 1%, 2%, and 5% ofthe distribution of citations using a three-year rollingwindow; in addition, at the 2% cutoff, we comparerolling time windows of three, five, and seven years (aswell as no rolling window). These alternatives speci-fications confirm our results.5

4. Discussion and ConclusionEven though past research has recognized the impor-tance of stars as collaborators, we have only a limitedunderstanding of how the creative aspects of collab-orating with stars differ from those of collaboratingwith nonstars and, as a consequence, of how starcollaborations affect the emergence of future stars(Groysberg and Lee 2009, Azoulay et al. 2010, Oettl2012). Our paper addresses this gap in the literature bybuilding a theory of how the creative aspects of col-laborating with stars differ from their nonstar collab-oration counterparts; we do this by showing that suchdifferences result in a higher likelihood of star emer-gence and by identifying the conditions under whichcollaborating with creative stars is most effective. Ourfindings have a number of implications as this papercontributes to the literature on creativity managementand interpersonal collaboration.

The literature on creativity has long wonderedwhether individual creativity is contagious (Perry-Smithand Shalley 2003, McFayden and Cannella 2004). Ourpaper shows that it may very well be. We observe thatindividual creativity is enhanced not only by an in-novator collaborating with others to access diversesources of information, but also, and more lastingly, bycollaborating with others to learn creative synthesisskills. We argue that such a transfer of skills is morelikely to occur when a focal innovator collaborates withcreative stars.

This theory and its supporting empirical evidencealso identifiy a previously unreported link betweenthe emerging field of creative synthesis within thecreativity literature (Harvey 2014, 2015; Chen andAdamson 2015) and the extant literature on interper-sonal collaboration. We show how interpersonal col-laboration is the conduit for transmitting knowledgeabout creative synthesis. Absent such a pathway, onemust assume that creative synthesis skills emergespontaneously and are discovered independently bydifferent innovators. Having identified star collabora-tions as a conduit for the transmission of creative

synthesis skills, we see more systematic patterns in theemergence of such skills. We remark that a star-drivenemergence and dissemination of creative synthesisskills is consistent with anecdotal evidence that orga-nizations known to employ effective creative synthesisapproaches have, indeed, been led by creative stars.Examples include John Lasseter at Pixar AnimationStudios and Ferran Adria at the former El Bulli, a top-rated restaurant serving vanguard cuisine; these ex-tremely creative individuals put creative synthesis intoaction and helped others develop their own creativesynthesis skills (Svejenova et al. 2007, Catmull andWallace 2014).Our work speaks directly and primarily to the lit-

erature on interpersonal collaboration in creative set-tings by challenging the widely held assumption thatthere is no need to differentiate collaborators in terms ofquality—in other words, that collaboration is most ac-curately viewed as a homogeneous construct. However,identifying the sources of collaborators’ observed het-erogeneity is necessary to advance our understanding ofhow social network mechanisms actually operate. Forexample,McEvily et al. (2012) use seniority as a source ofheterogeneity in collaborative relationships and findthat lawyers benefit primarily from the collaborativeties, established at the beginning of their careers, withsenior lawyers. Our work advances the social networkliterature by establishing that variation in collaborators’relevant characteristics (i.e., their respective quality) hassignificant effects that merit more systematic study.Consider, for example, our findings that social andknowledge network constructs (viz., social networkcohesion and expertise similarity) affect star emergencecontingent upon the collaborator’s quality. Both of theseconstructs benefit the innovator collaboratingwith a starbecause they create an environment ideally suited to thepractice of creative synthesis and the transfer of thoseskills; yet such cohesion and similarity are detrimental tothe innovator collaborating with a nonstar because theylimit access to diverse sources of information. Thesecontrasting results call for scholars to exhibit cautionwhen tempted to view either social or knowledgestructures as homogenous. Our research demonstratesthat, at least in creative settings, the quality of collab-orators matters significantly and should, therefore, betaken into account, both theoretically and empirically,when studying collaborative knowledge networks.We argue along similar lines that these contrasting

effects reveal a logical connection between two streamsof the interpersonal collaboration literature that haveheretofore developed separately and have often beenconstrued as advancing opposed messages. Our find-ings on the role of both social network cohesion andexpertise similarity as facilitators in the transfer ofcreative synthesis skills in star collaborations is entirelyconsistent with previous work studying knowledge

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1165

transfer from the source to the recipient (Reagans andMcEvily 2003) and on the assimilation of cross-boundaryknowledge that would otherwise be difficult for therecipient to acquire (Tortoriello and Krackhardt 2010,Tortoriello et al. 2012). At the same time, our work isconsistent with the literature that focuses on the idea-generation aspects of collaboration. The detrimental ef-fects of social network cohesion and expertise similarityon star emergence in a nonstar collaboration mirrorpreviouswork,which has documented that exchanges inopen (as compared with closed) social network struc-tures are more likely to generate potentially creativeideas (Burt 2004, Fleming et al. 2007) as are more ex-changes of diverse information (Rodan and Galunic2004, Sosa 2011). We can reconcile these seeminglycontradictory effects by pointing to collaborator qualityas the factor that determines which account is moreapplicable. Hence, we can offer practical guidelines forwhen researchers and practitioners should apply eachtheory.

From an empirical standpoint, examining the effectsof interpersonal collaboration is a daunting exercise.There is one issue in particular, however, that warrantsfurther discussion. Recall that social attention mecha-nisms definitely factor into an innovator’s emergence asa star following collaborationwith a star.We, therefore,faced the empirical challenge of providing evidence forthe transfer of creative synthesis skills—from the star tothe focal designer—above and beyond any social at-tention mechanisms. We address that challenge byway of three different approaches. First, we introducea control for the increased social attention to whicha star exposes a focal innovator. We find, as expected,that the social attention resulting from both star andnonstar collaborators plays an important role in thefocal innovator’s emergence as a star; however, it doesnot fully explain the observed variation in the proba-bility of becoming a star. Second, we suppose thatdifferent creative mechanisms are triggered by col-laborators of different quality (here, stars versus non-stars). Indeed, we report empirical evidence that aninnovator exhibits a significant behavioral change thatis consistent with having learned the stars’ mechanismafter having collaborated with a star—a behavioralchange that is not observed when the focal innovatorcollaborates with a nonstar. More specifically, bytracking the reuse of new class combinations in thedesign patent database, we discover that star designersare more likely than nonstars to reuse their own novelideas—an indication that star designers dwell longeron making refinements to an initial invention and alsoconstruct more exemplars of a new paradigm (Harvey2014). We then use a DID approach to show that a focaldesigner who collaborates with a star (rather thana nonstar) is more likely to emulate star behavior asevidenced by a significant increase in the tendency of

such designers to reuse their previously created novelcombinations. (Employing a DID framework, we alsoconfirmed that a focal designer significantly increasesthe focal designer’s patenting productivity after col-laborating with a star—in line with the within-dyadmechanisms driving our hypotheses.) Third, on a con-ceptual level, our contingencies are somewhat moreconsistent with a knowledge-sharing mechanism thanwith an attention transfer mechanism. We know thatexpertise similarity improves the learning environment(and, thus, enables knowledge sharing) but also hindersaccess to the diverse information needed to generatenovel ideas. However, it is theoretically unclear howexpertise similarity would be associated with the transferof social attention.An important caveat to this study concerns its re-

liance on archival data derived from design patents, anapproach that entails several constraints common to allresearch based on patent activity. For example, our find-ings are based solely on successful collaborations—thatis, those resulting in a patent (or patents) being granted.Yet it is also possible, of course, for designer–star collab-orations to fail. Because our paper focuses on how col-laboration affects exceptional outcomes (i.e., becominga star), our conclusions shouldnot be invalidatedbyournothaving data related to failures. As usual, fully estab-lishing causality in archival data are complicated by thepossibility of unobserved characteristics driving boththe exposure to and the outcome of the treatment (be-coming a star). Although the nature of our data are notsuch that all doubt can be removed, our robustnesschecks provide evidence that our results should not beviewed as spurious.In reviewing the lives of eminent philosophers from

ancient China and Greece, Collins (1998) found thatphilosophers of comparable creative eminence tend toappear in the same generation. There is anecdotalevidence from historians and sociologists that greatfigures, early in their careers, studied under prominentindividuals of their era. Our paper is the first one thatemploys an extensive design patent data set as a meansto quantify the effects of collaborating with a star de-signer and thereby help explain the phenomenon ofstar emergence. A unique aspect of this study is ourfinding that knowledge transfer in such collaborationsis strongly affected by the quality of collaborators.A firm could use these results when seeking to improveits cultivation of future stars; doing so should increaseits design competitiveness and could well lead to ben-eficial changes in the firm’s focus and direction.

AcknowledgmentsThe authors thank the senior editor, Gautam Ahuja, and thereferees for constructive remarks throughout the reviewprocess. The authors appreciate the insightful feedback fromMartin Gargiulo, Henning Piezunka, Phanish Puranam and

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1166 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

Michelle Rogan on an earlier version of this paper. This paperhas also benefitted from comments received in researchseminars at the Academy of Management Meetings StrategicManagement Society Conferences and the Darden & Cam-bridge Judge Entrepreneurship and Innovation ResearchConference.

Endnotes1Another mechanism that could plausibly explain the differencesbetween star and nonstar collaborations is the possibility of socialattention being transferred from the star to the star’s collaborators(Simcoe and Waguespack 2011), which might (or might not) becorrelated with the transfer of creative skills we invoked when ar-guing for H1. Other innovators could view the focal innovator asmore capable simply because of the focal innovator’s association witha star. In this case, a star’s association lends public credibility to thefocal innovator because a star’s endorsement carries more weight inthe community than do endorsements from nonstars (Stewart 2005).It is likely that the “social attention transfer” mechanism, whichoperates outside the focal dyad, is mostly independent of the within-dyad mechanism of transferring tacit creative synthesis knowl-edge, which is why we do not discuss the former mechanism froma theoretical viewpoint. However, we account for it in our empiricalsection.2We adopt and extend the inventor-matching algorithm of Fleminget al. (2007) to disambiguate the names of designers in the designpatent database. That algorithm has been employed in a number ofrelated studies and has been extensively refined over the years(Trajtenberg et al. 2009, Singh and Fleming 2010).3 It is reassuring to know that our method correctly identified suchwell-known star designers as Yves Behar, Robert Brunner, Jony Ive,Steve Jobs, Frank Nuovo, and Phillippe Starck—to name (alphabeti-cally) just a few.4This proportion is not at odds with our definition of a star: the 7%figure is the average percentage of stars over all our observations,whereas the 2% figure is the “instantaneous” percentage for any giventime.5The coefficient forNetwork cohesion with stars loses significance in thecase of a 5% cutoff, three-year rolling window. Such a loss of sig-nificance should not be surprising because the more we relax thedefinition of stars, the more the effects that set them apart from thenonstars should wash out.

ReferencesAhmed S, Wallace KM, Blessing LT (2003) Understanding the dif-

ferences between how novice and experienced designers ap-proach design tasks. Res. Engrg. Design 14(1):1–11.

Ahuja G (2000) Collaboration networks, structural holes, and in-novation: A longitudinal study. Admin. Sci. Quart. 45(3):425–455.

Ahuja G, Katila R (2001) Technological acquisitions and the in-novation performance of acquiring firms: A longitudinal study.Strategic Management J. 22(3):197–220.

Ahuja G, Lampert C (2001) Entrepreneurship in the large corporation:A longitudinal study of how established firms create break-through inventions. Strategic Management J. 22(6–7):521–543.

Ainley P, Rainbird H, eds. (2014) Apprenticeship: Toward a New Par-adigm of Learning (Routledge, London).

Angrist J, Pischke J (2009) Mostly Harmless Econometrics: An Empiri-cist’s Companion (Princeton University Press, Princeton, NJ).

Argote L, McEvily B, Reagans R (2003) Managing knowledge inorganizations: An integrative framework and review of emergingthemes. Management Sci. 49(4):571–582.

Audia PG, Goncalo JA (2007) Past success and creativity over time:A study of inventors in the hard disk drive industry.ManagementSci. 53(1):1–15.

Azoulay P, Graff Zivin J, Wang J (2010) Superstar extinction.Quart J. Econom. 125(2):549–589.

Bartunek JM (1984) Changing interpretive schemes and organiza-tional restructuring: The example of a religious order.Admin. Sci.Quart. 29(3):355–372.

Bechky B (2003) Sharing meaning across communities: The trans-formation of understanding on a production floor. Organ. Sci.14(3):312–330.

Bédard J, ChiMTH (1992) Expertise.Current Directions Psych. Sci. 1(4):135–139.

Blossfeld HP, Rohwer G (1995) Techniques of Event History Mod-eling: New Approaches to Casual Analysis (Lawrence Erlbaum,Mahwah, NJ).

Bode C, Singh J, Rogan M (2015) Corporate social initiatives andemployee retention. Organ. Sci. 26(6):1702–1720.

Bordage G, Zacks R (1984) The structure of medical knowledge in thememories of medical students and general practitioners: Cate-gories and prototypes. Medical Ed. 18(6):406–416.

Brown T (2008) Design thinking. Harvard Bus. Rev. (June):84–92.Brunner R, Emery S, Hall R (2008) Do You Matter? How Great Design

Will Make People Love Your Company (FT Press, Upper SaddleRiver, NJ).

Buchanan R (1992) Wicked problems in design thinking. DesignIssues 8(2):5–21.

Burt RS (1992) Structural Holes: The Social Structure of Competition(Harvard University Press, Cambridge, MA).

Burt RS (2004) Structural holes and good ideas. Amer. J. Sociol. 10(2):349–399.

Burt RS (2005) Brokerage and Closure: An Introduction to Social Capital(Oxford University Press, Oxford, UK).

Carlile PR (2004) Transferring, translating, and transforming: Anintegrative framework for managing knowledge across bound-aries. Organ. Sci. 15(5):555–568.

Catmull E, Wallace A (2014) Creativity, Inc.: Overcoming the UnseenForces That Stand in the Way of True Inspiration (Random House,New York).

Chan TH, Mihm J, Sosa ME (2017) On styles in product design: Ananalysis of U.S. design patents. Management Sci. 64(3):1230–1249.

Chen J, Adamson C (2015) Innovation: Integration of random vari-ation and creative synthesis. Acad. Management Rev. 40(3):461–464.

Chi MTH, Feltovich PJ, Glaser R (1981) Categorization and repre-sentation of physics problems by experts and novices. CognitiveSci. 5(2):121–152.

Cohen WM, Levinthal DA (1990) Absorptive capacity: A new per-spective on learning and innovation. Admin. Sci. Quart. 35(1):128–152.

Coleman JS (1988) Social capital in the creation of human capital.Amer. J. Sociol. 94:S95–S120.

Collins R (1998) The Sociology of Philosophies: A Global Theory of In-tellectual Change (Belknap Press of Harvard University Press,Cambridge, MA).

Cox D (1972) Regression models and life-tables. J. Royal Statist. Soc.Series B 34(2):187–220.

Cross N (2003) The expertise of exceptional designers. Cross N,Edmonds E, eds. Expertise in Design (Creativity and CognitionPress, University of Technology, Sydney, Australia), 427–441.

Cross N (2004) Expertise in design: An overview. Design Stud. 25(5):427–441.

Csikszentmihalyi M (1996) Creativity: Flow and The Psychology ofDiscovery and Invention (HarperCollins Publishers, New York).

Dahan E, Mendelson H (2001) An extreme-value model of concepttesting. Management Sci. 47(1):102–116.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1167

Dahlander L, Piezunka H (2014) Open to suggestions: How orga-nizations elicit suggestions through proactive and reactive at-tention. Res. Policy 43(5):812–827.

Dougherty D (1992) Interpretive barriers to successful product in-novation in large firms. Organ. Sci. 3(2):179–202.

Eckert CM, Stacey MK (2003) Adaptation of sources of inspirationin knitwear design. Creativity Res. J. 15(4):355–384.

Ernst H (2001) Patent applications and subsequent changes of per-formance: Evidence from time-series cross-section analyses onthe firm level. Res. Policy 30(1):143–157.

Ernst H, Leptien C, Vitt J (2000) Inventors are not alike: The distri-bution of patenting output among industrial R&D personnel.IEEE Trans. Engrg. Management 47(2):184–199.

Fleming L (2001) Recombinant uncertainty in technological search.Management Sci. 47(1):117–132.

Fleming L, Mingo S, Chen D (2007) Collaborative brokerage, gen-erative creativity, and creative success. Admin. Sci. Quart. 52(3):443–475.

Fujiwara-Greve T, Greve HR (2000) Organizational ecology and jobmobility. Soc. Forces 79(2):547–585.

Gargiulo M, Ertug G, Galunic C (2009) The two faces of control:Network closure and individual performance among knowledgeworkers. Admin. Sci. Quart. 54(2):299–333.

Girotra K, Terwiesch C, Ulrich KT (2010) Idea generation and thequality of the best idea. Management Sci. 56(4):591–605.

Godart FC, Maddux WW, Shipilov AV, Galinsky AD (2015) Fashionwith a foreign flair: Professional experiences abroad facilitate thecreative innovations of organizations. Acad. Management J. 58(1):195–220.

Grigoriou K, Rothaermel FT (2014) Structural microfoundations ofinnovation: The role of relational stars. J. Management 40(2):586–615.

Groysberg B, Lee LE (2009) Hiring stars and their colleagues: Ex-ploration and exploitation in professional service firms. Organ.Sci. 20(4):740–758.

Groysberg B, Polzer JT, ElfenbeinHA (2011) Toomany cooks spoil thebroth: How high-status individuals decrease group effective-ness. Organ. Sci. 22(3):722–737.

Hansen MT (1999) The search-transfer problem: The role of weak tiesin sharing knowledge across organization subunits. Admin. Sci.Quart. 44(1):82–111.

Hargadon A, Sutton RI (1997) Technology brokering and innovationin a product development firm. Admin. Sci. Quart. 42(4):716–749.

Hargadon AB, Bechky BA (2006) When collections of creatives be-come creative collectives: A field study of problem solving atwork. Organ. Sci. 17(4):484–500.

Harvey S (2014) Creative synthesis: Exploring the process of extraor-dinary group creativity. Acad. Management Rev. 39(3):324–343.

Harvey S (2015) Synthesis in action: Response to Chen andAdamson.Acad. Management Rev. 40(3):464–466.

Ho CH (2001) Some phenomena of problem decomposition strategyfor design thinking: Differences between novices and experts.Design Stud. 22(1):27–45.

Iacus SM, King G, Porro G (2012) Causal inference without balancechecking: Coarsened exact matching. Political Anal. 20(1):1–24.

Jaffe AB, Trajtenberg M, Romer PM (2002) Patents, Citations, and In-novations: A Window on the Knowledge Economy (MIT Press,Cambridge, MA).

Janis IL (1972) Victims of Groupthink: A Psychological Study of Foreign-Policy Decisions and Fiascoes (Houghton Mifflin, Boston).

Katila R,Ahuja G (2002) Something old, something new:A longitudinalstudy of search behavior and new product introduction. Acad.Management J. 45(6):1183–1194.

Kogut B, Zander U (1992) Knowledge of the firm, combinative capa-bilities, and the replication of technology. Organ. Sci. 3(3):383–397.

Lloyd P, Scott P (1994) Discovering the design problem. Design Stud.15(2):125–140.

Long-Lingo EL, O’Mahony S (2010) Nexuswork: Brokerage on creativeprojects. Admin. Sci. Quart. 55(1):47–81.

Maddux WW, Galinsky AD (2009) Cultural borders and mentalbarriers: The relationship between living abroad and creativity.J. Personality Soc. Psych. 96(5):1047–1061.

McEvily B, Jaffee J, TortorielloM (2012)Not all bridging ties are equal:Network imprinting and firm growth in the Nashville legalindustry, 1933–1978. Organ. Sci. 23(2):547–563.

McFayden MA, Cannella AA (2004) Social capital and knowledgecreation: Diminishing returns of the number and strength ofexchange relationships. Acad. Management J. 47(5):735–746.

Mednick SA (1962) The associative basis of the creative process.Psych. Rev. 69(3):220–232.

Nicolini D, Mengis J, Swan J (2012) Understanding the role of objectsin cross-disciplinary collaboration. Organ. Sci. 23(3):612–629.

Obstfeld D (2005) Social networks, the tertius iungens orientation, andinvolvement in innovation. Admin. Sci. Quart. 50(1):100–130.

Oettl A (2012) Reconceptualizing stars: Scientist helpfulness and peerperformance. Management Sci. 58(6):1122–1140.

Perry-Smith JE (2006) Social yet creative: The role of social relationshipsin facilitating individual creativity. Acad. Management J. 49(1):85–101.

Perry-Smith JE, Shalley CE (2003) The social side of creativity: A staticand dynamic social network perspective. Acad. Management Rev.28(1):89–107.

Reagans R (2005) Preferences, identity, and competition: Predictingtie strength from demographic data. Management Sci. 51(9):1374–1383.

Reagans R, McEvily B (2003) Network structure and knowledgetransfer: The effects of cohesion and range. Admin. Sci. Quart.48(2):240–267.

Reagans R, Singh PV, Krishnan R (2015) Forgotten third parties:Analyzing the contingent association between unshared thirdparties, knowledge overlap, and knowledge transfer relation-ships with outsiders. Organ. Sci. 26(5):1400–1414.

Rodan S, Galunic C (2004) More than network structure: Howknowledge heterogeneity influences managerial performanceand innovativeness. Strategic Management J. 25(6):541–562.

Schumpeter JA (1934) The Theory of Economic Development (HarvardUniversity Press, Cambridge, MA).

Seidel VP, O’Mahony S (2014) Managing the repertoire: Stories,metaphors, prototypes, and concept coherence in product in-novation. Organ. Sci. 25(3):691–712.

Simcoe TS, Waguespack DM (2011) Status, quality, and attention:What’s in a (missing) name? Management Sci. 57(2):274–290.

Simonton DK (1999) Origins of Genius: Darwinian Perspectives onCreativity (Oxford University Press, Oxford, UK).

Singh J, Agrawal A (2010) Recruiting for ideas: How firms exploit theprior inventions of new hires. Management Sci. 57(1):129–150.

Singh J, Fleming L (2010) Lone inventors as sources of breakthroughs:Myth or reality? Management Sci. 56(1):41–56.

Sosa ME (2011) Where do creative interactions come from? The roleof tie content and social networks. Organ. Sci. 22(1):1–21.

Sosa ME (2014) Realizing the need for rework: From task interde-pendence to social networks. Production Oper. Management 23(8):1312–1331.

Stewart JA (2005) Social status in an open-source community. Amer.Sociol. Rev. 70(5):823–842.

Sutton R, Hargadon A (1996) Brainstorming groups in context : Effec-tiveness in a product design firm. Admin. Sci. Quart. 41(4):685–718.

Svejenova S,MazzaC, PlanellasM (2007) Cooking up change in hautecuisine: Ferran Adria as an institutional entrepreneur. J. Organ.Behav. 28(5):539–561.

Tortoriello M (2014) The social underpinnings of absorptive capacity:The moderating effects of structural holes on innovation gen-eration based on external knowledge. Strategic Management J.36(4):586–597.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar Collaborators1168 Organization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS

Tortoriello M, Krackhardt D (2010) Activating cross-boundaryknowledge: The role of simmelian ties in the generation of in-novations. Acad. Management J. 53(1):167–181.

Tortoriello M, McEvily B, Krackhardt D (2015) Being a catalyst ofinnovation: The role of knowledge diversity and network clo-sure. Organ. Sci. 26(2):423–438.

Tortoriello M, Reagans R, McEvily B (2012) Bridging the knowledgegap: The influence of strong ties, network cohesion, and networkrange on the transfer of knowledge between organizational units.Organ. Sci. 23(4):1024–1039.

Trajtenberg M (1990) A penny for your quotes: Patent citations andthe value of innovations. RAND J. Econom. 21(1):172–187.

Trajtenberg M, Shiff G, Melamed R (2009) The “names game”:Harnessing inventors, patent data for economic research. Ann.Econom. Statist. 93/94:79–108.

Ulrich KT (2011) Design: Creation of Artifacts in Society (University ofPennsylvania, Philadelphia).

USPTO (2015) United States Patent and Trademark Office. AccessedJanuary 10, 2018, https://www.uspto.gov/patents-getting-started/general-information-concerning-patents.

Uzzi B, Spiro J (2005) Collaboration and creativity: The small worldproblem. Amer. J. Sociol. 111(2):447–504.

Uzzi B, Mukherjee S, Stringer M, Jones B (2013) Atypical combina-tions and scientific impact. Science 342(6157):468–472.

Vera D, CrossanM (2005) Improvisation and innovative performancein teams. Organ. Sci. 16(3):203–224.

vonHippel E, TyreM (1995) How “learning by doing” is done: Problemidentification in novel process equipment. Res. Policy 24(1):1–12.

Weisberg RW (1999) Creativity and knowledge: A challenge of the-ories. Sternberg RJ, O’Hara LA, eds. Handbook of Creativity(Cambridge University Press, Cambridge, UK), 226–250.

Wuchty S, Jones BF, Uzzi B (2007) The increasing dominance of teamsin production of knowledge. Science 316(5827):1036–1039.

Zucker LG, Darby MR, Brewer MB (1998) Intellectual capital andthe birth of U.S. biotechnology enterprises. Amer. Econom. Rev.88(1):290–306.

Haibo Liu is an assistant professor of management at theSchool of Business, University of California, Riverside. Hereceived his PhD degree from INSEAD. His research focuseson innovation and organizational learning with an emphasison extreme performance, such as extremely creative andinfluential designers and successes and failures of franchiseorganizations.

Jürgen Mihm is an associate professor of technology andoperations management at INSEAD. He received his PhD de-gree from WHU’s Otto Beisheim School of Management, Val-lendar, Germany. His research investigates coordination andincentive issues in new product development organizations.

Manuel E. Sosa is an associate professor of technology andoperations management at INSEAD. He received his PhDdegree from Massachusetts Institute of Technology. His re-search investigates coordination and innovation networks innew product development organizations.

Liu et al.: Where Do Stars Come From? Star vs. Nonstar CollaboratorsOrganization Science, 2018, vol. 29, no. 6, pp. 1149–1169, © 2018 INFORMS 1169