networks, creativity, and time: staying creative through

NETWORKS, CREATIVITY, AND TIME:STAYING CREATIVE THROUGH BROKERAGE AND

NETWORK REJUVENATION

GIUSEPPE SODABocconi University

PIER VITTORIO MANNUCCILondon Business School

RONALD S. BURTUniversity of Chicago

In this paper, we adopt a dynamic perspective onnetworks and creativity to propose thatthe oft-theorized creative benefits of open networks and heterogeneous content are lesslikely to be accrued over time if the network is stable. Specifically, we hypothesize thatopen networks and content heterogeneity will have a more positive effect on creativitywhennetwork stability is low.Webase our prediction on the fact that, over time, networkstability begets cognitive rigidity and social rigidity, thus limiting individuals’ ability tomake use of the creative advantages provided by open networks and heterogeneous con-tent. On the contrary, new ties bring a positive “shock” that pushes individuals in the net-work to change the way they organize and process knowledge, as well as the way theyinteract and collaborate—a shock that enables creators to accrue the creative advantagesprovidedbyopennetworkstructuresandheterogeneouscontent.Wetestandfindsupportfor our theory in a study on the core artists who worked on the TV series Doctor Whobetween 1963 and 2014.

Nurturingandpreservingindividualemployees’cre-ativityover timehasbecomeincreasingly important forfirm innovation and success (Amabile & Pratt, 2016;Anderson, Poto�cnik, & Zhou, 2014; Zhou & Hoever,2014). In today’scompetitiveenvironment, in fact, pro-ducing a single creative contribution might not beenough: as the cycles of innovation–exploitation areshortening, bumpy dynamics in employees’ creativitycan generate negative performance consequences andfinancial troubles for organizations (Ahuja & Lampert,2001; Tortoriello & Krackhardt, 2010). This trendimposes on individuals and organizations alike tofind a way to guarantee a sustainable flow of ideas

over time—something that, in reality, seems extremelychallenging (Simonton, 1984b, 1988).

Individuals’ ability to stay creative over time isshapedbymanyfactors,but theirsocialsystemofrela-tionshipsplaysaparticularlycentralrole(Brass,1995;Brothers, 2018; Burt, 2004; Simonton, 1984a; Perry-Smith & Mannucci, 2017). Research has shown thatindividuals whose network structure (i.e., whomthey talk to) or network content (i.e., what they areexposed to) gives them access to nonredundant per-spectivesandideasaremorelikelytogeneratecreativeideas (Burt, 2004; Carnabuci & Di�oszegi, 2015; Flem-ing, Mingo, & Chen, 2007; Rodan & Galunic, 2004).Specifically, thisnonredundancycancomefromhav-ing a network rich in structural holes, bridging other-wise disconnected social circles (Burt, 2004; Burt &Soda, 2017), or from a network that provides accessto diverse, heterogeneous knowledge (Aral & VanAlstyne, 2011; Goldberg, Srivastava, Manian, Mon-roe, & Potts, 2016; Zaheer & Soda, 2009).

Overall, thiswould suggest that the recipe tomain-taina certain level of creativityover time is tokeepupnetwork nonredundancy, in terms of both structureand content. However, this is no easy task. Ties

The authors extend their gratitude to associate editorGurneeta Vasudeva and three anonymous reviewers fortheir invaluable feedback throughout the review process.We also thank the participants at the INSEAD NetworkEvolution Conference, the ION Social Networks Confer-ence at Gatton College, University of Kentucky, and theAcademy of Management Conference for their helpfulcomments and suggestions on previous drafts. Finally,the authors would like to thank Emanuele Pezzani andXiaoming Sun for their precious help with data collection.

1164

Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder's express

written permission. Users may print, download, or email articles for individual use only.

rAcademy of Management Journal

2021, Vol. 64, No. 4, 1164–1190.

https://doi.org/10.5465/amj.2019.1209

bridgingstructuralholesarefragile(Baum,McEvily,&Rowley, 2012; Burt, 2002; Burt &Merluzzi, 2016; Sto-vel, Golub, &Milgrom, 2011), and are thus character-ized by diminishing returns over time (Soda, Usai, &Zaheer, 2004). Similarly, knowledgeand informationtend to homogenize quicklywithin a network (Aral &VanAlstyne, 2011).

This poses a conundrum: If structural holes andcontent heterogeneity are difficult to maintain andtheir creative returns decay, what is the best strategytokeep them “alive” and conducive to creative ideas?Despite the recognition that individuals’ ability toaccrue advantages from their network is contingenton how they reconfigure the network over time(Burt,Kilduff,&Tasselli,2013;Cannella&McFadyen,2016), networks and creativity scholars so far havemainly looked at the benefits of having a certain net-work structure at a given point in time.

In this paper, we attempt to solve this issue by pro-posingthatthecreativebenefitsofopennetworkstruc-tures andheterogeneous content at any givenpoint intime are less likely to be accrued if actors do not addnew ties. Specifically, we theorize that structuralholesandcontentheterogeneitywillhaveamorepos-itiveeffectonindividualcreativitywhennetworksta-bility is low—that is, when individuals rejuvenateover time the composition of their network by addingnewties.Webase thispredictionontheidea that,overtime, stable networks result in a homogenization andentrenchmentofcognitivestructures (cognitive rigid-ity) and interaction patterns (social rigidity). If net-work composition does not change, over time, thecreative advantages provided by structural holes andheterogeneous content will thus be lost, due to theincreased mental closure toward new perspectivesandknowledgeandtheincreasedrigidityincoordina-tion and collaboration patterns. On the contrary, theaddition of new ties introduces a positive “shock”that pushes individuals in the network to change theway they look at and process knowledge, as well asthe way they interact and collaborate (Ferriani, Cat-tani, & Baden-Fuller, 2009; Rand,Arbesman,&Chris-takis, 2011; Shirado & Christakis, 2017). These new,fresh outlooks and collaboration patterns in turnenable them to accrue the creative advantages pro-vided by opennetwork structures and heterogeneouscontent.

Wetest and findsupport forourhypotheses ina set-ting specifically suited for our research question: thepopulationof core artists behind theBritishTV seriesDoctor Who, the longest-running sci-fi series in theworld (e.g., Moffat, 2017; Moran, 2007; Petruzzella,2017).

THEORY AND HYPOTHESES

Network Structure, Network Content,and Creativity

Creativity occurs when a person breaks free fromtheir previous way of thinking, which can happenfor a variety of individual as well as social reasons.Network theory focuses on the social: breaking freefrom your usual ways is more likely when you areexposed to people whose opinions and behaviors aredifferent fromyourown.Others’opinions andbehav-iors canbedismissedas irrelevant, or engaged so as toseewhatyouknowinanewway.When thishappens,new ideas arise like “productive accidents”: the wayone person makes money with product X becomes arevelation to a person selling product Y, so a newway to distribute productY is born.

Networktheoryhasarguedandfoundthat,themoredisconnected the people in an individual’s network,themoreheterogeneoustheirknowledgeandperspec-tives, and thus the higher the chance of a productiveaccident inwhichdifferingopinionsorbehaviorscol-lide toproducea good idea (Burt, 2004; Fleming et al.,2007; Hargadon & Sutton, 1997; Lingo & O’Mahony,2010). For example, Picasso’s innovations in Cubismwere vastly the byproduct of him being embedded ina diverse, disconnected network (Sgourev, 2013). Onthe contrary, themore homogenous the opinions andbehaviors in a network, the lower the chance of crea-tive accidents. Highly interconnected people aredrawn together by similarity in their opinions andbehaviors, and socialize one another into even moresimilaropinionsandbehaviors (Festinger,Schachter,& Back, 1950; Katz & Lazarsfeld, 1955). A closed net-work of interconnected colleagues implies limitedvariation inopinions andpractices, aswell asempha-sis within the network on the propriety of discussionlimited to the socially accepted opinions andpractices.

In the aforementioned studies, network openness(closure) was often conflated with knowledge andcontent heterogeneity (homogeneity)—a well-known creativity booster (hinderer) (e.g., Mannucci& Yong, 2018; Taylor & Greve, 2006). The underlyingassumption was that structure always embodies andreflects content, and thus structural holes reflect het-erogeneous content and closure reflects homogenouscontent. Recently, however, this equation has beencalled intoquestion,with scholars arguing and show-ing that structure does not necessarily embody con-tent, and thus the two dimensions, while deeplyinterconnected, can also act independently and sohave similar yet distinct effects. For example, Zaheer

2021 Soda, Mannucci, and Burt 1165

and Soda (2009) used the content of TV scripts to cat-egorize content heterogeneity in the networks of TVproduction teams, and showed that network contenthomogeneity and structural holes had separate andeven opposite effects on team performance. Aral andVan Alstyne (2011) used the information content ofemail messages among people in an organizationand showed that, while networks bridging structuralholes do carry more diverse information, networkand information diversity have separate positiveeffects on performance. Furthermore, there is moreto the effect of information heterogeneity than is cap-tured by network structure: performance is enhancedbydiverseinformationprovidedeitherbyanopennet-work, or by one very strong connection (“diversity-bandwidth trade-off”). Finally, Goldberg and col-leagues (2016) analyzed email networks and contentover a five-year period among several hundredemployees and discovered a trade-off between net-work and content homogeneity: people in closednet-works received less positive job evaluations whenthey exchanged information using a language thatwas homogenous in terms of style and topics to theircolleagues’, while people in open networks obtainedmore positive job evaluations when they exhibitedthis language homogeneity. Building on theseinsights, networks researchers have argued that thebenefits of brokerage go beyond content: brokerageprovides a vision advantage, a flexibility in cognitionandpractices thatallowsbrokers to“seethings,” spot-ting connections that others do not see (Burt, 2008;Burt & Soda, 2017). This issue is particularly relevantfor creativity: in the words of Steve Jobs, “When youask creative people how they did something, theyfeel a little guilty because they didn’t really do it,they just saw something. It seemed obvious to themafter a while” (Wolf, 1996, “GrokKing design,”para. 3).

While creativity scholars have studied the effectsof these twodimensionsinisolation, theyhaveyet toprecisely disentangle the creative consequences ofnetwork structure (whom individuals talk to andcollaboratewith) fromtheeffectsofnetworkcontent(what type of knowledge they are exposed to). Con-sidering themtogether is thusneeded tounderstandwhether the effect of structure on creativity isentirely dependent on content (e.g., Rodan &Galunic,2004),or if thecreativebenefitsofopennet-works go above and beyond the effect of content, inthat they provide a vision advantage (Burt, 2004).We thus focus on network structure and networkcontent as separate predictors in our theorizingand analysis.

The Moderating Role of Network Stability

Extant studies on networks and creativity havemostly adopted an agnostic view on the role of net-work change in shaping the creative returns of non-redundant network structure and content. Forexample, papers looking at creative outcomes suchas academic publications or patents conceptualizecreativity as the aggregate sumof these outcomespro-ducedwithin a certain time period (e.g., three years),even when they adopt a longitudinal angle (e.g.,Burt, 2004; Fleming et al., 2007; McFadyen & Can-nella, 2004). By focusing on aggregated patterns, welose sight of how network composition changes orremains the same over the years—something thatvaries significantly across creative individuals(Phelps, Heidl, & Wadhwa, 2012; Simonton, 1988,1997). Adopting a dynamic perspective is highlyimportant because it puts into question whether thecreative benefits provided by network nonredund-ancy can be taken for granted also over time. Extantresearch shows, in fact, that the benefits of open net-works and heterogeneous content are more easilyaccrued in the short term than in the long term (Aral& Van Alstyne, 2011; Baum et al., 2012; Burt, 2002;Soda et al., 2004). Brokerage positions are fragile(Burt, 2002; Stovel et al., 2011) and subject to change(Burt & Merluzzi, 2016; Sasovova, Mehra, Borgatti, &Schippers, 2010), and knowledge and content tendto homogenize quickly and have diminishing returns(Aral & VanAlstyne, 2011).

The question thus becomes whether an open net-work would yield creative advantages over time andunder which conditions. We argue that answeringthis question requires considering the compositionof the network and how it evolves—that is, networkstability. We define “network stability” as the degreeto which individuals maintain their existing ties oradd new ones. Brokers can in fact maintain theiropennetworkstructureseitherbyretainingtheirexist-ing brokerage positions with the same people, or bycreating new ones through the addition of new ties(Sasovova et al., 2010). These strategies have verydif-ferent implications for the accrual of creative returnsover time.

Scholars have argued that network stability canhave both positive and negative effects on socialexchanges and, consequently, performance. On theone side, network stability provides coordinationand communication advantages (Ferriani et al.,2009; Perretti & Negro, 2006) that are beneficial forthe efficiency of social interactions, especially oncomplex tasks (Ferriani et al., 2009; Soda et al.,

1166 Academy of Management Journal August

2004), and can thus facilitate an actor’s ability toexchange knowledge and execute their work. More-over, recurring ties are “old timers” who possessmore expertise in the task and in the social domainmorebroadly, something that their contacts canbene-fit from (Perretti & Negro, 2006, 2007). On the otherside, however, network stability can alsomake socialinteractions excessively rigid and routinized,makingteams increasingly rely on the same exchange andinteractionpatterns,withoutexploringnewones(Fer-riani et al., 2009; Soda et al., 2004). On the contrary,new ties can “shake up” existing cognitive patternsand thus push individuals to reconsider their waysof mentally organizing and using knowledge, as wellas engender the reshaping of collaboration patternsthroughtheir sheerpresence (Morrison,2002;Perretti& Negro, 2006). These advantages are present regard-lessofwhethernewtiesbringnewcontent(oneoftheiroft-argued, yet never tested, advantages) or not (Shir-ado&Christakis,2017).Moreover, researchhascalledinto questiononeof the benefits of stability—namely,that it improves collaboration quality. In a series oflarge-scale experiments, scholars have shown thatnetworks with high stability yield no collaborationbenefits (Traulsen, Semmann, Sommerfeld, Kram-beck, & Milinski, 2010), and that networks that arenot rewired through the addition of new ties actuallyseecooperation sharplydeclineovertime (Randetal.,2011).

Weargue that,whenitcomesto themoderating roleofstabilityontherelationshipbetweennonredundantnetwork structure and content on creativity, thedownsides of stability will prevail. Specifically, wepropose that thiswill happenbecausenetwork stabil-ity engenders homogenization and entrenchment of(a) mental models and structures (cognitive rigidity)and (b) interaction patterns (social rigidity)—allwhich severelyundermine, to thepoint ofpotentiallyeliminating, the creative advantages providedbybro-kerage and content heterogeneity.

Network structure. As mentioned above, oneadvantage of brokerage beyond access to heteroge-neous content is premised on having contacts thatcome from different social circles, and that thus holddiverse worldviews and mental models. This pro-vides the broker with a diversity of viewpoints thatallows them to look at things in different ways andadoptmultiple angles to address the same issue, thusfostering cognitive flexibility (Burt, 2004). If thosecontacts remain the same over time, however,mentalmodels andcognitive structures are likely tohomoge-nize and become more rigid (Morrison, 2002; Sodaet al., 2004). This increased cognitive rigidity will

hamper the vision advantages that brokers enjoythankstotheirposition(Burt,2004,2008), thusdimin-ishing their ability to generate creative ideas. More-over, network stability is also likely to reduceindividuals’ability toengagewithandevenrecognizedifferentpointofviews.Researchhasinfact theorizedand shown that highly stable collectives tend to becharacterized by rigidity and resistance to new per-spectives and approaches (Dunbar, 1993; Perretti &Negro, 2006, 2007; Rollag, 2004; Skilton & Dooley,2010; Sytch & Tatarynowicz, 2014). Having an opennetwork with highly stable membership would thusresult instructuralholesprovidinglittle tonocreativeadvantage: brokers will increasingly fixate on theirwaysofdoing things, thus limiting their ability to rec-ognizeandutilizethenonredundantperspectivesandviews they are exposed to, to the point of ignoringthem entirely.

Newties,ontheotherside,stimulatetheadoptionofnew perspectives and ways of seeing (Ferriani et al.,2009; Morrison, 2002), thus fostering brokers’ visionadvantage and ability to successfully apply oldnotions in different ways. This advantage of new tiesisnotpremisedontheirsocialcapital,andspecificallyon them bringing new content: it is instead rooted inthe fact that they do not possess the shared mentalmodels and views that characterize the existing net-worktheyareentering.Becauseofthis, theyaskissuesthat others do not see and take for granted (March,1991). It is precisely their “naïvet�e” that ensures thatindividuals in the network reconsider their ways ofdoingthingsandrestructuretheirmentalmodels.Net-work reconfiguration should thus benefits brokers byincreasing the likelihood that they consider newframes and “lenses” to see the world, allowing themto recognize new opportunities and new potentialrecombinations, even within the same knowledgebase. Moreover, being exposed to “new” actors whobelong topreviouslyunexploredsocialcirclesshouldincrease an individual’s psychological readiness tonew perspectives and mental frames (Perry-Smith,2014). This reasoning is consistent with both empiri-cal and anecdotal evidence on how being exposed tosomething or someone new leads to the reconfigura-tion of mental structures. Taking on unusual workassignments (Kleinbaum, 2012),migrating to a differ-entcountry(e.g.,Hunt&Gauthier-Loiselle,2010),andinteracting with people from different cultures (e.g.,Maddux & Galinsky, 2009) have been shown to favorthese processes. Similarly, creatives at Pixar identifythe moment they began working with Brad Bird, thefirst director to join them as an “outsider” after hisexperiences atWarnerBros andFox, as a keymoment


for their continued creativity, as his addition forcedthem to change the ways they looked at things (Rao,Sutton, &Webb, 2008).

Another advantage of structural holes resides in thefact that interactions with disconnected individualsincrease the chance of creative friction (Burt, 2004)because of the sheer fact of interactingwith others thathavedifferentmodesofwork.Having a stablenetwork,however, can lead to an increased social rigidity androutinization of interaction patterns, both in terms ofwhom actors interact with and how they interact. Thisroutinization will result in individuals becomingentrenched and fixated in their ways of collaboratingand coordinating (Morrison, 2002; Perretti & Negro,2006).Theywillthusbecomeblindtonewwaysofcoor-dinatingandworkingtogether, losing inpartorentirelythe potential creative sparks that result from having toreconsider one’s interaction and collaboration habits(Ferriani et al., 2009; Skilton&Dooley, 2010).

On the contrary, the addition of new ties to an exist-ing network represents a positive shock that pushesindividuals in the network to reconsider theway theywork together and coordinate. Once again, the abilityof new ties to generate this shock is not premised onthe novelty and nonredundancy of content they candirectly provide. The mere addition of new people isinfactenoughtoforceotherindividuals inthenetworkto reconsider theway theydo things, if only to explainthem to the newcomers. In so doing, they are forced toexplore, cognitively or practically, new coordinationpaths, thuschanging theoldwaysand“shaking thingsup.” Consistent with this reasoning, Shirado andChristakis (2017) have shown that even the additionof new agents without any competence (such as“noisy” bots) to a network is enough to change thewaynetworkmembersinteractandorganizetoexecutecomplex tasks. The addition of new agents shapes notonly the interactions of other actors with them, butalso the way other actors interact among themselves,changing their coordination strategies and routines.

All in all, these arguments suggest that networkreconfiguration to create new structural holes repre-sentsamoreeffectivestrategyforaccruingthecreativebenefits of structural holes compared to the stabiliza-tionofexistingholes.Wethusexpectnetworkstabilitytoweaken the creative benefits provided byopennet-works,whereasweexpectchangesinnetworkcompo-sition to strengthen them.

Hypothesis 1. Network stability moderates the rela-tionship between open networks and creativity. Thepositive association between open networks and crea-tivity is weaker in more stable networks, and strongerin less stable ones.

Contentheterogeneity.Asimilarreasoningappliesto the heterogeneous content shared through the net-work. One creative advantage of the exposure to het-erogeneous content is premised on providing new“raw materials” that fuel the recombinatory processat theheart of the generationofnovel anduseful ideas(Campbell, 1960; Mannucci & Yong, 2018; Taylor &Greve,2006).Maintainingthesamenetworkcomposi-tionovertimecanleadtoheterogeneouscontenttoagemore quickly and become obsolete (Aral & VanAlstyne,2011), thuslimitingboththenoveltyanduse-fulness of generated ideas (Soda et al., 2004). Further-more, the likelihood of content to change over time,both in terms of composition and how it is structuredand organized, is lower if the network is stable. Thecreative returns of heterogeneous content are likelyto diminish over time if it does not change, as thereare only so many creative permutations that you canderive fromthesamecontentandcognitivestructures(Campbell, 1960; Simonton, 2003). Finally, networkstability is likely to engender rigidity inmental struc-tures, hampering even the mere ability to recognizeand use new content (Schulz-Hardt, Frey, L€uthgens,& Moscovici, 2000; Scholten, van Knippenberg, Nij-stad, & De Dreu, 2007). Always interacting with thesame alters creates inert cognitive structures, whichin turn reduces individuals’ ability to identify andwillingness to integrate diverse knowledge and con-tent (Morrison, 2002; Skilton & Dooley, 2010). Thisresistance means that, even if exposed to heteroge-neous content, individuals in stable networkswill beless receptive to it andmay ignore it entirely (Ferrianiet al., 2009; Perry-Smith, 2014).

On the contrary, reconfiguring the network by add-ingnewtiesshouldensure that theadvantagesofferedby heterogeneous content are accrued. New ties aremore likely to bring points of view (Morrison, 2002;Perretti & Negro, 2006, 2007; Sytch & Tatarynowicz,2014), and can thus shake up mental structures,changingthewaythecreator looksatavailableknowl-edge.The“elementsofingenuity”broughtbynewties(Perretti & Negro, 2006: 761) shake up individuals’mental structures and pressure them to reconsiderwhat they thought they knew and look at it in newways. Moreover, being exposed to “new” actors(belonging to previously unexplored social circles)can increase an individual’s psychological readinessto attend to and use heterogeneous, diverse content(Perry-Smith, 2014). Consistently, research hasshown that the addition of uninformed individualstosocialgroupsensures thatall information isequallyattended to, eliminating biases toward dominantpoints of view and content (Couzin et al., 2011).


Anotherreasonwhystabilitycouldhampertherela-tionship between heterogeneous content and creativ-ity liesinthe fact that itmaydiminishthechances thatthis content is actuallyshared.The routines andoper-ating procedures for coordination and knowledgesharingshapealsothetypeofknowledgethatisshared(Hansen, 1999; Reagans & McEvily, 2003). The rigid,routinized procedures that characterize stable net-works thus lead to the sharing of commonly ownedknowledge, turning the advantage of having accessto heterogeneous knowledge from actual to potentialand consequently reducing its creative returns.

Overall, these arguments suggest that network sta-bility should weaken the creative benefits providedby heterogeneous content, whereas changes in net-work composition should strengthen them. Thus, wepredict:

Hypothesis 2. Network stability moderates the rela-tionship between content heterogeneity and creativity.The positive association between the exposure to het-erogeneous content and creativity is weaker in morestable networks, and stronger in less stable ones.

METHODS

Setting: The Doctor Who Production World

Testing our hypotheses required a research settingcharacterized by creatives who continuously engagein collaborations to generate creative outcomes. Wefound such a setting in the network of creativesinvolved in the realization of the episodes of DoctorWho, a British science-fiction television show andthe longest running in the world. Since its launch in1963,DoctorWhohas been a groundbreaking successin British television (Howe, Stammers, & Walker,1994). It is currently broadcasted in more than 50countries and is one of the top grossing shows pro-duced by the BBC (O’Connor, 2008). The series tellsthe adventures of an extra-terrestrial being called“the Doctor” who explores the universe thanks to aspaceship called the TARDIS, which allows him totravelinspaceandtime.Heis joinedinhisadventuresbyavarietyofcompanions,whohelphimfightingfoesin different planets, times, and civilizations.

The increased importance, scope, and success ofDoctor Who over the years has led the showrunnersto elaborate anarrativeploy tokeep the showrunningevenwhentheactor interpreting theDoctordecides toquit:whenhe ismortallywounded, theDoctor’sbodyregeneratestotakeonadifferentappearance.Regener-ation is thusat thecoreofDoctorWho in termsofchar-acters, plots, and themes. The showhas attracteda lot

of praise for its creativity and ability to reinvent itself(e.g., Moran, 2007; Petruzzella, 2017). For example,this is how StevenMoffat, one of the most successfulshowrunners inBritish television, described the clas-sic series ofDoctorWho in a recent interview:

Theclassic series … hasmoregood ideas in it, theclas-siconesofDoctorWho, thananyother television seriesin history. They invented the TARDIS! Somebody satin a room and said: “It’s bigger on the inside and lookslike a police telephonebox.”They invented theDoctorwho’snever caught,whosename is showntobeDoctorWhobut isn’tDoctorWho,whichisinitselfaweirdandcharming idea. They invented the regeneration, theyinvented the Daleks, they invented the Cybermen,they invented a different version of a show where theDoctorwas a benevolent alien living on Earthworkingthrough the UNIT and saving the planet. All these aredifferent series contained within Doctor Who. …

There aremore goodideas there than in, look,BreakingBad, theWestWing, and theseare twothingsamongthebest things television has ever done. Doctor Who hasmore ideas in a couple of episodes than I have everhad in an entire life. (Moffat, 2017)

DoctorWho is also an ideal context in that it repre-sents a single cultural product realized for a longperiod of time within the same company (BBC). Assuch, it provides a controlled context for creativityand it allows us to rule out product-specific orcompany-specific characteristics that could be affect-ingcreativity (e.g.,Cattani&Ferriani,2008;Mannucci& Yong, 2018; Soda et al., 2004). Moreover, focusingjust on Doctor Who enables us to identify preciseboundaries for defining collaboration networks andcontent domains (see Clement, Shipilov, & Galunic,2018, for a similar approach),1 while at the sametimecontrolling forcreators’collaborationsandexpo-sure to content outside these boundaries. Finally, thetime required for creating and shooting Doctor Whoepisodes was very important for our focus, as itallowed for a fine-grained exploration of the stabilityversus change in network composition, with timewindows covering only few months rather than oneormore years.

1More broadly, this approach is consistentwith the largemajorityof extantnetworkstudies incultural industries thatfocus on a single product (e.g., movies, television shows,Broadway shows; Cattani & Ferriani, 2008; Soda et al.,2004; Uzzi & Spiro, 2005), and thus do not consider thework artists might have done in other fields. For example,an actor playing a role in a TV show might have workedalso in a movie at the same point in time.


Data and Sample

The sample consisted of the entire population ofcore crewmembers who worked in at least one of the273 episodes produced between 1963 (the year theshowstarted) and2014.While recognizing that a tele-visionepisodeistheresultofthecreativeeffortofmul-tipleprofessionals,we followedadiffusedpractice innetwork and creativity research (e.g., Cattani & Fer-riani, 2008;Mannucci&Yong, 2018; Perretti &Negro,2007; Soda&Bizzi, 2012) and focusedon the individ-ual artists who were in charge of the most criticalaspects of creative work. The “core” artists for eachepisode included three creative roles: one producer(sometimes called a “showrunner”), one or moredirectors, and one or more writers. In our sample,“core” teams vary in size from two to five, with themajority containing three people (81%).

Weidentified individualsassociatedwitheachrolebylookingat thecreditsofeachepisodeasreportedontheBBCwebsite.Wethencrosscheckedthereliabilityofthisinformationwithothersources,suchasspecial-izedpublicationsonDoctorWho (e.g.,Fleiner&Octo-ber, 2017; Howe, Stammers, & Walker, 1992, 1993,1994) andDoctorWho-dedicatedWikis (e.g.,TARDISWiki).We thencleaned thedata, removingduplicatesand checking for other inconsistencies. As not everyartistwas involved in every episode, the final sampleincluded 866 observations for 200 individual artists.

Social Network Structure of Doctor Who andArtists’ Cohorts

Tounveil the socialnetwork structureof theDoctorWho production world, we analyzed the affiliationnetwork between artists and episodes.An “affiliationnetwork” is a network of vertices connected by com-mon group memberships such as projects, teams, ororganizations. Examples studied in the past haveincluded collaborations among television professio-nals(Sodaetal.,2004),Broadwayartists(Uzzi&Spiro,2005), and Hollywood film professionals (Cattani &Ferriani, 2008). In our network, a link between anytwo artists thus indicates that they had workedtogether on themaking of an episode.

Like many cultural industries—and, in particu-lar, television—the Doctor Who collaboration net-work is structured as a “latent organization”(Starkey, Barnatt, & Tempest, 2000), with an inter-play of artistswho come together for a givenproject,seemingly disband, and then come together foranother project at a later date. Artists come to workon these projects in different ways: sometimes,they self-propose for a project, and sometimes the

content buyer actively pursues them. In DoctorWho, for example, Neil Gaiman self-nominated forwriting the episode “The Doctor’s Wife,” but it wasBBC executives who selected Verity Lambert as thefirst producer of the show (Fleiner & October,2017; Howe, Stammers, &Walker, 1992).

Within latent organizations, the great majority ofcollaborations take place within the project bound-aries, akin towhat happenswithina regularorganiza-tion (Starkey et al., 2000). Consistent with previouswork (e.g., Clement et al., 2018), we thus defined theboundaries of our network as the production worldof our focal product, thus limiting our analysis toartists’ collaborations while working Doctor Who.With such an extended run, the social network ofartists working on Doctor Whowas naturally charac-terized by different cohorts based on the time theseartists worked on the show. Figure 1 is a sociogramoftheartists involvedinDoctorWhoduringourobser-vationperiod(1963–2014).Symbolsrepresentthe200artists distinguished for their primary role (by color)andprimarycohort(bysymbolshape).Largersymbolsdistinguish artists who worked on more episodes.Thin lines connect artists who worked together ononly one episode, while bold lines connect artistswhoworkedtogetherontwoormoreepisodes.Artistsare located in the space close to other artists withwhom they worked (spring embedding algorithm;Borgatti, 2002). We use Graeme Harper (red trianglein the center of the sociogram) as an example to illus-trate what network connectionsmean in our context.Graemedirected a total of 14 episodes, threeofwhichwere produced by John Nathan Turner in the secondcohort (yellow square in the center of the secondcohort cluster). The bold line connecting Graemeand John indicates that they worked on more thanone episode together. The thin lines connectingGraeme with three other artists indicate that theyworked together on one episode.

ThesociogramofcollaborationsinFigure1displaysfour clusters. These clusters empirically identify fourartistcohortsthatcorrespondtodifferent timeperiodsof the shows.Artists aremoredensely interconnectedwithin cohorts, and each cohort is connected only byoccasional bridge relations between artists belongingto multiple cohorts. “Cohort 1” artists are clusteredtogether to the west (circles). Below them are the“Cohort 2” artists (squares). To the right of them isthe cluster of “Cohort 3” artists (triangles), and, tothefurtherright,istheclusterof“Cohort4”artists(dia-monds). The artists’ population that created theDoc-tor Who episodes is thus more precisely a set of fourseparate populations, variably overlapping, and


orderedintime.2Figure1showsthatthefewinstancesof artists working across cohorts generate numerousinterpersonal collaborations across cohorts, but thecohorts remain visible as separate populations. Thetable in Figure 1 shows that most interpersonal

collaborations arewithin cohort, with almost no con-nection between artists in the first two cohorts versusthe last two. The latter is due to the fact that the showwas effectively cancelled in 1989 because of fallingviewing numbers and a less prominent transmissiontime (Ley, 2013). This resulted in a 14-year gapbetween Cohort 2’s last episode in 1989 and Cohort3’s first episode in 2004—a gap depicted in Figure 1by the deep structural hole between the first twocohorts and the last two, spanned only by GraemeHarper, who connects Cohorts 2 and 3.

Data Set Construction: Cross-Sectional vs. Panel

Weconstructed twodata sets. The first one is cross-sectional, constructed by taking the approach com-mon in networks and in creativity research (e.g.,Burt,2007;Simonton,1984a)ofmeasuringanindivid-ual’s network over a given observation period andaggregating all outputs (in this case, their creative

FIGURE 1Doctor Who Collaboration Network

Phil Collinson(47 episodes, 24highly creative,27 colleagues)

Collaborationswithin andbetween cohorts

Graeme Harper(15 episodes, 8

highly creative, 14colleagues)

First cohort(1963–1979)

Second cohort(1980–1989)

Third cohort(2004–2008)

Fourth cohort(2009–2014)

Cohort 1

Cohort 2

Cohort 3

Cohort 4109

32

0

0

4

54

4

1

3

87

13

2

293

1

John Nathan Turner (50 episodes,5 highly creative, 47 colleagues)

Producer (18)

Director (94)

Writer (88)

2TheDoctorWhonetwork inFigure1meets thecriteriaofbeinga“smallworld,” inthat (a) theaveragenetworkdensityaround individual artists is much higher than would beexpected by random chance and (b) the path distance (i.e.,the shortest chain of indirect connections linking artists) isabout as short as would be expected by random chance(Watts & Strogetz, 1998). The average density of collabora-tive ties between artist contacts in Figure 1 is 65.6%—twothirds of the average artist’s contacts have collaboratedwith one another. The expected average density if thesame number of collaborations were distributed at randomwould be a much lower 3.1%. The average path distancebetween any two artists in Figure 1 is 3.8 steps, which isabout the same as the 3.1 steps expected if the same numberof collaborations were distributed at random.


contributiontoeachepisode)theyrealizedduringthistime. This data set thus consisted of the aggregationover time of all network and creativity data, with the200 artists as the unit of analysis. We constructedthisdataset for tworeasons.First,wewanted toverifythat the well-known positive relationships betweennonredundant network or content and creativity thatare usually identified when taking this aggregativeapproachwerepresent inour setting.Given thepecu-liarity of our setting, there was the chance that someidiosyncrasies related to the setting could affect ourhypothesis testing. Replicating these well-knownrelationships would make us confident that ourresults were not driven by these idiosyncrasies. Sec-ond, we wanted to offer an in-depth overview anddescription of the collaboration network that devel-opedoveryearsofcreativeproductionofDoctorWho.

The second data set is a panel that we used to con-duct ourmain analyses and test our hypotheses. Thisdata set is an unbalanced panel, with number of epi-sodes per artist ranging from 1 to 50, with an averageof 4, and included 866 artist–episode pairs as unitsof analysis.

Dependent Variable: Creativity

We measured creativity following the consensualassessment technique, a well-established method increativity research (Amabile, 1982, 1983). Thismethod is rooted in the idea that creativity is not anobjective property: in order to be considered creative,a product has to be judged as such by appropriateexpert observers belonging to the field (Amabile,1996; Csikszentmih�alyi, 1999). We thus recruitedtwo expert judges to assess the artists’ creativity.Judges were recruited for their expertise with Britishtelevision and with Doctor Who in particular. Bothwere critics withmany years of experience, and bothhadwrittenessaysandarticlesonthehistoryofDoctorWho. The judges provided their assessment consis-tentlywith the suggestedbest practices in the consen-sual assessment technique (Amabile, 1982). First, toestablish similar frames of reference, they were pro-vided with a definition of “creativity” as the genera-tion of novel and appropriate outcomes. Second,they provided their assessments independently.

Televisionepisodesarethesumofthecreativeeffortof different individual creators, each contributingwith their specialized knowledge and talents. Thisfeature allows experts such as our judges to identifyand isolate each individual’s creative contribution,independently from the overall creativity of the epi-sode (see Cattani & Ferriani, 2008, and Mannucci &

Yong, 2018, for a similar approach). For example, anepisode can feature outstanding directing but a poorscript. For each episode, judges were thus asked torate the creativity of the producer, of the director, ofthe writer, and overall episode creativity3 on a5-point scale (1 5 not creative, 5 5 very creative).The fact that the ratingswereprovided twoyears afterthe last episode was broadcasted allows us to mini-mize issues of reverse causality (see Mannucci,2017).However, the time separation could also creatememory problems: we therefore asked the judges tore-watch episodes they had not watched in morethan three years.

We provide the frequency counts of the creativityratings in the Online Appendix (Table A1). We mea-sured interrater agreement using Cohen’s (1960)weighted kappa, which is more appropriate in thepresence of ordinal variables (Bakeman & Gottman,1997).Thekappascores for the three rolesandtheepi-sodes varied between .79 and .83, appreciably higherthan the threshold of .61 generally accepted as agood level of overall agreement (Kvalseth, 1989).

For the panel data set, we used the creativity of thecreator’s role as a measure of their creativity in thegiven episode. If the creator covered more than onerole, we took the average of the two scores. For thecross-sectional values, we computed four differentmeasures of an artist’s creativity over their careerwithin Doctor Who, two measured at the individuallevel and twomeasured at the episode level: (1)maxi-mum individual creativity exhibited by the artist, (2)maximum episode creativity for episodes the artisthasworked on, (3) number of highly creative individ-ual contributions, and (4) number of highly creativeepisodes the artist has worked on. We considered acontribution as “highly creative”when the creativityscore as assessed by our two judges was equal to orhigher than 4.5.

Independent Variables

Forthepaneldataset,weconstructedthenetworkofeach artist at time t as the network composed of everyother person who worked with the artist over a four-episode timewindow—the episode at time t plus theimmediatelypreceding three episodes.Asanepisodeis produced in about one month, a four-episode win-dow could be seen, on average, as a four-month time

3Wetreatedepisode creativity as akin tobeinga coauthorof a significantwork, andassigned thesameepisodecreativ-ity rating to all artists whoworked in a given episode.


window.4Weransensitivityanalysisbyreducingandexpanding the four-episode window, but found nosubstantive differences in results. We thus reportonly analyses with the four-episode windows. Forthe cross-sectional data set, we constructed the net-work of each artist by looking at the network com-posed of every other person who worked on thesame episodes as the artist over their time workingonDoctorWho. The connection between each pair ofpeople in the network was the number of episodesonwhich they everworked together.The size of these200networksrangedbetweentwoand47,withameanof 5.93 and amedian of four.

Network openness. We computed network con-straint to measure the extent to which an artist’s net-work was closed (Burt, 1992). Constraint increasedfrom 0 to 1 according to the extent to which a personhad few contacts (size), those contacts were stronglyconnected directly to one another (density), orstrongly connected indirectly through their connec-tions to the same other person in the network (hierar-chy). Scores approached 1 when an artist workedwith collaborators who often worked with oneanother. Scores approached 0 when an artist workedwith different people who themselves worked withdifferent people. We computed constraint withinfour-episode time windows for the panel data set,and over all of an artist’s time with Doctor Who forthe cross-sectional data set. The patterns of these twomeasures are illustrated in theOnlineAppendix (Fig-ure A3). To ease interpretation, we operationalizedour independent variable as 1 2 constraint, so thathigh scores reflect openness and low scoresreflect closure.

Content heterogeneity. Wemeasured content het-erogeneity in terms of how similar an episode’s con-tent was compared to other episodes the artist hadworked on. To compute this variable, we first identi-fied content categories that we could use to describeeach Doctor Who episode. Following an approachalready validated in other studies set in the cultural

industries (e.g., Cattani & Fliescher, 2013; Taylor &Greve,2006) and in the television industry inparticu-lar (e.g.,Clementetal.,2018;Zaheer&Soda,2009),weconsulted domain-specific sources to establish rele-vant content categories. Specifically, we searchedthrough published reference works and essays (e.g.,Fleiner & October, 2017; Howe et al., 1992, 1993,1994), magazines (e.g., Doctor Who Magazine, RadioTimes) and online sources (e.g., Tardis Fandom)focusing on Doctor Who. By cross-comparing thesesources, wewere able to identify four content catego-ries that were consistently used to classify DoctorWho episodes: story type, setting, incarnation of theDoctor, and type of alien foe.

We then followed a two-step procedure to corrobo-rate the appropriateness of these content categories.First, we reviewed the plots of each episode to ascer-tain that the four content categories could be indeedapplied to each episode, and verified that this wasthe case (see Zaheer & Soda, 2009, for a similarapproach). Second, and most importantly, we askedour expert judges to separately validate our list of cat-egories. They confirmed both that these four catego-ries were capturing the “language, messages,narrative, and identity” of each episode (Zaheer &Soda, 2009: 16), and that they significantly affectedepisodes’ key features, such as narrative style, visualappearance, and characters. For example, episodeswith the seventh incarnation of the Doctor have adarker, secretive atmosphere, whereas episodeswiththe third incarnation are characterized by moredown-to-earth, investigative plots. Similarly, epi-sodes that include the aliens called Daleks often takeplace inwar-ridden planets and sets, with a gloomiercinematography; whereas episodes that include thealiens called Time Lords take place in luxurious,sci-fi interiors, with a cinematography characterizedby saturated colors (Howe et al., 1992; Howe &Walker, 1998). Table 1 provides a detailed descrip-tion of each content category and of the relativesubcategories.

The second author and a research assistant blind tothe research hypotheses then used these four catego-ries to independently code the content of each epi-sode. In the few instances in which disagreementarose (about2%of the cases), they resolved it throughdiscussion.

We then split eachcategory into a set of binary vari-ables, each describing one subcategory. Each episodewas thus characterized by a profile of 41 binary varia-bles: three of the variables distinguished story type,three distinguished story setting, 12 distinguishedincarnations of the Doctor, and 23 distinguished the

4 It is important to note that, while producers oftenworked on consecutive episodes (the record being JohnNathan Turner, who produced 50 consecutive episodes—seeFigure1), directors andwriters typicallyworkedonnon-consecutive episodes. Table F1 shows how unusual it wasfor a director or writer to work on consecutive episodes.Evenwhenadirectororwriterworkedononly twoepisodes,the episodes were separated by amedian of three—mean ofnine—intervening episodes. This supports the idea that theDoctorWhocollaborationnetwork isa“latentorganization”(Starkey et al., 2000).


kind of alien opposing the Doctor. The content onwhichanartisthadworkedwasthusdefinedbyMcon-tent profiles, where M is the number of episodes theartist had contributed to, either within the four-episodetimewindow(paneldataset)oracrossalltheirwork onDoctorWho (cross-sectional data set). To theextent that an artist’sM content profiles were identi-cal, the artist had a history of homogeneous content;the more the artist’s M content profiles differed, themore they had a history of heterogeneous content.

WeusedJaccardcoefficients tomeasuredissimilar-ity between pairs of the M profiles, which togetherdefine an (M, M) symmetric matrix of associationlike a correlation table.We averaged theM2 elementsin the table tomeasure an artist’s content heterogene-ity. In our setting, a low coefficient means the artistworkedonstoriesof thesametype,inthesamesetting,with the same Doctor protagonist, against the samekindofalien.Theresultingmeasureshowedconstructvalidity both in terms of what is assumed in networktheory and what should be expected from previousresearch: content heterogeneity increased with thelevel of network openness (r5 .92, compared to, e.g.,r5 .71 inAral & VanAlstyne, 2011: 118).5

The way the Jaccard index was computed meantthatone-episodeartistswouldnaturally receiveahet-erogeneity score of 0. Given that the minimummeanJaccard for multi-episode artists was .333, this scorewould set one-episode artists far apart from the restof the population, potentially creating outlier prob-lems in the analysis. We thus shifted the content het-erogeneity score for single-episode artists from 0 to a.33,whichput themat the lowest level of content het-erogeneity, but only just below the rest of thepopulation.

Network stability. In the panel data set, networkstability was measured as 12 (nnew ties / maxnew ties).New ties were computed as the number of new faceson the creative team the artist was working with ona given episode, where a collaborator was treatedas“new” if theartisthadnotworkedwith thembeforethe current episode. Given the small team sizes, thenumber of new faces in a given episode were typi-cally one or two, with many instances of no newfaces (13.39%) and an equal number of three orfour (12.89%). For the cross-sectional data set, we

TABLE 1Doctor Who Content Categories

Category Description Number of subcategories

1 Type of story Describes whether the episode is about an

historical event or has a sci-fi plot

3 (1 5 historic, 2 5 pseudo-historic, 3 5 sci-fi)

2 Type of setting Describes where the episode is mostly set,in terms of location

3 (1 5 Earth, 2 5 alien planet, 3 5 spaceship)

3 Type of alien Describes the type of alien the Doctor is

facing as foe in the focal episode

23

4 Doctor Describes which incarnation of the Doctoris the protagonist of the episode

12 (one per incarnation of the Doctor)

TABLE F1Consecutive Episodes per Artist

Artist’s no. ofepisodes (N)

Artists withconsecutiveepisodes

Artists with morethan one episode in

one season

Min. no. of episodesbetween

consecutiveepisodes

Median no. ofepisodes between

consecutiveepisodes

Max. no. of episodesbetween

consecutiveepisodes

One (65) 65 65 1 1 1

Two (45) 9 25 2 5 60

Three (21) 1 2 3 11 62More (51) 0 4 7 47 109

5 We also ran a principal component analysis of eachartist’s (M,M)matrix of Jaccardcoefficients as an alternativeway to summarize content homogeneity (ratio of first

eigenvalue to M). The principal component analysis andmean Jaccard measures were so highly correlated (r 5 .99)that we report only results with the more widely used Jac-card measure.


computednetworkstabilityastheaverageofthepanelmeasure across the episodes on which the artisthadworked.

Control Variables

Weincludedcontrolvariables toaccount for factorsthat can influence the creators’ likelihood of generat-ing a creative contribution or the characteristics oftheir network structure and content.

Panel data set. We controlled for artists’ level ofexpertise, a well-known creativity precursor (e.g.,Amabile,1983;Dane,2010;Simonton,2003).Thevar-iablewascomputedas thenumberofepisodes theart-isthadworkedonuptothefocalone.Wealsoincludedameasure for inputnonredundancy, to control for theexperiencesof thepeople in the focal artist’snetwork.We computed it as the number of content elementsthattheartist’saltershadexperienceinwhiletheartistdid not, divided by the total number of content ele-ments alters had experience in. We also controlledfor an artist’s outside experience, measured as thenumber of TV shows outside Doctor Who the artisthad worked on during the focal year. Including thiscontrol was warranted for two reasons. First, non-redundant content and thinking styles can in factcome not only from network position and exposure,butalsofromworkinginunrelatedareasandproducts.Second, while focusing on the collaboration networkof a singleTVshowallowedus tocontrol forpotentialconfounds at the product or company level, it did notallow us to assess the role played by outside experi-ence, a potentially powerful creativity precursor(e.g., Perry-Smith & Shalley, 2014; Reagans, Zucker-man, &McEvily, 2004).

Additionally, we controlled for the previous crea-tivity of the artist, measured as the average creativityof their prior contributions as rated by our judges.6

The inclusion of this control was warranted becauseprior creativity can affect current creative perfor-mance (e.g., Audia & Goncalo, 2007). Moreover,including this variable allowed us to control for

unobservedvariablesand forotherpotentially impor-tant, but omitted, predictors of creativity (Greene,2011).

We also controlled for previous outside collabora-tionsbetweenthecreatorandtheirties,andfortheout-side ties of each creator with other artists in thetelevision industry outside Doctor Who. As men-tioned above, focusing on the collaboration networkof a specific product is standard practice in general(e.g., McFadyen & Cannella, 2004), and in studies setin the cultural industries in particular (e.g., Clementetal.,2018).However,ourfocusonnewtiespromptedus to control for the number of preexisting ties due tocollaborations outside Doctor Who. We computedpreviousoutsidecollaborationsas thenumberofpeo-pleineachartist’snetworkwhomtheartisthasalreadyworked with on other projects outside Doctor Whoprior to the focal episode. Controlling for outside tiesis a standard practice in network research, and innetworks–creativity in particular (e.g., Fleming et al.,2007; Perry-Smith, 2006; Tortoriello & Krackhardt,2010), as it allows balancing the need to set up bound-aries for mapping the focal network with the need toaccount for actors’ outside experience (Laumann,Marsden, & Prensky, 1989).Wemeasured outside tiesas the number of people not included in the networkthat thefocalcreatorhadworkedwithonotherproduc-tions during each four-episode timewindow.

Finally, we included dummy variables for the rolecovered by the artist in the focal episode and for thecohort to which the focal episode belonged. Control-ling for roleswas important because they have impli-cations for the way artists work. Producers wereoften hired towork on a sequence of consecutive epi-sodes. In contrast, directors andwriters were usuallyhiredonaper-episodebasis.Asaconsequence,athirdof the writers and directors worked on only one epi-sode (35.7%), and another third worked on only twoor three episodes (36.3%). Controlling for cohortswas also relevant, for two reasons. First, we observedsignificant differences between cohorts. The firstcohort created and established the template for theshow.Artistsinthefirstcohortproducedthemostepi-sodes (108, vs. 59 in the second most active cohort),involving the largest number of different artists (86,vs.46 in thesecond largest cohort).Thesecondcohortwas instead characterized by the presence of a singleproducer, John Nathan Turner (the large yellowsquare in Figure 1), compared to eight in the firstone, with directors and writers experiencing shorteremployment periods than in the other cohorts. Halfof the writers and directors in the second cohortworkedonasingleepisode(48%),comparedtoathird

6 For the first contribution, where no prior rating wasavailable, we tried two different specifications of this vari-able. First, we assigned to the first contribution a value of0, in order to reflect the fact that no contribution had yettaken place. Second, we assigned to the first contribution avalue of 3—that is, the mid-point of the scale on whichjudges rated artists’ creativity. Results for our focal relation-ships remained identical across the two specifications. Theeffect of the prior creativity variable was also the same, interms of direction and significance, across specifications.We report results based on the first specification.


in the other cohorts (35%, 30%, and 28%, respec-tively, in the first, third,and fourthcohorts).The thirdcohortenteredafterthe14-yearhiatusintheshowpro-duction:artists in thiscohort thusenjoyedsomeof thefreedom and license enjoyed by the first cohort. Pro-duction in the thirdcohortwasalso relativelycentral-ized in a single producer, Phil Collinson (yellowtriangle in Figure 1), who produced 84% of the thirdcohort’s episodes. The fourth cohort followed imme-diately inthewakeof the third,embeddedintheopin-ion and practice of the third cohort without theleadership of Phil Collinson’s experience. Second,controlling for the cohort was particularly importantin the panel data set because each cohort representsanetwork community.A stable community is charac-terized by high connectivity and high knowledgeflow, and is at higher risk of homogenization (Gulati,Puranam, & Tushman, 2012; Sytch & Tatarynowicz,2014). Thus, transitions between cohorts can be dis-ruptive experiences that make artists in the subse-quent cohort more likely to rethink previousopinions and behaviors. Conversely, the shorter andless disruptive the transition from one cohort toanother, themore the subsequent cohort is embeddedin the first, making only incremental adjustments toestablished opinion and behaviors.

Cross-sectional data set. For the cross-sectionaldata, we controlled for an artist’s level of expertise,outside experience, creative role, and cohort. Exper-tisewas computed as thenumber of episodes anartisthadworked onduring their entire run inDoctorWho.Wemeasuredoutsideexperienceas thenumberofTVshowsanartisthadworkedonduring their career thatwere not related toDoctorWho.

Wealso includeddummyvariables for creativeroleand cohort, in order to control for unobserved role-specificandcohort-specificcharacteristics. Ifanartist

had worked in more than one role or cohort, weassigned them to the role they more frequently cov-ered and to the cohort they spentmore time in.

RESULTS

Preliminary Analysis (Cross-Sectional Data Set)

Table 2 presents the correlations and descriptivestatistics for our variables in the cross-sectional dataset. The correlations ofnetworkopenness andknowl-edge heterogeneitywith our fourmeasures of creativ-ity are positive and significant, ranging between .540and .572 for constraint (p , .001) and .519 and .593for heterogeneity (p , .001). This suggests that opennetworks and nonhomogeneous knowledge are posi-tively related to creativity in our setting, too, replicat-ing thewell-known relationships in extant research.

Figure 2presents a visual depictionofour findings:themore closed the network around an artist, the lesscreative their work.7 Figure 2A compares artists withrespect to their most creative work, while Figure 2Bcompares artists in terms of the number of creativecontributions. There is a linear association with thenumber of very creative works up to about 70 pointsof network openness, above which there is a concen-tration of creative work in the artists with the mostopen networks.

To further explore this relationship, we also con-ducted regression analyses. While our cross-sectionaldesignandvariableaggregationdonotallowus to claimcausality, regression analyses offer amorerobust test of the relationship between our predictorsand creativity. We used ordinary least squares (OLS)

TABLE 2Mean, Standard Deviations, and Correlations for the Cross-Sectional Data Set

Variable Mean SD 1 2 3 4 5 6 7 8

1 Max role creativity 4.05 1.022 Max episode creativity 4.04 1.00 .91

3 N Highly creative contributions 1.21 2.56 .42 .40

4 N Highly creative episodes 1.36 2.65 .43 .45 .94

5 Network openness 0.32 0.29 .54 .55 .51 .576 Content heterogeneity 0.50 0.18 .52 .54 .52 .59 .92

7 Network stability 0.55 0.15 .31 .31 .32 .34 .27 .49

8 Expertise 4.33 6.59 .36 .36 .81 .88 .66 .68 .35

9 Outside experience 10.42 10.80 .06 .03 .08 .05 .07 .12 .12 .04

Note: All correlations higher than |.26| significant at p , .01.

7To facilitate Figure 2’s interpretation,wemultiplied thefractional constraint scores by 100. This allowed us to dis-cuss points of constraint.


regressions for the analyses focusing on maximumcreativity, and Poisson regressions to predict the fre-quencywithwhichanartist producedhighlycreativework.8Foreachmeasureofcreativity,weenteredvar-iables into the analysis at two hierarchical steps: (1)control variables, (2) predictor variables.

Table 3 presents the regressions analyses. Theresults are highly consistent across different opera-tionalizations of creativity. Looking atModels 2, 4, 6,and8,wecanseethatnetworkopennesshasapositiveand significant effect across all the operationaliza-tionsofcreativityexceptthenumberofhighlycreativeepisodes (p, .01 formaximumrolecreativity;p, .05formaximumepisodecreativity;p, .01fornumberofhighly creative individual contributions). On theotherside,contentheterogeneitydidnothaveasignif-icant effect on any of our operationalizations of crea-tivity. Finally, it is worth noting that the effect of

network stability is always positive (p, .05 for bothmaximum creativity measures, p, .01 for both crea-tivecontributionsmeasures): artistsworking instablenetworks generatedmore creativework.

Wealsoconductedanalysesenteringeachpredictorseparately, both with and without control variables.These analyseswerewarranted, given the high corre-lation between two of our predictors (network open-ness and content heterogeneity). Full results are notreported here due to space constraints, but are avail-able in the Online Appendix (Table A2). The effectof brokerage and stability when added in isolationwas almost identical to that found in ourmain analy-ses:bothvariablesdisplayedapositiveandsignificanteffect for all operationalizationsof thedependentvar-iable (p , .001 for all), both when control variableswerepresentandwhentheywereabsent.Contenthet-erogeneity, though,displayedadifferentpatterncom-pared to theone reported inourmainmodels (Models2,4,6,and8).Itseffectwassignificantforalloperation-alizations of the dependent variable (p, .001 for all),both when control variables were present and whenthey were absent, suggesting that the nonsignificanteffect found in the main analysis is due to brokerage“washing out” its effect.

FIGURE 2Association between Network Constraint and Creativity: Cross-Sectional Data Set

Highest Creativity Frequency of High CreativityA B

Mean maximum episode creativity

Mean of other two means

Mean maximum role creativity

Mean number of highly creative episodes

Mean of other two means

Mean number of high creativity in role

r = �.81

5.0

4.5

4.0

3.0 0

2r = �.84

4

6

8

10

10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100

3.5

Network Closure(100 × network constraint)

Hig

hest

Cre

ati

vit

y R

ati

ng

Fre

qu

en

cy

of

Hig

h-C

reati

vit

y W

ork

Network Closure(100 × network constraint)

8 Given that our dependent variable was overdispersed,we initially tried a negative binomial specification. How-ever, the dispersion parameter alpha was not significantlydifferent from 0, thus suggesting that the data were betterestimated using a Poisson rather than a negative binomialmodel.


Main Analysis (Panel Data Set)

Table4 reports the correlations anddescriptive sta-tistics for thepaneldataset.For thisdataset, theuseoflinearregressionwasnotappropriate, for tworeasons.First, the presence of repeated observations for thesame creators over time violates the OLS assumptionof independence of observations. Second, the vari-ance of the error termsmight beheterogeneous acrossdifferentcross-sectionalunits,presentingaheterosce-dasticity issue. The final model for the panel data setwasthereforeaconditionalfixed-effects linearmodel,which controlled for inherent differences in creator’sskills and ability. We performed a Hausman (1978)test to choose between fixed- and random-effects

models. The test was significant, indicating that therandom-effects estimator was not consistent. Wereport significance levels based on Huber–Whiterobuststandarderrorstocontrolforanyresidualheter-oscedasticity across panels. Using robust standarderrorswas equivalent to clustering on the creator, fur-ther accounting for the presence of repeated observa-tions (Arellano, 2003;Wooldridge, 2016).

We entered the variables into the analysis at fourhierarchical steps: (1) control variables, (2) predictorvariables, (3) each interaction separately, (4) bothinteractions together.Wedid not center the predictorvariables, and thus our interaction coefficients can beinterpreted as the effect of the independent variable

TABLE 4Mean, Standard Deviations, and Correlations for the Panel Data Seta

Variable Mean SD 1 2 3 4 5 6 7 8 9

1 Creativity 3.66 1.02

2 Network openness 0.22 0.27 2.06

3 Content heterogeneity 0.47 0.31 2.02 .364 Network stability 0.62 0.22 .06 2.01 .37

5 Previous creativity 2.95 1.71 .08 .36 .77 .52

6 Previous collaborations 0.68 0.97 2.06 .13 .13 2.06 .08

7 Outside ties 1.11 2.48 2.05 2.23 2.04 2.08 2.10 .078 Expertise 7.66 9.59 2.02 .53 .44 .23 .35 .07 2.18

9 Input nonredundancy 0.51 0.34 2.02 2.53 2.68 2.40 2.69 2.04 2.18 2.63

10 Outside experience 0.87 1.02 .00 2.31 2.12 2.08 2.14 .08 .42 2.17 .14

a All correlations greater than |.10| are significant at p , .01.

TABLE 3OLS and Poisson Models Predicting Creativity: Cross-Sectional Data Set

Variables

Model 1Max. rolecreativity

Model 2Max. rolecreativity

Model 3Max. episodecreativity

Model 4Max. episodecreativity

Model 5N creative

contributions

Model 6N creative

contributions

Model 7N creativeepisodes

Model 8N creativeepisodes

Network openness 2.064��

(0.730)

1.783�

(0.728)

2.518��

(0.886)

1.530

(0.861)Content heterogeneity 20.691

(1.315)

0.022

(1.285)

20.962

(1.388)

1.324

(1.330)

Average network stability 1.532�

(0.618)

1.335�

(0.553)

3.820��

(0.953)

3.078��

(0.832)Expertise (N episodes) 0.051��

(0.013)

20.004

(0.008)

0.050��

(0.016)

20.007

(0.007)

0.079��

(0.011)

0.038��

(0.007)

0.078��

(0.010)

0.036��

(0.008)

Outside experience 0.002

(0.007)

20.001

(0.006)

20.001

(0.007)

20.005

(0.005)

0.015

(0.008)

0.004

(0.007)

0.012

(0.008)

20.000

(0.007)Ego role dummies Yes Yes Yes Yes Yes Yes Yes Yes

Cohort dummies Yes Yes Yes Yes Yes Yes Yes Yes

R2 .187 .358 .179 .366 .393 .468 .408 .500

Notes: Unstandardized coefficients. Huber–White robust standard errors are in parentheses. Models 1–4 are OLS regressions.Models 5–8 are

Poisson regressions, with the pseudo R2 reported.� p , .05

�� p , .01


when the moderator is equal to 0 (Allison, 1977).Table5summarizes theresults.Wecheckedformulti-collinearity by computing the collinearity diagnosticprocedures illustrated by Belsley, Kuh, and Welsch(1980), themost appropriate approach for computingcollinearity using panel data (Hill & Adkins, 2001).These procedures examine the “conditioning” of thematrix of independent variables, producing a condi-tion number that is the largest condition index. Thecondition number for the full model was 13.11, farbelow the value of 30 consideredproblematic by con-ventional standards (Belsley, 1991), indicating thatcollinearitywas not an issue.

Model 1 includes all the control variables.We findthat previous creativity (p, .01) has a negative effecton creativity. Model 2 shows the results after weenteredthe independentvariablesandthemoderator.Neithernetworkopenness nornetwork stability has asignificant effect on creativity, but content heteroge-neity has a positive and significant effect (p , .05).9

Model3 reports theresultsafter including the interac-tionbetweennetworkopennessandnetworkstability.Asexpected,thecoefficientisnegativeandsignificant(p, .01), indicatingthattheeffectofopennetworksorstructural holes becomes less positive as network sta-bility increases. This is consistent with our Hypothe-sis 1. Model 4 reports the results after including theinteraction between content heterogeneity and net-work stability. The interaction coefficient is negativeand significant (p, .01), indicating that the effect ofcontent heterogeneity becomes less positive as net-work stability increases. This is consistent with ourHypothesis 2. Model 5 presents the results afterincluding both interaction variables: results are con-sistentwiththoseofModels3and4,withbothinterac-tioncoefficientsremainingnegativeandsignificant (p, .01forboth).Theoverall fitofthemodelimprovesascompared to the baseline, but also with respect toModel 2, indicating that the full model better fits thedata. The F test for one degree of freedom shows that

TABLE 5Panel Regression Predicting Creativity

Variables Model 1 Model 2 Model 3 Model 4 Model 5

Previous creativity 20.110��(0.030)

20.180��(0.043)

20.224��(0.042)

20.260��(0.056)

20.289��(0.052)

Previous collaborations 0.015

(0.045)

0.028

(0.050)

0.036

(0.050)

0.038

(0.049)

0.043

(0.049)

Outside ties 20.010(0.020)

20.014(0.017)

20.016(0.017)

20.020(0.016)

20.020(0.016)

Expertise (N episodes) 20.014†

(0.007)

20.017�

(0.008)

20.018�

(0.008)

20.018�

(0.008)

20.018�

(0.008)

Input nonredundancy 20.234(0.216)

20.066(0.231)

20.051(0.233)

0.036(0.230)

0.040(0.233)

Outside experience 20.024

(0.062)

20.014

(0.061)

20.013

(0.060)

20.022

(0.063)

20.020

(0.062)Network openness 20.225

(0.240)

0.989�

(0.464)

20.306

(0.245)

0.703

(0.467)

Content heterogeneity 0.558�

(0.242)

0.613��

(0.228)

2.195��

(0.562)

2.092��

(0.561)Network stability 0.267

(0.209)

0.816��

(0.255)

1.641��

(0.466)

1.969��

(0.441)

Openness 3 Stability 21.899��

(0.593)

21.567��

(0.591)Heterogeneity 3 Stability 22.465��

(0.678)

22.241��

(0.694)

Ego role Yes Yes Yes Yes YesCohort Yes Yes Yes Yes Yes

Notes: Unstandardized coefficients. Huber–White robust standard errors are in parentheses.† p , .10

� p , .05

�� p , .01

9Aswith thecross-sectionaldata set,we triedalsoaddingour core predictors separately, one by one. Results on their

main effects were identical to those reported above and inTable 5, Model 2.


Model5improvessignificantlyonModel2(Pr.F is,.001).

Figure 3 and Figure 4 plot the marginal averageeffects.Figure3showsthat theeffectofnetworkopen-nessispositivewhennetworkstabilityislow(p, .05),andbecomesincreasinglylesspositiveasnetworksta-bility increases, to thepoint of turning negativewhen

the network is characterizedbyzero change (p, .01).Figure4shows that theeffectofcontentheterogeneityis positive and significant when network stability islow (p, .01), but becomes increasingly less positiveas network stability increases. The analysis of mar-ginal effects provides further support for ourHypoth-eses 1 and 2.

Figure 5A andFigure 5Bprovide a visual depictionof the observed pattern of results bymapping the dis-tributionofcreativity ratings acrossdifferent configu-rations of network openness and network stability(Figure5A) andofcontentheterogeneity andnetworkstability(Figure5B).Figure5Ashowsthatcreativityis

FIGURE 4Marginal Effects of Content Heterogeneity on Indi-vidual Creativity at Different Levels of Network

Stability: Panel Data Set

0

�1

01

23

.25 .75 1.5

Network stability

Note: 95% confidence intervals.

FIGURE 5(A) Distribution of Creativity Ratings across Com-binations of Network Openness and Network Sta-bility: Panel Data Set. (B) Distribution of CreativityRatings across Combinations of Content Homoge-

neity and Network Stability: Panel Data Set

–0.10

EpisodeNetwork

NewTeammates

Many Two One or None

Open(78)

Closed(25)

Open(142)

Closed(203)

Creativity Highest in Open Networksversus New Teammates

Creativity Highest with HeterogeneousContent and New Teammates

Open(200)

Closed(218)

0.10

–0.05

0.00

0.05

0.20

0.15

–0.10

Mean

Rela

tive C

reati

vit

y R

ati

ng

Mean

Rela

tive C

reati

vit

y R

ati

ng

Content

Homogeneity

NewTeammates

Many Two One or None

Low(83)

High(20)

Low(211)

High(134)

Low(138)

High(280)

0.10

–0.05

0.00

0.05

0.20

0.15

A

B

FIGURE 3Marginal Effects of Network Brokerage on Indi-vidual Creativity at Different Levels of Network

Stability: Panel Data Set

0

�2

�1

01

2

.25 .75 1.5

Network stability

Note: 95% confidence intervals.


highest forpeople embedded inopennetworkswork-ing with multiple new teammates (far left columns).Thebenefitsofmaintaininganopennetworkarelowerif half the team is unchanged (middle columns), andreverses to negative if the team is unchanged or con-tainsonlyonenewmember(farrightcolumns).Asim-ilar pattern can beobserved for content heterogeneityin Figure 5B: creativity is highest for people withdiverse content experience working with multiplenew teammates, and decreases as the number of newteammates decreases.

Robustness Checks

First,wecontrolled for the effect of adifferent oper-ationalization of our independent variable. We thuscomputed structural holes using the effective sizemeasure developed by Burt (1992). This measurereflected the number of nonredundant contacts in anindividual’s networks, and is thus posited to haveopposite effects as compared to constraint (whichinstead measures the degree to which an individualis “constrained” within a redundant network). Theresults were identical to the ones presented above,with the interaction coefficient between effectivesize and network stability being negative and signifi-cant (p , .01 when entered alone, p , .05 whenentered with the interaction between homogeneityand stability): the effect of structural holes becomeslesspositiveasnetworkstabilityincreases.Thecoeffi-cient of the interactionbetweencontenthomogeneityand network stability stayed positive and significant(p, .01).

Second, we controlled for the presence of survivalbias in our model. It might in fact be that, in the longrun, successful individuals would be more likely tostay in the sample, whereas unsuccessful creatorswould have relatively less opportunities to displaytheir creativity in the future.While ourpast creativitycontrol variable should already account for this(Greene, 2011), we decided to directly test for thepotential effect of selection based on how successfula creator’s episode had been. We adopted Lee’s(1983) modified version of the Heckman (1979)model.Sincewewerelookingatindividualsdroppingout of the sample because of low success, we used anaccelerated failure time model with an exponentialdistribution to estimate the likelihood that a creatorwould leave the production (and thus the sample) inyear t1 1 (see Henderson, Miller, & Hambrick, 2006,andMannucci & Yong, 2018, for a similar approach).We used audience ratings for the focal episode as theselection condition. Audience ratings represent the

most salient measure of success in the televisionindustry, and are the most likely determinant ofwhether an individual would continue to work onDoctor Who or leave the production team. Weobtained audience ratings from the BBC website andcorroborated them using archival sources (e.g.,Howe et al., 1992). The selectionmodel included ourtwopredictors (constraintandknowledgehomogene-ity) and audience ratings. Results from the first stepregressionarereportedintheOnlineAppendix(TableA3).

We followed the procedure detailed byHendersonandcolleagues(2006)tocalculatetheselectionparam-eter, or inverseMills ratio, and then added it as a con-trol to the fullmodel. The inverseMills ratio does nothaveanysignificanteffectoncreativitywhenaddedtothe full model. Moreover, our main results are robustand consistent with those presented in Model 5,withbothinteractionsstayingpositiveandsignificant(p, .05 forconstraintandp, .01 forcontenthomoge-neity),andtheanalysisofmarginaleffectsbeingvirtu-ally identical to that presented above. This providesevidence that survival bias was not affecting ourresults (Certo, Busenbark,Woo, & Semadeni, 2016).

Third, we empirically verified one assumption ofour theorizing on the moderating effects of stabil-ity—namely, thataddingnewtieswouldbebeneficialregardless of the human capital owned by these newties, andinparticular regardlessofwhether theybringnonredundant content. This notion was rooted inresearch suggesting that the value of new ties inreshaping mental models and social interactions isindependent of their expertise, knowledge, and com-petences (e.g., Rand et al., 2011; Shirado&Christakis,2017).Weputthisassumptiontothetestbymeasuringtwo typesofhumancapital ownedbynewties: exper-tise and content nonredundancy. We measured newties’ expertise aswe did for focal actors: we first com-puted the number of episodes each new tie hadworked on; then, we took the average of these valuesand used it as our measure of new ties’ expertise. Wecomputed new ties’ knowledge nonredundancy as avariation of our control variable input nonredund-ancy10—that is, as the ratio between the number ofcontent elements that the artist’s new ties had experi-enceinwhiletheartistdidnot,andthetotalnumberofcontent elements new alters had experience in. We

10 This variable was unsurprisingly highly correlatedwith our measure of input nonredundancy (r5 .96). Thus,in order to avoid collinearity issues,wedropped input non-redundancy and used only the new ties’ knowledge nonre-dundancymeasure in this specific analysis.


then computed two separate model for each of thesevariables, for a total of four models. Specifically, foreach human capital variable, we tested one model inwhichwe added a three-way interaction between thehuman capital variable, network stability, and net-work openness; and one in which we tested thethree-wayinteractionbetween thehumancapitalvar-iable, stability, and content heterogeneity.

The three-way interaction with new ties’ expertiseand stability was negative and not significant for net-workopenness (p5 .110), andpositiveandnot signif-icant for content heterogeneity (p5 .366). The three-way interaction with new ties’ knowledge nonre-dundancy and stability was negative and not signifi-cant for network openness (p 5 .188), and positiveand not significant for content heterogeneity (p 5

.938). The full results are reported in the OnlineAppendix (TableA4).Overall, these findingsprovidesupport for our assumption that adding new ties willfoster the network openness or content heterogeneityon creativity regardless of new ties’ human capital.

Finally, we also controlled for the possibility thatprior experiencehadacurvilineareffectoncreativity,as prior research on creative careers would suggest(see Simonton, 1988, 1997). We thus added thesquared term of prior experience to our main model.The coefficient of the squared term was positive andsignificant (p, .01), indeed suggesting the presenceof a curvilinear effect. However, adding the squaredtermdid not affect ourmain results, with the interac-tions staying positive and significant (p , .01 forboth network openness and content heterogeneity)and the analysis of marginal effects being virtuallyidentical to that presented above.

DISCUSSION

Creativity often manifests as a bolt of lightning—something that strikes once and forever changes thecourseofwhatfollows.Storiesaboundaboutcreativeswho shaped the fieldwith only onememorable pieceof work. For example, Harper Lee’s novel To Kill aMockingbird won the Pulitzer Prize in 1960, andremains the only work she published during her life-time (2015’s Go Set a Watchman being a first draft ofMockingbird). However, as the demand for creativeideas keeps growing in organizations, producing asingle creative contribution might not be enoughto ensure organizational success. Understandinghow employees can preserve their creativity overtime is thus becoming increasingly important fororganizations.

In the present paper, we adopted a social networklens to address this issue. We suggested that the oft-found positive association between open networksandheterogeneousknowledgewithcreativity isques-tionable when one takes a long-term view thataccounts for changes innetworkcomposition.Asnet-work openness and content heterogeneity are unsta-ble and characterized by diminishing returns, theway individuals maintain them over time becomesrelevant for their continued creativity. We theorizedthat constantly rejuvenating network compositionby adding new ties, rather than maintaining existingones, ensures thecontinuedenjoymentof thecreativeadvantages provided by network openness and con-tent heterogeneity.

We tested and found support for our predictions byanalyzing 866 creative contributions from 200 artistsinvolved in the realization of 233 episodes of the tele-vision showDoctorWho. Open networks and hetero-geneous content were found to foster creativity onlywhen they were coupled with low network stabil-ity—that is,with the addition of new ties. These find-ingscontribute to researchonnetworksandcreativityand on creativity over timemore broadly.

Theoretical Contributions

Our study challenges and enriches our currentunderstandingofhowsocialnetworks shape individ-ual creativity. First, we offer a theoretical frameworkandempirical testofhowchangeinnetworkcomposi-tion plays into the networks–creativity relationship.In sodoing,we answer the long-standing call to intro-duce adynamic focus to researchon creativity in gen-eral (Anderson et al., 2014; Shalley, Zhou, &Oldham,2004)andonsocial structuresandcreativity inpartic-ular (Perry-Smith & Mannucci, 2015; Phelps et al.,2012). Specifically, scholars have suggested that sta-bility in network composition can be a potentially“important contingency variable in explaining whena particular type of structure (i.e., closed vs. open)will improveactorknowledgecreation” (Phelpsetal.,2012: 37). Our theory also speaks to research that hasexploredthedynamicsofstructuralholes(Burt&Mer-luzzi, 2016; Sasovova et al., 2010) and of structuralholesandperformanceincreativecontexts(Sodaetal.,2004; Zaheer & Soda, 2009).

Altogether,our findingspinpoint the importanceofconsidering the interactiveeffectofnetworkstructureor content, on one side, and network stability on theother in order to understand how it is possible to staycreative over time. Stability begets rigidity in mentalstructures and modes of interaction, leading to the


risk of rejecting nonredundant perspectives and con-tent and to the rigidity of collaboration patterns, thustaking away the creative spark deriving from creativeabrasion. Network change instead brings a shockthat forces individuals to reconsider their cognitivestructures and collaboration modes, increasing theirflexibility and thus enhancing the chance that theyconsider and utilize new frames and knowledge. It isinteresting to note that this effect is not contingenton whether new ties bring nonredundant content:the mere presence of new faces disrupts existingwaysofworkinganddoingthings, forcingindividualsto reconsider how they interact, share, and integrateknowledge.

Similarly,networkchangepersedoesnotengenderbenefits for thecreativityof agivenoutcome:our find-ingsshowthatitneedstobecoupledwithcertaintypesof structures and content in order to be conducive tocreativeperformance. In sodoing,weextend andcor-roborate extant literatureon theeffects of theadditionof new ties on individual creativity. First, we showthat the benefits of addingnew ties for any given crea-tor are contingent on the creator being able to “tap”intononredundantperspectives, frames,andcontent;while it is often assumed that new ties bring newcon-tent, thisassumptionisrarelytested.Moreover,extantresearchhasalsosuggestedpotentialdownsidestotheaddition of new ties in the disruption of coordinationand routines.Our studyprovides apotential explana-tion for these inconsistencies.Our results suggest thatnew ties are beneficial for individual creativity onlywhen coupled with network structures and contentthat provide the raw materials that ensure that the“shock” they bring is a positive one—one that acti-vates generative recombination and reconfigurationprocesses rather than just being disruptive. To use ametaphor, if new ties are the spark that ignites a crea-tive reaction, open networks and heterogeneous con-tent constitute the chemical ingredients. Thisfinding is consistentwith extantwork that has shownthat theadditionofnewties isbeneficial forcollectiveproblem-solving only when these ties are added to aspecific network structure—specifically, the core ofthe network (Shirado & Christakis, 2017), where theyare likely to stimulate the vision advantage that thisposition confers.

Overall,ourstudypointsout theneedtoaddnetworkstabilityandchangetotheequationinordertodevelopagoodnetworktheoryofcreativity.Thelong-lastingintu-itionofthebenefitsofbrokerageisreinforcedbytheideathat structural and content-based network advantagescannot be decoupled from network composition interms of old versus new ties—and, thus, from an actor-

based view. Future research should further explorethese issues by adopting a more in-depth view on net-work and creative trajectories, looking at howdifferentnetwork configurations engender different trajectoriesin creative productivity. One interesting avenue couldbe to look beyond the focal actor’s network to examinehowtheevolutionand“breakingandmaking”ofalters’ties shape the actor’s creativity. This alter-centricapproachhasalreadybeenadopted innetworkandcre-ativity research (e.g., Grosser, Venkataramani, & Labi-anca, 2017; Venkataramani, Richter, & Clarke, 2014)and could be fruitfully extended to focus on networkdynamics and creativity over time.

A second contribution of our study is thatwe bringthe structure–content debate to creativity research.We show that being embedded in an open networkhas a positive effect on creativity above and beyondits benefits in terms of content heterogeneity. Thisfinding is consistentwith research that has suggestedthatopennetworksprovidenotonlycontentheteroge-neity,butalsoa“vision”advantage,changing thewayindividuals thinkandsee thingsandallowing themtospot opportunities otherwise unseen (Burt, 2004,2008; Burt & Soda, 2017). Maintaining an open net-work is “valuable as a forcing function for the cogni-tive and emotional skills required to communicatedivergent views. It is the cognitive and emotionalskills produced as a by-product of bridging structuralholes that are the proximate source of competitiveadvantage” (Burt, 2008: 963). Our results corroboratethis view in the realm of creativity: the interactionbetweennetworkopennessandstability is significantevenwhen controlling for content heterogeneity, andboth the effect of openness and the effect of heteroge-neity are contingent on network stability. Given thecentrality of the structure–content relationship inshaping creativity, future research could furtherexplore their relatedyet independenteffectsby focus-ing on other types of structures or content. For exam-ple, scholars could explore whether and how theeffect of network size and centrality is independentfrom the one of content: while the majority of extanttheorizing on the creative benefits of centrality centeron increased access to nonredundant content (Perry-Smith & Shalley, 2003), it is true that centrality isalso linked to increased sense of power (Bonacich,1987),which in turnhasbeenshownto lead togreatercreativity (Galinsky, Magee, Gruenfeld, Whitson, &Liljenquist, 2008). Similarly, we focus on a specifictype of heterogeneity, capturing the degree to whicha creator isworking on somethingdifferent from theirpast.Futureresearchcouldfocusonothertypesofhet-erogeneity, such as the heterogeneity of the inputs


received through the social network, for example byfocusingonemailexchanges(e.g.,Aral&VanAlstyne,2011).

A third contribution of our study is that it indicatesthat our focal relationships could play out differentlydepending on the breadth of the time horizon beingconsidered. Our results suggest that aggregating crea-tive outcomes and networks over long time horizonscan result in a neglect of how network micro-dynamics in the short termshape thecreativebenefitsof open network structures in the long run. Extantresearch usually focuses on time horizons of at leastthree to four years—a period during which indi-viduals can adopt significantly different networkmaintenancestrategies,whichinturncanhavesignif-icantlydifferentimplicationsfortheirabilitytoaccruecreativeadvantages fromtheirnetworkstructure.Theexistence of these differences and their implicationshasalreadybeenpointedoutbynetworksandcreativ-ity scholars alike (e.g., Burt et al., 2013; Simonton,1988), but has so far been overlooked empirically.While extant evidenceemerging fromfocusingonrel-ativelylongtimehorizons(andreplicatedinourcross-sectional data set) shows that open networks arerelated to creative achievement over time, our mainfindings using time windows of about four monthssuggest that the “network recipe” behind thismacro-pattern is maintaining nonredundant networksthrough short-term cycles of rejuvenation in networkcomposition (i.e., the addition of new ties).

Similarly,our findingssuggest that also theeffectofnetwork stability could vary depending on the timehorizonconsidered:wefoundstabilitytobepositivelyrelated to creativity over a long time horizon, while ithas no direct effect on single creative contributions.Understandingthereasonsunderlyingtheseobserveddifferences could provide an explanation for thedivergentfindingsontheeffectsofstabilityoncreativ-ity and innovation (e.g., Ferriani et al., 2009;Kumar&Zaheer, 2019; Sytch & Tatarynowicz, 2014). Specifi-cally, we believe that understanding what changeand stability mean for these different time horizonscouldprovideinsightsintothispuzzle.Changewithinshort time horizons can come from adding a reducednumber of new ties—something that would result ina rejuvenation of the network without causing toomuchdisruption.Onthecontrary,lownetworkstabil-ity over a longer timeperiodmeans that the artist con-tinuously changes their network composition, thuspotentially jeopardizing their ability to maintaineffective collaboration patterns. Future researchcould explore this issue, focusing on how muchchange is “toomuch” across different time horizons.

Practical Implications

Our study has relevant implications for practi-tioners. First, we pinpoint the importance of net-work renewal in order to accrue the creativeadvantages of nonredundant structures andcontentinthelongrun.Ourresultsshowthatbuildinganon-redundant network might not be enough to sustaincreativity in the long term if it is not coupled withthe systematic broadening of the network withnewpeople.Preservingabrokerage role in the struc-tureofcollaborationisnotenoughtosustaincreativ-itywithout the intellectual friction and the shock tomental structure and routines associatedwith new-comers. This finding has implications for both indi-vidual creators and managers who want to reducecreativity fluctuations. These fluctuations come,in fact, with high risks and liabilities. Creators whoarenotabletomaintainaconstantfluxofideasmightbe excluded from interesting projects and careeradvancement opportunities, or even lose their jobaltogether. The history of cultural industries is fullof once-successful creators whose inability to con-tinuously generate creative ideas made it impossi-ble for them to find valuable employment: forexample, director Michael Cimino was ostracizedfrom Hollywood after the failure of Heaven’s Gate,despite having previously directed the criticallyacclaimed The Deer Hunter.

Similarly,managerswhocannotguaranteearegu-largenerationofcreativeideastotheirorganizationsrun the riskofhurting theorganizations’chances forsurvival, given the increased importanceofemploy-ees’ creativity for organizational competitiveness.Fromanorganizationalstandpoint, thekey implica-tion of our study is therefore thatmanagers can helptheir employees’ creativity by actively encouragingthemtomaintaintheircollaborativestructuresopento newcomers, even when products are successful,ensuring the right balance betweennonredundancyand stability. We believe that this idea extendsbeyond cultural products, as the need for fresh per-spectives is a feature of creativity across manydomains.

Our setting provides a perfect illustration of thisidea. One of the most interesting features of DoctorWho is that this TV show survived, from both a busi-ness and a cultural perspective, across generations ofviewers, industry and technological disruptions,andcompanyreorganizations.Over theyears, thecre-atives involved in this adventure have been able totransform and reinvent the “product” by capturingand adapting it to spirit of the time.


Limitations and Directions for Future Research

Notwithstanding its contributions, this study hassome limitations. First, the characteristics of our set-ting might limit the generalizability of our findings.The Doctor Who production world is characterizedby a strong focus on creativity, and by the adoptionof collaborative project-based structures. The latterfeature means that our conceptualization and opera-tionalizationofstabilityarerootedinthefact thatadd-ing a tie often implies breaking another one, given thefixednessofroles:adirectorworkingwithanewwriteron an episode likely means that that director will notbe working with someone they used to work with.Moreover, our setting is a creative industry, and wefocus on the work of highly qualified professionalsworking together to produce creative outcomes. Ourfindingsmay applymore closely to settingswith sim-ilar characteristics, such as consulting, scientificresearch, and new product design, but not to settingswhere creativity is not so central, roles aremore flexi-ble, or individuals are less qualified. However, thereis reason to believe in the generalizability of our find-ings to a wider range of settings. Results in our cross-sectional replicated the well-known relationshipsbetween open networks, nonredundant content, andcreativity. Second, many of the problems faced byemployeesinculturalindustriesarebecomingincreas-ingly common in other industries given the increasedcentralityofcreativityandinnovationtocompanysuc-cess and survival (Ahuja & Lampert, 2001; Lampel,Lant, & Shamsie, 2000). Finally, the “shock” effect ofnew ties is not premised on their quality, but just onthe fact that their addition disrupts established rou-tines and cognitive structures (Shirado & Christakis,2017). That said, we cannot definitively rule out thatthe phenomenon of interest plays out differently inother settings.

A second limitation stems from the fact that wefocused on the artists’ creative efforts and collabora-tion patterns within a single product category (i.e.,Doctor Who series). This choice was made followingestablished practices (e.g., Clement et al., 2018;McFadyen & Cannella, 2004), as this feature of thecollaboration network allowed us to rule out con-founding effects (e.g., different product categories,different companies) that usually affect networksfocusing on multiple products. At the same time,though, it could be that taking into account multipleproducts and the relative collaboration networkswouldpaint a different picture. For example, it couldbethatbrokerswhosenetworkspansdifferentproductdomains have a lower need to renew their network in

order to accrue creative benefits from their brokerageposition. However, there are three reasons that makeus believe that this should not affect the robustnessand generalizability of our findings. First, we con-trolled for creators’cumulatedexperienceandcollab-orationpatterns in otherTVseries. Second, this effectof spanning different domains should be true, at leastto some extent, also for creators whose network con-tactsspandifferentcategorieswithinthesameproductdomain. Our results show that, while controlling forthisnonredundancy, themoderatingeffectofnetworkstability was still significant. Third, the definitionitself of what constitutes a product domain is largelysubjective (Csikszentmih�alyi, 1999), and this samelogic could thus be applied to extant studies focusingon an entire industry. For example, network studieshave focused on network of creators within culturalindustries such as the movie industry (e.g., Cattani &Ferriani, 2008), the television industry (e.g., Sodaet al., 2004), and Broadway musicals (e.g., Uzzi &Spiro, 2005). However, many creatives involved inthese industries are likely to work in another one; forexample,screenwriterswriteforbothcinemaandtele-vision, and theater directors work on both musicalsand plays. This means that, even when consideringentire industries, creators can have networks span-ning multiple fields. Given the importance of clearlyidentifying network boundaries and meaningfulproduct categories for our analysis, we believe thatthebenefits of this choiceoutweigh itsdisadvantages,andthatour focusonasingleproductcategoryshouldnot significantly limit the generalizability of our find-ings. Scholars could further explore this issue byfocusing on networks spanning different productdomains.

Finally, thearchivalnatureof thedatapreventedusfromempiricallymeasuring themechanisms throughwhich structural holes, network stability, and theirinteractionshapecreativitywithindifferenttimehori-zons. Future research could identify these mecha-nisms through the use of designs such as laboratorystudies or ad hoc surveys.

CONCLUSION

Notwithstanding these limitations, we believe thatourstudyoffersafreshperspectiveontherelationshipbetween networks and creativity, and on creativitymore broadly. By introducing a dynamic lens tonetwork–creativity research, we hope to pave theway formore studiesexploring thedynamicsofsocialnetworks and creativity across different time


horizons, aswell as to researchexploring thebestnet-work strategies to sustain creativity over one’s career.

REFERENCES

Ahuja, G., & Lampert, C. M. 2001. Entrepreneurship in thelarge corporation: A longitudinal study of how estab-lished firms create breakthrough inventions. StrategicManagement Journal, 22: 521–543.

Allison, P. D. 1977. Testing for interaction in multipleregression. American Journal of Sociology, 83: 144–153.

Amabile, T.M. 1982.Social psychologyof creativity:Acon-sensual assessment technique. Journal of Personalityand Social Psychology, 43: 997–1013.

Amabile, T.M. 1983. The social psychology of creativity: Acomponential conceptualization. Journal of Personal-ity and Social Psychology, 45: 357–376.

Amabile, T. M. 1996. Creativity in context. Boulder, CO:Westview Press.

Amabile,T.M.,&Pratt,M.G.2016.Thedynamiccomponen-tialmodelofcreativityandinnovationinorganizations:Making progress, makingmeaning. Research in Orga-nizational Behavior, 36: 157–183.

Anderson,N., Poto�cnik,K., & Zhou, J. 2014. Innovation andcreativity in organizations: A state-of-the-sciencereview, prospective commentary, and guiding frame-work. Journal ofManagement, 40: 1297–1333.

Aral, S., & Van Alstyne, M. 2011. The diversity-bandwidthtrade-off.AmericanJournalofSociology, 117:90–171.

Arellano,M. 2003.Panel data econometrics. Oxford,U.K.:Oxford University Press.

Audia, P. G., & Goncalo, J. A. 2007. Past success and creativ-ityover time:Astudyof inventors in theharddiskdriveindustry.Management Science, 53: 1–15.

Bakeman,R.,&Gottman, J.M.1997.Observing interaction:An introduction to sequential analysis. Cambridge,U.K.: Cambridge University Press.

Baum, J. A., McEvily, B., & Rowley, T. J. 2012. Better withage? Tie longevity and the performance implicationsof bridging and closure. Organization Science, 23:529–546.

Belsley, D. A. 1991.Conditioning diagnostics, collinearityand weak data in regression. New York, NY: JohnWiley & Sons.

Belsley, D. A., Kuh, E., &Welsch, R. 1980.Regressiondiag-nostics. New York, NY:Wiley & Sons.

Bonacich, P. 1987. Power and centrality: A family of meas-ures.American Journal of Sociology, 92: 1170–1182.

Borgatti, S. P. 2002.NetDraw software for network visual-ization. Lexington, KY: Analytic Technologies.

Brass,D. J. 1995.Creativity: It’s all in your social network. In

C. M. Ford & D. A. Gioia (Eds.), Creative action in

organizations: 94–99. Thousand Oaks, CA: SAGE.

Brothers, T. 2018.Help!: TheBeatles, DukeEllington, and

themagic of collaboration. NewYork,NY:W.W.Nor-

ton & Company.

Burt, R. S. 1992. Structural holes: The social structure of

competition. Cambridge, MA: Harvard University

Press.

Burt, R. S. 2002. Bridge decay. Social Networks, 24: 333–

363.

Burt, R. S. 2004. Structural holes and good ideas.American

Journal of Sociology, 110: 349–399.

Burt, R. S. 2007. Secondhand brokerage: Evidence on the

importance of local structure for managers, bankers,

and analysts. Academy of Management Journal, 50:

119–148.

Burt,R. S. 2008. Informationandstructural holes:Comment

onReagans andZuckerman. Industrial andCorporate

Change, 17: 953–969.

Burt, R. S., Kilduff, M., & Tasselli, S. 2013. Social network

analysis: Foundations and frontiers on advantage.

Annual Review of Psychology, 64: 527–547.

Burt, R. S., & Merluzzi, J. 2016. Network oscillation. Acad-

emy ofManagement Discoveries, 2: 368–391.

Burt, R. S., & Soda,G. 2017.Social origins of great strategies.

Strategy Science, 2: 226–233.

Campbell,D.T.1960.Blindvariationandselectiveretention

in creative thought as in other knowledge processes.

Psychological Review, 67: 380–400.

Cannella, A. A., & McFadyen, M. A. 2016. Changing the

exchange: The dynamics of knowledgeworker ego net-

works. Journal of Management, 42: 1005–1029.

Carnabuci, G., & Di�oszegi, B. 2015. Social networks, cogni-

tive style, and innovative performance: A contingency

perspective. Academy of Management Journal, 58:

881–905.

Cattani,G.,&Ferriani,S.2008.Acore/peripheryperspective

on individual creative performance: Social networks

and cinematic achievements in the Hollywood film

industry.Organization Science, 19: 824–844.

Cattani, G., & Fliescher, C. M. B. 2013. Product category

dynamics in cultural industries: Spaghetti Westerns’

renewal of the Hollywood Western movie genre. In L.

Lazzeretti (Ed.), Creative industries and innovation

in Europe: Concepts, measures and comparative

case studies: 212–231. London, U.K.: Routledge.

Certo, S. T., Busenbark, J. R., Woo, H. S., & Semadeni, M.

2016. Sample selection bias and Heckman models in

strategic management research. Strategic Manage-

ment Journal, 37: 2639–2657.


Clement, J., Shipilov, A., & Galunic, C. 2018. Brokerage as a

public good: The externalities of network hubs for dif-

ferent formal roles in creative organizations. Adminis-

trative Science Quarterly, 63: 251–286.

Cohen, J. 1960. A coefficient of agreement for nominal

scales.EducationalandPsychologicalMeasurement,

20: 37–46.

Couzin, I. D., Ioannou, C. C., Demirel, G., Gross, T., Torney,

C. J., Hartnett, A., Conradt, L., Levin, S. A., & Leonard,

N. E. 2011. Uninformed individuals promote demo-

cratic consensus in animal groups. Science, 334:

1578–1580.

Csikszentmih�alyi, M. 1999. Implications of a systems per-

spective for the study of creativity. In R. J. Sternberg

(Ed.), Handbook of creativity: 313–335. New York,

NY: Cambridge University Press.

Dane, E. 2010. Reconsidering the trade-off between exper-

tise and flexibility: A cognitive entrenchment perspec-

tive.Academy ofManagement Review, 35: 579–603.

Dunbar, K. 1993. Concept discovery in a scientific domain.

Cognitive Science, 17: 397–434.

Ferriani, S., Cattani, G., & Baden-Fuller, C. 2009. The rela-

tional antecedents of project-entrepreneurship: Net-

work centrality, team composition and project

performance. Research Policy, 38: 1545–1558.

Festinger, L., Schachter, S., & Back, K. 1950. Social pres-

sures in informal groups: A study of human factors

in housing. Oxford, U.K.: Harper.

Fleiner, C., & October, D. 2017. Doctor Who and history:

Critical essays on imagining the past. Jefferson, NC:

McFarland.

Fleming, L., Mingo, S., & Chen, D. 2007. Collaborative bro-

kerage, generative creativity, and creative success.

Administrative Science Quarterly, 52: 443–475.

Galinsky,A.D.,Magee, J.C.,Gruenfeld,D.H.,Whitson, J.A.,

&Liljenquist,K.A.2008.Power reduces thepressof the

situation: Implications for creativity, conformity, and

dissonance. Journal of Personality and Social Psy-

chology, 95: 1450–1466.

Goldberg,A., Srivastava, S.B.,Manian,V.G.,Monroe,W.,&

Potts, C. 2016. Fitting in or standing out? The tradeoffs

of structural and cultural embeddedness. American

Sociological Review, 81: 1190–1222.

Greene, W. H. 2011. Econometric analysis. Upper Saddle

River, NJ: Prentice Hall.

Grosser, T. J., Venkataramani, V., & Labianca, G. J. 2017. An

alter-centric perspective on employee innovation: The

importance of alters’ creative self-efficacy and network

structure. Journal of Applied Psychology, 102: 1360–

1374.

Gulati, R., Puranam, P., & Tushman, M. 2012. Meta-organi-

zationdesign: Rethinkingdesign in interorganizational

and community contexts. Strategic Management

Journal, 33: 571–586.

Hansen,M.T.1999.Thesearch-transferproblem:Theroleof

weak ties in sharing knowledge across organization

subunits. Administrative Science Quarterly, 44:

82–111.

Hargadon, A. B., & Sutton, R. I. 1997. Technology brokering

and innovation inaproductdevelopment firm.Admin-

istrative Science Quarterly, 42: 716–749.

Hausman, J. A. 1978. Specification tests in econometrics.

Econometrica, 46: 1251–1271.

Heckman, J. J. 1979. Sample selectionbias as a specification

error. Econometrica, 47: 153–161.

Henderson, A. D., Miller, D., & Hambrick, D. C. 2006. How

quicklydoCEOsbecomeobsolete? Industrydynamism,

CEO tenure, and company performance. Strategic

Management Journal, 27: 447–460.

Hill, R. C.,&Adkins, L. C. 2001.Collinearity. InB.H.Baltagi

(Ed.),A companion to theoretical econometrics: 257–

278. Chicago, IL: JohnWiley & Sons.

Howe, D. J., Stammers, M., & Walker, S. J. 1992. Doctor

Who—The handbook: The fourth doctor. London,

U.K.: Virgin.


Who—The handbook: The sixth doctor. London,

U.K.: Virgin.


Who—The handbook: The first doctor. London,

U.K.: Virgin.

Howe, D. J., & Walker, S. J. 1998. Doctor Who—The hand-

book: The seventh doctor. London, U.K.: Virgin.

Hunt, J., & Gauthier-Loiselle, M. 2010. How much does

immigration boost innovation? American Economic

Journal. Macroeconomics, 2: 31–56.

Katz, E., & Lazarsfeld, P. 1955. Personal influence. New

York, NY: Free Press.

Kleinbaum, A. M. 2012. Organizational misfits and the ori-

ginsofbrokerageinintrafirmnetworks.Administrative

Science Quarterly, 57: 407–452.

Kumar, P., & Zaheer, A. 2019. Ego-network stability and

innovation in alliances. Academy of Management

Journal, 62: 691–716.

Kvalseth,T.O. 1989.NoteonCohen’s kappa.Psychological

Reports, 65: 223–226.

Lampel, J., Lant,T.,&Shamsie, J. 2000.Balancingact:Learn-

ing from organizing practices in cultural industries.

Organization Science, 11: 263–269.

Laumann, E. O., Marsden, P. V., & Prensky, D. 1989. The

boundary specification problem in network analysis.

In L. C. Freeman, A. Kimball Romney, & D. R. White


(Eds.), Research methods in social network analysis:

61–87. Fairfax, VA: GeorgeMasonUniversity Press.

Lee, L.F. 1983.Generalizedeconometricmodelswith selec-

tivity. Econometrica, 51: 507–512.

Ley,S. 2013,May29.WasDoctorWho rubbish in the1980s?

BBC News. Retrieved from https://www.bbc.co.uk/

news/entertainment-arts-22628484

Lingo,E.L.,&O’Mahony,S.2010.Nexuswork:Brokerageon

creative projects. Administrative Science Quarterly,

55: 47–81.

Maddux, W. W., & Galinsky, A. D. 2009. Cultural borders

and mental barriers: The relationship between living

abroad and creativity. Journal of Personality and

Social Psychology, 96: 1047–1061.

Mannucci, P. V. 2017. Drawing SnowWhite and animating

Buzz Lightyear: Technological toolkit characteristics

and creativity in cross-disciplinary teams. Organiza-

tion Science, 28: 711–728.

Mannucci, P. V., & Yong, K. 2018. The differential impact of

knowledge depth and knowledge breadth on creativity

over individual careers. Academy of Management

Journal, 61: 1741–1763.

March, J. G. 1991. Exploration and exploitation in organiza-

tional learning.Organization Science, 2: 71–87.

McFadyen,M. A., & Cannella, A. A., Jr. 2004. Social capital

and knowledge creation: Diminishing returns of the

number and strength of exchange relationships.Acad-

emy ofManagement Journal, 47: 735–746.

Moffat, S. 2017, April 9. My times as showrunner [Radio

Times Hall of Fame induction speech]. BFI & Radio

TimesTelevisionFestival2017,BFISouthbank,London.

Moran, C. 2007, June 30. Doctor Who is simply masterful.

Times. Retrieved from https://entertainment.

timesonline.co.uk/tol/arts_and_entertainment/tv_and_

radio/article1989181.ece

Morrison,E.W.2002.Newcomer’s relationships:Theroleof

social network ties during socialization. Academy of

Management Journal, 45: 1149–1161.

O’Connor, G. 2008, July 13. Daleks speak to all nations.

Wales on Sunday. Retrieved from https://web.archive.

org/web/20080715043335/http://www.walesonline.co.

uk/news/wales-news/2008/07/13/daleks-speak-to-all-

nations-91466-21331585

Perretti,F.,&Negro,G.2006.Fillingemptyseats:Howstatus

and organizational hierarchies affect exploration ver-

sus exploitation in team design.Academy ofManage-

ment Journal, 49: 759–777.

Perretti, F., & Negro, G. 2007. Mixing genres and matching

people: A study in innovation and team composition

in Hollywood. Journal of Organizational Behavior,

28: 563–586.

Perry-Smith, J. E. 2006. Social yet creative: The role of socialrelationships in facilitating individual creativity.Academy of Management Journal, 49: 85–101.

Perry-Smith, J. E. 2014. Social network ties beyond nonre-dundancy: An experimental investigation of the effectof knowledge content and tie strength on creativity.Journal of Applied Psychology, 99: 831–846.

Perry-Smith, J. E., &Mannucci, P. V. 2015. Social networks,creativity, andentrepreneurship. InC. E.Shalley,M.A.Hitt,& J.Zhou (Eds.),TheOxfordhandbookofcreativ-ity, innovation, and entrepreneurship: 205–224. NewYork, NY: Oxford University Press.

Perry-Smith, J.E.,&Mannucci,P.V.2017.Fromcreativity toinnovation: The social network drivers of the fourphases of the idea journey. Academy ofManagementReview, 42: 53–79.

Perry-Smith, J.E.,&Shalley,C.E.2003.Thesocialsideofcre-ativity: A static and dynamic social network perspec-tive.Academy ofManagement Review, 28: 89–106.

Perry-Smith, J. E., & Shalley, C. E. 2014. A social composi-tion view of team creativity: The role of membernationality-heterogeneous ties outside of the team.Organization Science, 25: 1434–1452.

Petruzzella, O. 2017, March 9. Doctor Who e il suo grandepunto debole. Hall of Series. Retrieved from https://www.hallofseries.com/doctor-who/doctor-who-punto-debole

Phelps, C., Heidl, R., & Wadhwa, A. 2012. Knowledge, net-works,andknowledgenetworks:Areviewandresearchagenda. Journal of Management, 38: 1115–1166.

Rand,D.G.,Arbesman,S.,&Christakis,N.A.2011.Dynamicsocial networks promote cooperation in experimentswith humans. Proceedings of the National Academyof Sciences of the United States of America, 108:19193–19198.

Rao,H.,Sutton,R., &Webb,A. 2008,April1. Innovation les-sons from Pixar: An interview with Oscar-winningdirector Brad Bird. McKinsey Quarterly. Retrievedfrom https://www.mckinsey.com/business-functions/strategy-and-corporate-finance/our-insights/innovation-lessons-from-pixar-an-interview-with-oscar-winning-director-brad-bird

Reagans, R., & McEvily, B. 2003. Network structure andknowledge transfer: The effects of cohesion and range.Administrative Science Quarterly, 48: 240–267.

Reagans, R., Zuckerman, E., & McEvily, B. 2004. How tomake the team: Social networks vs. demography as cri-teria for designing effective teams.AdministrativeSci-ence Quarterly, 49: 101–133.

Rodan, S., & Galunic, C. 2004.More thannetwork structure:How knowledge heterogeneity influences managerialperformance and innovativeness. Strategic Manage-ment Journal, 25: 541–562.


Rollag, K. 2004. The impact of relative tenure on newcomer

socialization dynamics. Journal of Organizational

Behavior, 25: 853–872.

Sasovova, Z., Mehra, A., Borgatti, S. P., & Schippers, M. C.

2010. Network churn: The effects of self-monitoring

personality on brokerage dynamics. Administrative

Science Quarterly, 55: 639–670.

Scholten, L., vanKnippenberg, D., Nijstad, B.A., & DeDreu,

C.K.2007.Motivatedinformationprocessingandgroup

decision-making: Effects of process accountability on

information processing and decision quality. Journal

of Experimental Social Psychology, 43: 539–552.

Schulz-Hardt, S., Frey, D., L€uthgens, C., & Moscovici, S.

2000. Biased information search in group decision-

making. Journal of Personality and Social Psychol-

ogy, 78: 655–669.

Sgourev, S. V. 2013. How Paris gave rise to Cubism (and

Picasso): Ambiguity and fragmentation in radical inno-

vation.Organization Science, 24: 1601–1617.

Shalley, C. E., Zhou, J., & Oldham, G. R. 2004. The effects of

personal and contextual characteristics on creativity:

Where should we go from here? Journal of Manage-

ment, 30: 933–958.

Shirado, H., & Christakis, N. A. 2017. Locally noisy autono-

mous agents improve global human coordination in

network experiments.Nature, 545: 370–374.

Simonton, D. K. 1984a. Artistic creativity and interpersonal

relationships across andwithingenerations. Journal of

Personality and Social Psychology, 46: 1273–1286.

Simonton, D. K. 1984b. Creative productivity and age: A

mathematicalmodel basedona two-step cognitive pro-

cess.Developmental Review, 4: 77–111.

Simonton, D. K. 1988. Age and outstanding achievement:

What do we know after a century of research? Psycho-

logical Bulletin, 104: 251–267.

Simonton, D. K. 1997. Creative productivity: A predictive

and explanatory model of career trajectories and land-

marks. Psychological Review, 104: 66–89.

Simonton, D. K. 2003. Scientific creativity as constrained

stochasticbehavior:The integrationofproduct, person,

andprocess perspectives.Psychological Bulletin, 129:

475–494.

Skilton, P. F., & Dooley, K. J. 2010. The effects of repeat col-

laboration on creative abrasion.Academy ofManage-

ment Review, 35: 118–134.

Soda, G., & Bizzi, L. 2012. Think different?An investigation

ofnetworkantecedentsandperformanceconsequences

of creativity as deviation. Strategic Organization, 10:

99–127.

Soda,G.,Usai,A., &Zaheer,A. 2004.Networkmemory:The

influenceofpastandcurrentnetworksonperformance.

Academy ofManagement Journal, 47: 893–906.

Starkey, K., Barnatt, C., & Tempest, S. 2000. Beyond net-works and hierarchies: Latent organizations in the UKtelevision industry. Organization Science, 11: 299–305.

Stovel, K., Golub, B., & Milgrom, E. M. M. 2011. Stabilizingbrokerage. Proceedings of the National Academy ofSciences of the United States of America, 108:21326–21332.

Sytch, M., & Tatarynowicz, A. 2014. Exploring the locus ofinvention: The dynamics of network communitiesand firms’ invention productivity. Academy of Man-agement Journal, 57: 249–279.

Taylor, A., & Greve, H. R. 2006. Superman or the fantasticfour? Knowledge combination and experience in inno-vative teams. Academy of Management Journal, 49:723–740.

Tortoriello, M., & Krackhardt, D. 2010. Activating cross-boundary knowledge: The role of Simmelian ties inthe generation of innovations. Academy of Manage-ment Journal, 53: 167–181.

Traulsen, A., Semmann, D., Sommerfeld, R. D., Krambeck,H. J., & Milinski, M. 2010. Human strategy updating inevolutionary games. Proceedings of the NationalAcademyofSciencesof theUnitedStatesofAmerica,107: 2962–2966.

Uzzi, B., & Spiro, J. 2005. Collaboration and creativity: Thesmall world problem.American Journal of Sociology,111: 447–504.

Venkataramani, V., Richter, A. W., & Clarke, R. 2014. Crea-tivebenefits fromwell-connectedleaders:Leadersocialnetwork ties as facilitators of employee radical creativ-ity. Journal of Applied Psychology, 99: 966–975.

Watts, D. J., & Strogetz, S. H. 1998. Collective dynamics of“small-world” networks.Nature, 393: 440–442.

Wolf, G. 1996, February 1. Steve Jobs: The next insanelygreat thing.Wired. Retrieved from https://www.wired.com/1996/02/jobs-2

Wooldridge, J.M.2016.Introductoryeconometrics:Amod-ern approach. Mason, OH: Cengage Learning.

Zaheer,A., & Soda, G. 2009.Network evolution:Theoriginsofstructuralholes.AdministrativeScienceQuarterly,54: 1–31.

Zhou, J., &Hoever, I. J. 2014.Researchonworkplace creativ-ity:A reviewand redirection.AnnualReviewofOrga-nizational PsychologyandOrganizational Behavior,1: 333–359.

Giuseppe “Beppe” Soda ([email protected])is professor of organization theory and network analysisat Bocconi University and SDA Bocconi School ofManagement. His research focuses on network origins


and dynamics, and on performance consequences of theinterplay between organizational architectures andorganizational networks.

Pier Vittorio Mannucci ([email protected]) is anassistant professor of organizational behavior at LondonBusiness School. He received his PhD in managementfrom HEC Paris. His research focuses on creativity and itsdynamics, with a particular attention to the role of socialnetworks, technology, and culture.

Ronald S. Burt (www.ronaldsburt.com) is the Charles M.Harper Leadership Professor of Sociology and Strategy atthe University of Chicago Booth School of Business, andsenior professor in the Department of Management andTechnology at Bocconi University. He received his PhDin sociology from the University of Chicago. His workdescribes cooperation and achievement as a function ofthe surrounding social network.


Copyright of Academy of Management Journal is the property of Academy of Management

and its content may not be copied or emailed to multiple sites or posted to a listserv without

the copyright holder's express written permission. However, users may print, download, or

email articles for individual use.

networks, creativity, and time: staying creative through

Documents