Strategic Management Journal Strat. Mgmt. J., 34: 666–686 (2013) Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2044 Received 28 March 2011 ; Final revision received 5 March 2012 CUTTING THE GORDIAN KNOT: THE EFFECT OF KNOWLEDGE COMPLEXITY ON EMPLOYEE MOBILITY AND ENTREPRENEURSHIP MARTIN GANCO * Department of Strategic Management and Entrepreneurship, University of Minnesota, Minneapolis, Minnesota, U.S.A. Employee entrepreneurship and employee moves to rival firms (employee mobility) have both been recognized as critical drivers of the transfer of knowledge. Drawing on a unique database of intra-industry inventor entrepreneurship and mobility events in the U.S. semiconductor industry, I examine the effect of the complexity of inventors’ prior patenting activities on their decisions to join a rival firm or found a start-up. The findings show that even though complexity inhibits knowledge diffusion to rival firms through employee mobility, complex knowledge may be underexploited within existing organizations and may still flow to startups through employee entrepreneurship. This study sheds new light on how technology shapes patterns of employee entrepreneurship and mobility, with implications for knowledge flows and competitive dynamics. Copyright 2013 John Wiley & Sons, Ltd. INTRODUCTION Employee entrepreneurship — the intra-industry founding of a new venture by an individual who previously worked for an incumbent firm—has been heralded as a hallmark of innovation (Free- man, 1986; Klepper and Sleeper, 2005), a critical source of new firm capabilities and heterogeneity in performance (Agarwal et al ., 2004; Phillips, 2002), and an impetus to the creation and growth of industries and regional clusters of firms (Klep- per, 2007; Sorenson and Audia, 2000). Through employee entrepreneurship, a new venture inherits industry-specific knowledge and strategies that are based on the founders’ prior work experience (Agarwal et al ., 2004; Chatterji, 2009; Klepper and Thompson, 2010). Similarly, scholars have Keywords: complexity; employee entrepreneurship; emp- loyee mobility; knowledge diffusion; NK model *Correspondence to: Martin Ganco, Department of Strategic Management and Entrepreneurship, Carlson School of Manage- ment, University of Minnesota, 321 19th Avenue South, #3-365, Minneapolis, MN 55455, USA. E-mail: [email protected]Copyright 2013 John Wiley & Sons, Ltd. long recognized intra-industry employee mobility (i.e., individual moves to another firm in the same industry) as a powerful engine of knowledge diffusion between firms, established and start-ups alike (Almeida and Kogut, 1999; Rosenkopf and Almeida, 2003; Singh and Agrawal, 2011). At the heart of these issues is a question that relates to the underlying drivers of employee entrepreneurship and mobility. Scholars have noted that profitable opportunities frequently arise as a result of information asymmetries while emphasizing the role of individuals’ prior knowl- edge (Helfat and Lieberman, 2002; Shane, 2000). For instance, Federico Faggin, Intel employee and inventor of the original Intel 4004 microprocessor, founded Zilog in 1975 after discovering that signif- icant improvements to the Intel 8080 architecture were possible. His decision led to the famous Z80 microprocessor, improving the Intel 8080 in terms of both speed and costs (National Inventors Hall of Fame Foundation, Inc; Pitta, 1997). T. J. Rodgers founded Cypress Semiconductor in 1982 to exploit his experience with MOS designs acquired while
Published online EarlyView in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2044
Received 28 March 2011 ; Final revision received 5 March 2012
CUTTING THE GORDIAN KNOT: THE EFFECTOF KNOWLEDGE COMPLEXITY ON EMPLOYEEMOBILITY AND ENTREPRENEURSHIP
MARTIN GANCO*Department of Strategic Management and Entrepreneurship, Universityof Minnesota, Minneapolis, Minnesota, U.S.A.
Employee entrepreneurship and employee moves to rival firms (employee mobility) have bothbeen recognized as critical drivers of the transfer of knowledge. Drawing on a unique database ofintra-industry inventor entrepreneurship and mobility events in the U.S. semiconductor industry,I examine the effect of the complexity of inventors’ prior patenting activities on their decisions tojoin a rival firm or found a start-up. The findings show that even though complexity inhibitsknowledge diffusion to rival firms through employee mobility, complex knowledge may beunderexploited within existing organizations and may still flow to startups through employeeentrepreneurship. This study sheds new light on how technology shapes patterns of employeeentrepreneurship and mobility, with implications for knowledge flows and competitive dynamics.Copyright 2013 John Wiley & Sons, Ltd.
INTRODUCTION
Employee entrepreneurship—the intra-industryfounding of a new venture by an individual whopreviously worked for an incumbent firm—hasbeen heralded as a hallmark of innovation (Free-man, 1986; Klepper and Sleeper, 2005), a criticalsource of new firm capabilities and heterogeneityin performance (Agarwal et al ., 2004; Phillips,2002), and an impetus to the creation and growthof industries and regional clusters of firms (Klep-per, 2007; Sorenson and Audia, 2000). Throughemployee entrepreneurship, a new venture inheritsindustry-specific knowledge and strategies thatare based on the founders’ prior work experience(Agarwal et al ., 2004; Chatterji, 2009; Klepperand Thompson, 2010). Similarly, scholars have
Keywords: complexity; employee entrepreneurship; emp-loyee mobility; knowledge diffusion; NK model*Correspondence to: Martin Ganco, Department of StrategicManagement and Entrepreneurship, Carlson School of Manage-ment, University of Minnesota, 321 19th Avenue South, #3-365,Minneapolis, MN 55455, USA. E-mail: [email protected]
Copyright 2013 John Wiley & Sons, Ltd.
long recognized intra-industry employee mobility(i.e., individual moves to another firm in the sameindustry) as a powerful engine of knowledgediffusion between firms, established and start-upsalike (Almeida and Kogut, 1999; Rosenkopf andAlmeida, 2003; Singh and Agrawal, 2011).
At the heart of these issues is a question thatrelates to the underlying drivers of employeeentrepreneurship and mobility. Scholars havenoted that profitable opportunities frequently ariseas a result of information asymmetries whileemphasizing the role of individuals’ prior knowl-edge (Helfat and Lieberman, 2002; Shane, 2000).For instance, Federico Faggin, Intel employee andinventor of the original Intel 4004 microprocessor,founded Zilog in 1975 after discovering that signif-icant improvements to the Intel 8080 architecturewere possible. His decision led to the famous Z80microprocessor, improving the Intel 8080 in termsof both speed and costs (National Inventors Hall ofFame Foundation, Inc; Pitta, 1997). T. J. Rodgersfounded Cypress Semiconductor in 1982 to exploithis experience with MOS designs acquired while
Employee Mobility and Entrepreneurship 667
working at AMD and American Microsystems(Cypress Semiconductor). Similarly, John Birknerand H. T. Chua founded QuickLogic in 1988 as away of exploiting their invention of programmablearray logic, which they invented while work-ing at Monolithic Memories (Computer HistoryMuseum; QuickLogic). These anecdotes illustratethat employees working at existing organizationshave preferential access to knowledge related totechnologies, markets, and environmental drivers(Agarwal et al ., 2004; Freeman, 1986; Sørensen,2007). Such work experience may also help theindividuals to acquire critical networks and bolsterconfidence in their success (Sorenson and Audia,2000). Employee entrepreneurship and mobilitythus represent key mechanisms for exploitation ofvaluable resources and opportunities outside theboundaries of parent firms. Indeed, extant studiessuggest that technologically more advanced firmsgenerate more entrepreneurs (Brittain and Free-man, 1980; Franco and Filson, 2006) and thatunderexploited technological opportunities oftenlead to employee entrepreneurship (Agarwal et al .,2004; Freeman, 1986; Garvin, 1983; Klepper andThompson, 2010).
While the drivers of employee entrepreneur-ship and mobility have been extensively stud-ied, they are largely examined in isolation ofeach other, some exceptions notwithstanding (e.g.Campbell et al ., 2012). To the extent that somecharacteristics of knowledge may affect employeeentrepreneurship and mobility differently, we cangain unique insights by examining the antecedentsof employee entrepreneurship and mobility jointly.In this study, I focus on an area that has receivedrelatively little attention, namely the micro-levelvariation in the nature of knowledge gainedwhile solving technological problems. Specifically,I examine the effect of technological complex-ity on employees’ choices about mobility andentrepreneurship. In other words, how does thecomplexity of the technological problems impactinventors’ decisions to move to another firm orstart up a new venture, either individually or inteams?
I develop a theory connecting knowledge com-plexity with employee entrepreneurship and mobil-ity, based on the conceptualization of knowledgeas a recipe, and inventors as carriers of the recipesthat they acquire while solving technological prob-lems (Dosi and Grazzi, 2010; Nelson and Winter,1982; Sorenson, Rivkin, and Fleming, 2006). I
define complex knowledge as a recipe that hasmany interdependencies between its ingredients.Two ingredients are considered interdependent ifa change in one ingredient affects performance ofanother ingredient.
The empirical context of the study is theU.S. semiconductor industry from 1973 to 2003,a canonical example of an industry driven bytechnological intensity, knowledge spillovers, andemployee mobility and entrepreneurship (Agarwalet al ., 2009; Freeman, 1986; Macher, Mowery, andHodges, 1998). To operationalize the complexityof knowledge held by individuals, I examine thepatent-level complexity of inventors’ prior activi-ties in the incumbent firm. The complexity of priorpatents thus serves as a proxy for the complexity ofrecipes that the inventor has acquired. I measurethe complexity using NK methodology (Flemingand Sorenson, 2001; Kauffman, 1993; Sorensonet al ., 2006), which I also validate for my empiricalcontext with the help of industry experts. To isolatethe effect of knowledge complexity on employeemobility and entrepreneurship from existing expla-nations, I employ a stringent empirical approach.I focus only on large public firms as sources ofemployee moves and utilize firm-year fixed effects.The resulting estimation is based on comparingindividuals within the same focal firm in the sameyear. This approach also simplifies the estimationsince all time-varying firm level controls are sub-sumed in the fixed effects.1
To briefly foreshadow the results, I theorize andfind that inventors whose patents reflect highercomplexity are less likely to join rival firms, butare more likely to become entrepreneurs. Further,I find that knowledge complexity increases thelikelihood of both team mobility (moving torival firms together with co-inventors) and teamemployee entrepreneurship (starting a new firmtogether with co-inventors). I also find that theimpact of complexity is significantly stronger in itseffect on team employee entrepreneurship relativeto its effect on team mobility.
Understanding how differences in knowledgecomplexity affect mobility and entrepreneurship
1 The controls only need to capture the individual leveldifferences within the ‘firm-year’. I also employ a multitude ofrobustness tests including an alternative measure of knowledgecomplexity and patent class-year fixed effects (which meanscomparing inventors who patent in the same technological classin the focal year).
outcomes has multiple practical and theoreticalimplications. From the practical perspective, theawareness of which technological areas spawnmost employee entrepreneurs may help incumbentfirm managers recognize underexploited opportu-nities. Further, such an understanding can directlyinform managerial practices with respect to indi-vidual employees. To harness the employees’ inno-vative talent, these practices may need to varywith the complexity of the technological prob-lems. In cases when the employee exit occurswithout the approval of the incumbent firm, thefirms need to mitigate potential misappropriationthreats. Incumbent firm managers could employdifferent strategies depending on the complexityof the employees’ tasks. Similarly, incumbent firmemployees may better understand why and whentheir ideas would face hurdles within the existingorganizations while realizing that disagreementsmay have structural underpinnings and are not per-sonal in nature. A better understanding could helpthe employees in their negotiations with employ-ers or facilitate their transition to other firms or toentrepreneurship.
Theoretically, the paper adds a contingency tothe traditional view that complexity is a barrierto knowledge diffusion (Fleming and Sorenson,2004; Sorenson, Rivkin, and Fleming, 2006;Williams, 2007). Even though complexity mayprevent spillovers to rivals through mobility, Ishow that complex knowledge can still flow tonew firms. Such a finding is important given thatemployee entrepreneurship has a detrimental effecton parent firm performance (Campbell et al ., 2012;Phillips, 2002; Wezel et al ., 2006). The resultsalso imply that the assignment of a technologicaltask can facilitate or inhibit the transition toentrepreneurship. The study thus provides a steptowards a ‘contextual’ theory of entrepreneurship(Aldrich and Fiol, 1994; Elfenbein, Hamilton,and Zenger, 2010; Shane, 2000; Sørensen, 2007;Sorenson and Audia, 2000; Stuart and Sorenson,2003). Such a ‘contextual’ approach to entre-preneurship suggests that entrepreneurial outcomesmay be driven by knowledge or organizationalcontext as opposed to entrepreneurial traits(Busenitz and Barney, 1997; Sarasvathy, Simon,and Lave, 1998). This focus on the context as adriver of entrepreneurship extends and comple-ments the dominant theories of entrepreneurship(Kirzner, 1997; Sarasvathy, 2001; Shane andVenkataraman, 2000).
THEORETICAL FRAMEWORKAND HYPOTHESES
Knowledge has been identified as one of the moststrategically important resources of a firm (Grant,1996; Kogut and Zander, 1996; Nickerson andZenger, 2004). According to this view, knowledgeis generated and held by individuals and applied tothe production of goods and services through thecoordination facilitated by the firm. Similarly, Iassume that while solving technological problems,inventors acquire knowledge. I conceptualize theirknowledge as a set of recipes (Dosi and Grazzi,2010; Nelson and Winter, 1982; Sorenson et al.,2006). These recipes include the ‘list of potentialingredients [that] encompasses both physical com-ponents and processes’ and ‘details [about] howto combine these ingredients’ (Sorenson et al.,2006: 997) or a ‘set of actions that need to betaken to achieve the desired outcome’ (Dosi andGrazzi, 2010: 173). In my context, for instance, aninventor may have acquired knowledge of how todesign a semiconductor device with certain spec-ifications. It is possible that the codified part ofthe recipe is complemented by an important tacitknowledge (Agrawal, 2006; Lowe, 2002; Polanyi,1983) that cannot be codified but is embodied inthe practice of the individual (Nelson and Winter,1982).
Employees not only acquire and hold knowledgebut also carry it across organizational boundaries.Employee entrepreneurship and mobility have bothbeen recognized as critical drivers that mitigate thedifficulties associated with the transfer of knowl-edge. The literature on employee entrepreneur-ship emphasizes that founders with prior workexperience in the focal industry bring with themhighly relevant industry-specific knowledge (Agar-wal et al ., 2004; Freeman, 1986; Klepper andSleeper, 2005; Phillips, 2002). Similarly, the lit-erature on mobility shows that mobility is akey driver of knowledge diffusion across exist-ing organizations (Marx, Strumsky, and Flem-ing, 2009; Rosenkopf and Almeida, 2003). Impor-tantly, the recipient firms ‘learn by hiring’—theknowledge brought in by mobile inventors dif-fuses throughout hiring firms (Singh and Agrawal,2011). The robust implication of the employeeentrepreneurship and mobility literatures is thatemployees carry knowledge that is valuable acrossorganizations.
Building on the notion that knowledge is a keyresource, the prior literature has conceptualizedthe creation of knowledge as a search for newrecipes over a space of possible combinations ofingredients (Nelson and Winter, 1982) or as asearch over a problem landscape (Nickerson andZenger, 2004; Sorenson et al ., 2006). Consistentwith this approach, the knowledge brought intothe recipient organization may need to be adapted(Williams, 2007). The received recipe, combinedwith the existing recipes of the hiring organization,provides a starting point for subsequent searching.
Perceiving knowledge as a recipe, Sorensonet al . (2006) develop a logic connecting complex-ity with the knowledge diffusion across actors andorganizational boundaries. According to this view,the innovative search for new recipes takes placein a rugged landscape, with ruggedness being anal-ogous to complexity (Kauffman, 1993; Levinthal,1997). The rugged landscape is a ‘problem space’searched by agents, who are, in the context ofthe current research, individual inventors solvinga given technological problem. The agents areassumed to be bounded in their ability to searchthis space and, thus, have to search for thesolutions by iterative experimentation—that is,by a local search. As the ruggedness of the spaceincreases, the problem becomes more difficultto solve (Rivkin, 2000). The boundedness of theagents’ search behavior may lead to ‘lock-in,’ ora cessation of searching before the best recipeis found. Kauffman (1993) showed that theruggedness (complexity) of a problem spaceincreases with the density of interdependenciesbetween individual components—i.e., betweenthe ingredients of the recipe.
The transfer of complex knowledge is alsoprone to difficulties. Rivkin (2000) showed the-oretically that, because of interdependencies, evensmall errors in the transfer process can lead tolarge performance penalties. To be functional,complex recipes either have to be kept intactor a coordinated change in multiple ingredientsneeds to be performed to improve the existingrecipe (Rivkin and Siggelkow, 2002). Transfer-ring complex knowledge thus imposes greatercoordinative challenges (Grant, 1996; Nickersonand Zenger, 2004). Because complex recipes mayembody knowledge of many interacting ingredi-ents, a higher proportion of such knowledge may
be tacit, which further complicates the knowledgecoordination and transfer. Additionally, the num-ber of local optima and the variance of their per-formance increase with complexity (Fleming andSorenson, 2001; Levinthal, 1997), which couldmake the ex-ante evaluation of outcomes asso-ciated with subsequent searches more uncertain(Gavetti and Levinthal, 2000).
Consequently, the transfer of a complex recipeand, in particular, the transfer between its carrierand other parties, may be ineffective. Consistentwith such a view, studies have reported that thecomplexity of knowledge inhibits its transfer toother contexts (Sorenson et al ., 2006; Williams,2007). However, it is possible that complexity alsoaffects the coordination, transfer, and exploitationof knowledge within existing firms. This may leadto differences between the mechanisms of knowl-edge transfer as they operate through employeeentrepreneurship and mobility. Knowledge com-plexity may affect not only which knowledge isabsorbed by other actors but also the knowledgethat is exploited within original firms. To examinethese differences, I combine the logic connect-ing complexity and knowledge diffusion (Sorensonet al ., 2006) with the notions of employee mobilityand entrepreneurship, occurring individually and inteams.
Knowledge complexity and the useof knowledge within existing firms
The mechanisms operating within the source firmwill determine the set of recipes that individ-uals acquire and potentially transfer to otherorganizational settings. Importantly, these mech-anisms may also determine which recipes are usedby the existing organizations. In an early work,Freeman (1986) argued that individuals work-ing for incumbent firms have the opportunity tolearn about badly served markets. Their parentorganizations fail to exploit these opportunitiesbecause they lack speed or are unable to allo-cate resources efficiently. The more recent model-ing literature suggests that employees may prefernot to disclose certain inventions (Anton and Yao,1995) or they may discover new ideas throughexploratory searching, which will remain unre-warded by the incumbent firm (Hellman, 2007).Further, the presence of opportunities (i.e., unusedrecipes) has been attributed to intrafirm frictionsin knowledge transfer (Franco and Filson, 2006),
underexploitation of knowledge (Agarwal et al .,2004), or information asymmetries (Klepper andThompson, 2010). The overarching implication ofthis literature is that profitable, unexploited oppor-tunities exist in organizations. I proceed to explorehow knowledge complexity affects the likelihoodthat a new recipe discovered by an inventor is notutilized by the original firm.
When solving a complex problem, inventorssearch a landscape that is rugged—has manypeaks and valleys (Kauffman, 1993). In such acase, the implementation of a newly discoveredcomplex recipe requires a coordinated change inmultiple ingredients relative to the currently usedsolution. When the problem is simple, the peaksare clustered together—discovered recipes dif-fer from each other only by a few ingredients(Kauffman, 1993; Rivkin and Siggelkow, 2002).When the problem is complex and the ingredientshave dense interactions, then peaks are distributedacross the search space—searching inventors maydiscover recipes that are more distant from thecurrently used one. Even though the newly dis-covered recipe may represent improvement overthe recipe currently used by the incumbent firm,the firm may be unwilling to make the transition.The coordinated change of many ingredients mayinterfere with complementarities across projects(Cassiman and Ueda, 2006), and the firm may viewthe new recipe as inconsistent with its long-termstrategic direction (Gavetti and Levinthal, 2000).Prior theoretical work (Cassiman and Ueda, 2006)showed that, even when the new project itself isviable, it is not necessarily optimal for the firmas a whole. Due to a lack of complementaritieswith other existing projects, the firm may will-ingly let the new project be exploited by others.The challenge to make the coordinated changesmay be even more pronounced when the inven-tor has a specialized knowledge and discovers anew but only partial complex recipe (Dosi andGrazzi, 2010). The complementary parts of theexisting recipe residing within the firm may beinadequate for the newly discovered recipe due tothe interdependencies. The firm needs to performfurther searches to find the new complementarypieces of the recipe, imposing additional burdensand increasing the likelihood that the recipe willnot be used.
There are additional mechanisms that mayamplify the relationship between the knowledgecomplexity and the potential for employees to
acquire unused recipes. The actual use of the dis-covered recipe by the incumbent firm requires atransfer of knowledge from the inventor to otherindividuals within the firm. Attributes of complexknowledge, including its sensitivity to small errors,tacitness, and uncertainty may increase the likeli-hood that the parent firm will be unable to evaluateand adapt the knowledge effectively. The transferineffectiveness may be further amplified by the factthat the newly discovered complex recipe is likelyto be distant from the currently used one.
The arguments above suggest that, in therepertoire of acquired recipes, an inventor solvingcomplex problems is more likely to have a recipethat will not be used by her employer. Importantly,the acquired knowledge—consisting of both usedand unused recipes—can be valuable in othercontexts. The recipes can provide the seed forsubsequent searches within another firm.
Knowledge complexity and employee mobility
From the recipient organization’s perspective, thepurpose of hiring a new individual from anotherfirm is the potential gain from knowledge diffusion(Singh and Agrawal, 2011). Such knowledge couldconsist of the recipes that were both used andunused by the original firm. However, key tothe value creation potential of the employee tothe recipient firm is whether such recipes can begainfully exploited within its own organizationalboundaries.
For established firms, one barrier to exploitingthese recipes is based on the fact that the situ-ations in the original and the potential recipientfirm are analogous. The existing organization hasan ongoing operation that relies on coordinationof its existing knowledge. The incoming recipesmay need to be adapted for solving new prob-lems, or they need to be integrated with existingrecipes (Sorenson et al ., 2006; Williams, 2007).The adaptation and integration process leads to anew search. The coordinated changes necessary toimprove or implement successfully the complexrecipes brought in by the mobile inventor may beproblematic given the constraints imposed by theexisting activities of the hiring firm or the pur-suit of complementarities. The ability to performthe search also requires a transfer of knowledgebetween the newly hired individual and other par-ties within the hiring firm. Because of the sen-sitivity to small changes of the ingredients, the
tacit nature of the knowledge, and the uncertaintyassociated with subsequent outcomes, the com-plexity of the knowledge may render such a trans-fer difficult. As a result, the existing organizationsare likely to experience difficulties when adapt-ing complex knowledge originating outside theirorganizational context for their own use (Hoetkerand Agarwal, 2007; Williams, 2007). Further, theability of the organization to learn from the newlyhired individual may be negatively affected by theknowledge complexity. Singh and Agrawal (2011)show only limited diffusion of the hired inven-tor’s knowledge within the recipient organization.The attributes of complex knowledge likely furtherinhibit such ‘learning-by-hiring’ and decrease theability of an organization to diffuse efficiently theknowledge within its structures.
Consequently, implementing complex recipesinto existing organizations is difficult. Even thoughinventors solving complex problems may acquirerecipes that are not used by their original employ-ers, the recipient organizations may be unableto take advantage of them either. The ability ofexisting organizations to absorb any kind of com-plex knowledge—used or unused by the originalfirm—is constrained. The efficiency with whichthe potential recipient firms can exploit the knowl-edge carried by the inventors in turn affects themobility choices that the inventors face. As shownby the modeling literature (e.g., Anton and Yao,1995), the viability of outside alternatives affectsexit and mobility decisions. Consistent with thelogic of these models, the complexity of recipesthat inventors carry is likely a barrier limiting thenumber and scope of job alternatives that would-bemobile inventors have. In other words, the com-plexity of the entire knowledge held by the inven-tors may be a more important determinant of theirmobility options than the fact that they hold somepotential entrepreneurial ideas. Such a mechanismleads to the following prediction:
Hypothesis 1. An employee is less likely to moveto a rival firm as the complexity of the employee’sknowledge increases .
Transfer of complex knowledge and teammobility
Multiple studies have shown that technologicalproblem solving and innovation is increasinglya team phenomenon (Singh and Fleming, 2010;
Wuchty, Jones, and Uzzi, 2007). The trend is con-sistent with the view that production requires bothknowledge specialization (Grant, 1996; Kogut andZander, 1996) and the technological recipes tobe distributed among multiple parties (Dosi andGrazzi, 2010). However, keeping the distributednature of knowledge constant, there may be higheror lower levels of interdependencies between theknowledge ingredients held by different individu-als. For distributed but less complex recipes, i.e.,when the interdependencies are low, inventors maystill be able to move individually and to transferrecipes effectively.
When the recipe is not only distributed amongmultiple individuals but there are also interde-pendencies between its components, then transferof a partial recipe carried by a single individ-ual may be ineffective.2 The recipient organizationneeds not only to adapt the incoming knowledgefor its own use but also to provide complemen-tary parts of the recipe for its basic functionality.When only a partial complex recipe is transferred,the need for a tight coordination between the ingre-dients held by the hired individual, and exist-ing knowledge within the hiring organization maysubstantially complicate the use of this knowl-edge. The coordinative challenge (Grant, 1996;Nickerson and Zenger, 2004) associated with theuse of complex knowledge is more pronouncedwhen the incoming recipe is partial. The transferof knowledge between the newly hired individ-ual and the organization has to occur. The abilityof the recipient organization to match the partialrecipe carried by the inventor with complementaryknowledge will be hindered by the attributes ofcomplex knowledge—its tacit nature and outcomeuncertainty.
The solution to this problem could be movementof a larger proportion of the recipe as embodied inthe joint mobility of a collaborating team. Whenthe knowledge is complex, the recipient organiza-tions may look to hire teams of innovators ratherthan individuals. Team mobility provides paral-lel channels for knowledge transfer—minimizingthe impact of tacitness and transfer errors. Team
2 The argument implies that knowledge complexity and thedistributed nature of knowledge are separate drivers of teammobility. Also note that Hypothesis 1 implies that complexityinhibits mobility directly and not only through its effect on teamsize. Empirically, I control for the size of inventor teams. I thankan anonymous reviewer for these insights.
movement allows groups of collaborators toremain intact, maintaining communication routinesand social interaction developed while working atthe parent firm and retaining the coordination thatis critical for implementing and improving com-plex recipes. In other words, team mobility allowsthe retention of team-specific private knowledgethat emerges due to interdependencies and thatwould be lost if inventors moved individually orif the team dissolved (Fortune and Mitchell, 2012;Hoetker and Agarwal, 2007).
Consequently, team mobility mitigates the detri-mental effect of complexity on knowledge transfer.The inventor teams may be incentivized by recip-ient organizations to move together, and the like-lihood of observing team mobility should increasewith knowledge complexity:
Hypothesis 2. An employee is more likely tomove to a rival firm with coworkers relative toindividually as the complexity of the employee’sknowledge increases .
Knowledge complexity and employeeentrepreneurship
Scholars have explained employee entrepreneur-ship as driven by parent-firm frictions that leadto unexploited, profitable opportunities (Francoand Filson, 2006; Freeman, 1986; Hellman, 2007;Klepper and Thompson, 2010). As I proposeabove, knowledge complexity can serve as anadditional friction, increasing the likelihood thatan inventor acquires a recipe that is not usedby her employer. The unused recipe could pro-vide the seed for a subsequent search within anew firm.
Consider the case of Garmin—a manufacturerof consumer-oriented global positioning systems(GPS). While working on aerospace navigation,AlliedSignal employees Gary Burrell and Min Kaorealized that it is possible to design a GPS sys-tem targeted at the consumer market. AlliedSignalliked the idea but felt that it did not fit with thecompany’s identity as an aerospace products man-ufacturer, leading to the founding of Garmin, Inc.(Corporate History). Similarly, Federico Fagginidentified that significant improvements to the Intel8080 architecture were possible. However, he leftIntel and founded Zilog because “Intel, still in 74,was a memory company. Microprocessors alwayswere taking second . . . second best . . . and I felt
not appreciated, frankly, at Intel.”3 Both of theseexamples illustrate that, even though the inventorsdiscovered viable solutions to complex problems,their employers did not want to change their cur-rent activities and fully commit to the proposedideas.
But just as transferring complex knowledgeto another existing firm is problematic, exploit-ing complex knowledge through entrepreneur-ship raises potential difficulties. The question iswhether the factors that prevent the exploita-tion of complex recipes by existing organizationsare potentially mitigated in cases of employeeentrepreneurship.
When the complex recipe does not need to beintegrated into an existing structure, there are notrade-offs driven by the complementarities withexisting activities. Coordinated changes necessaryto improve or implement the recipe do not interferewith current strategies, and the complex recipe canserve as the foundation of a new organization.The knowledge coordination that is associatedwith the use of complex knowledge is, thus, lesschallenging in a new organization. Entrepreneursalso create and optimize their new organization toexploit the knowledge they bring in, assemblingcomplementary assets that match the opportunitythey pursue (Freeman, 1986; Wezel et al ., 2006).4
Further, the carrier of knowledge controls thefirm so there is no need to convince other managersabout proposed changes or the viability of ideas.That being said, the start-up founders still facea bottleneck in the form of external funding.There is again evidence, however, that existingfirms’ managers evaluate ideas differently than doventure capitalists (VCs) (Dushnitsky and Shapira,2010), and VCs fund many projects that incumbentfirms have rejected (Kenney and Florida, 2002). Inparticular, the projects that are rejected by existingfirms because they do not fit with their currentactivities may be very attractive targets for VCfunding. Managers of existing firms evaluate ideas
3 Interview with Federico Faggin, Stanford University, March3, 1995. http://silicongenesis.stanford.edu/transcripts/faggin.htm[June 18, 2009].4 This argument assumes that such complementary assetsare available, being either transferred from an incumbentfirm (e.g., human assets) or bought on the market. If thenecessary complementary assets are not available, employeeentrepreneurship is very difficult. For instance, Mitchell’s (1991)study of the diagnostic imaging industry showed a persistentadvantage for incumbents due to their dominance of distributionchannels.
by considering trade-offs with existing activitieswhile such considerations are less likely to bepresent in an evaluation by an independent VC.Further, VC funding decisions are based ona broad range of characteristics, going beyondthe evaluation of a specific idea and includingrecommendations, the backgrounds of founders,market conditions, and general attractiveness of thetechnological domain (Kenney and Florida, 2002).Consequently, VCs may fund the exploitation of acomplex recipe even though they may be unable toperfectly evaluate the prospects of the recipe itself.
Although the ability to transfer complex recipesto other organizations decreases with complexity,the challenges are mitigated when a new organi-zation is established to exploit the complex recipe.Further, in the repertoire of acquired recipes, indi-viduals solving complex problems are more likelyto hold recipes that are not used in their existingorganizations. Both of these factors favor exploita-tion of complex recipes by establishing a new firmrelative to moving to a rival firm:
Hypothesis 3. An employee is more likely to starta new venture relative to move to a rival firm asthe complexity of the employee’s prior knowledgeincreases .
Transfer of complex knowledge and teamentrepreneurship
In light of the fact that the relevant knowledge islikely distributed among collaborating individuals(Singh and Fleming, 2010; Wuchty et al ., 2007),the interdependence between the ingredients heldby different individuals creates similar challengesfor the transfer of knowledge to a new organizationas it does for the transfer to an existing one.The founder, who may be the carrier of a partialrecipe, needs to assemble a team of individualsholding complementary pieces of knowledge whileachieving the tight coordination necessary forimplementing the complex recipe. Analogous tothe situation of when the recipient is an existingfirm, the attributes of complex knowledge mayrender the founder’s search for complementaryingredients problematic. The solution, again, maybe to found the firm together with individuals whohelped to co-develop the complex recipe within theincumbent firm.
The critical difference between team mobil-ity and team entrepreneurship, however, is that
in cases of entrepreneurship the organization isassembled afresh. The founding team membersnot only serve as important complementary assetsfor determining the survival of the new venture(Eisenhardt and Schoonhoven, 1990), but they alsoallow the coordination developed while collabo-rating within the parent organization to continue.In cases of team mobility to an existing organiza-tion, the recipient firm needs to integrate the entireteam while facing a dual challenge—maintainingthe coordination within the team and adaptingtheir knowledge to match the existing activi-ties. Importantly, the adaptation of the incomingcomplex recipe may change the optimal config-uration of the existing team and render someindividuals unnecessary. The recipient organiza-tion may also already own the complementarypieces of knowledge and may be unwilling toduplicate them by hiring individuals with similarknowledge.
The characteristics of complex knowledge,including its tacitness, difficulties in evaluation,and uncertainty may further inhibit the ability ofthe recipient organization to absorb a larger team.The recipient organization may have difficultiesunderstanding when it is necessary to hire the com-plementary parts of the knowledge embodied indifferent individuals.
Extending the examples above, Federico Fagginstarted Zilog in 1974 with Ralph Ungerman, whoworked as a manager for him while at Intel.6 IfMr. Faggin had decided to pursue his ideas withinanother organization, the exit of Mr. Ungermanmay not have occurred. Integration of Mr. Faggin’sknowledge with the knowledge of a recipientorganization may have rendered the team moveunnecessary. The recipient organization couldhave provided similar resources, and the teammove based purely on the fact that they hadworked together before would have been harder tojustify.
Consequently, (1) the knowledge complexityincreases the likelihood of team entrepreneur-ship, and (2) the effect of complexity on teamentrepreneurship is likely more pronounced rela-tive to its effect on the mobility of teams acrossexisting organizations:
Hypothesis 4a. An employee is more likely tofound a firm with coworkers relative to individ-ually as the complexity of the employee’s knowl-edge increases .
Hypothesis 4b. The effect of the employee’sknowledge complexity on team founding relativeto individual founding is greater than its effect onteam mobility to rival firms relative to individualmobility .
METHODS
Industry context and data description
The context of the study is the U.S. semiconduc-tor industry. This industry exhibits a high degree ofemployee entrepreneurship and mobility; and priorstudies have documented that such mobility facil-itates interfirm transfers of knowledge (Almeidaand Kogut, 1999; Singh and Agrawal, 2011). Firmsin this industry have a high propensity to filepatents (Hall and Ziedonis, 2001), a characteristicthat allows construction of a patent-based mea-sure of complexity (Fleming and Sorenson, 2001,2004). The semiconductor industry is also idealfor these research purposes because of its focus oncomplex technological innovation (Macher et al .,1998). Following the shift of the U.S. semicon-ductor industry post-1990 to ‘fabless’ firms thatdesign semiconductors and outsource manufac-turing, entry became relatively easy, which fur-ther fostered innovation (Macher et al ., 1998).5
The critical complementary assets required in newdesign firms were highly mobile human assets(Campbell et al ., 2012; Teece, 2003). When crit-ical complementary assets are not locked intoincumbent firms, the entrepreneurial ideas canbe more easily transferred to other organizations.Such characteristics highlight the importance ofknowledge as a determinant of entrepreneurshipand mobility patterns and provide an ideal settingfor this study.
Empirically, I trace the innovative activitiesof 649 U.S. semiconductor firms over a three-decade period, 1973–2003. The construction ofthe sample is analogous to that in prior studieson mobility (Agarwal et al ., 2009; Rosenkopf andAlmeida, 2003) in that firms that were potentialsources of inventive talent were distinguished fromfirms in the industry that were potential recipients(rival incumbents and start-ups). The source firm
5 Only 3 percent of the sample are post-1990 entrants thatestablished a foundry. Excluding these firms from the sampledoes not alter any of the results.
sample consists of large publicly traded U.S.firms that (1) competed primarily in semiconductorproduct markets and (2) were founded prior to1995. Restricting attention to firms that werepublic by the mid 1990s (n = 136) allowed asufficiently long window through which to viewpossible mobility and employee entrepreneurshipevents. Focusing on large public firms as potentialsources of employee mobility and entrepreneurshipevents was necessary to allow for firm-yearfixed effects. Only firms that have inventorswith different observable outcomes (staying vs.mobility, mobility vs. entrepreneurship) in thesame year can be used in the estimation. Thisrestrictive empirical design allowed me to isolatethe effects at the individual inventor level andcontrol for existing explanations of employeeentrepreneurship and mobility that operate at thefirm or regional level (Agarwal et al ., 2004;Almeida and Kogut, 1999; Klepper, 2007). Toassemble a larger pool of potential recipients ofinventive talent in the industry and to maximize thelikelihood of observing employee entrepreneurshipand mobility events within the industry, therecipient firm sample includes all firms from thesource firm sample and the following: (1) usingthe Venture One database, I added semiconductorfirms that were founded between 1980 and 2003(n = 454), and (2) using Compustat, I addedfirms in the industry (SIC 3674) that went publicbetween 1995 and 2003 (n = 59).
Since I was interested in an inventor-levelanalysis, but USPTO patent data do not providea unique individual identifier, I reconstructedindividuals’ patenting histories via a matchingalgorithm described in Agarwal et al . (2009)that creates inventor patenting and employmenthistories. This algorithm identifies 28,123 uniquenames listed in patents awarded to firms, of which25,339 appear in the source firm sample. Employeemobility was observable only when an inventorpatented at both a source and a recipient firm.This restriction eliminated 14 incumbent and 188start-up firms.6 The final recipient sample thereforeincludes 266 private start-ups and 181 incumbentpublic firms. The matching algorithm yields 1,166mobility events.
6 The disproportionate omission of start-ups is not surprising.Many start-ups in the larger sample failed or were acquiredat very young ages, which reduced the likelihood of observingpatent awards for them.
Searching press releases in Lexis-Nexis, ana-lyzing archived websites of the recipient firms(www.archive.org), and utilizing several onlineresources (e.g. smithsonianchips.si.edu) enabledthe identification of the founders of the recipientfirms. Since I was interested in how individuals’past patenting shaped the emergence of new firms,I needed to identify inventor-founders—thosewhose ideas led to founding start-ups—and notsimply early ‘board members’. Consequently,I defined founder status stringently, requiringthe word ‘founder’ or ‘cofounder’ to appearwith the person’s name on either the archivedcorporate website (as early as possible afterthe year of entry) or in early press releases orindustry materials. To look at how prior inventiveactivity affected the decision to start a new firm,I matched the founder names (after verifyingand cleaning the matches using Lexis-Nexis andcorporate websites to reconstruct precisely founderemployment histories) with the source-firm poolof 25,339 inventors. Using this procedure yielded141 inventor-founders who originated from 49source firms and founded 114 start-ups. Of these,10 were started by groups of 3 inventor-founders,19 by groups of 2 inventor-founders, and the restby single inventor-founders. It is important to notethat the identification procedure did not require afounder to be an inventor at the start-up. He or sheonly needed to appear as an inventor in the source-firm sample. Further, spin-offs (i.e., incumbentfirm divestitures) and start-ups receiving corporateventure capital from a parent firm in the industry,were excluded from the sample. Source-firmobservations in which the focal firm exited withinthe next two years were also excluded. Finally, toavoid possible confounding effects, excluded wereall mobility events that appeared to occur betweenfirms linked by a merger or an acquisition event.For the combined set of firms, I integrated finan-cial, founding, and exit year data from Compustat,Hoover’s Business Directories, VentureOne, 10-Kfilings, and Lexis-Nexis with patent datafrom Delphion and the National University ofSingapore.
Estimation strategy
I tested the hypotheses using discrete-timeconditional Logit analysis, with the employeeentrepreneurship or mobility events as the positiveoutcome. The models are estimated using pairwise
comparisons (staying vs. mobility, mobility vs.entrepreneurship, etc.) that assume that mobility,staying, and entrepreneurship are independent,non-sequential choices.7 The use of the firm-yearfixed effects significantly simplifies estimationsince all time-variant firm-level controls areabsorbed in the time-variant, firm-fixed effect.To control for individual-level differences, Ideveloped a set of patent-based measures. Thesample was constructed as an unbalanced panelwith the inventor-year observations. To check forthe robustness of the results, I re-estimated themodels using an alternative measure of innovationcomplexity and using the main patent class-timeperiod fixed effects. The class-year fixed effectestimation hinges on comparing individuals whopatent in the same patent class in the focal year.Using class-year fixed effects is a very stringenttest because the estimation hinges only on thewithin-class variation of the complexity measurewhile all across-class differences are subsumedin the fixed effects. However, it addressesthe concern that systematic differences acrosstechnological areas drive the results.
Variables
Dependent variables
The dependent variable for tests of Hypothesis 1was mobility , a binary indicator set to 1 if anemployment spell in a source firm in a focal yearwas followed by a move to a different firm in therecipient sample and 0 if the spell was followedby a further employment at the source firm. Thevariable team mobility (Hypothesis 2) was codedas 1 if the inventor patented together with thesame co-inventor within the parent firm and therecipient firm and 0 otherwise (sub-sample withemployee mobility = 1). For Hypothesis 3, thedependent variable was employee entrepreneur-ship. This binary variable was set to 1 if found-ing a start-up followed an employment spell in afocal year and 0 if joining a rival firm or furtheremployment with the same firm followed (depend-ing on the comparison group). The variable team
7 The multinomial Logit could be an alternative method ofestimation. However, pairwise estimation using conditionalLogit is superior because it allows conditioning on firm year. Theestimation passes the Hausman test of the IIA assumption withχ2 of 0.039 suggesting that nested Logit is not an appropriatemodel.
entrepreneurship (Hypothesis 4a) was coded as 1 ifthe inventor patented together with another inven-tor within the parent firm and both were listedas start-up cofounders and 0 otherwise (subsamplewith employee entrepreneurship = 1).
Main explanatory variable: knowledge complexity
Because a patent is essentially a codified recipe,the complexity of the patent can stand as a proxyfor the complexity of the recipes the inventoracquired while working on the innovation. Thedensity of interdependencies between functionalcomponents of a patent thus represents the den-sity of interdependencies between recipe ingre-dients that the inventor holds. In keeping withprior work (Fleming and Sorenson, 2001, 2004;Sorenson et al ., 2006), I measured knowledgecomplexity by relying on classification of patentsinto subclasses. The NK literature (Kauffman,1993) shows that the ratio between K (the num-ber of interdependencies per component) and N(the number of components) is the main driver ofperformance when solving complex problems.
In this research context, the measure ofinterdependence K is a single-industry measureanalogous to the cross-sectional one used inprior studies (Fleming and Sorenson, 2001,2004; Sorenson et al ., 2006). It is based on theinteraction matrix from Kauffman’s NK model(1993). The key idea behind the measure isthat when two underlying functions (representedby patent subclasses) are coupled, componentsbelonging to these classes are more likely to occurin a single invention. If the functions A and Bare highly coupled, if component a is classifiedin patent subclass A, a∈A, and if component bis in subclass B , b∈B , then one is more likelyto see subclasses a and b in a single invention.In other words, high interdependence between Aand B implies that whenever an inventor solves aproblem related to one of these functions, she/heneeds to redesign or include the coupled functionas well, and the components optimizing thesefunctions are likely to be observed together ina patent. Similarly, if the patent improves thearchitecture of multiple functions, all componentsthat correspond to these functions are likely to becoupled to the architecture. On the other hand,if A and B are independent with respect to eachother, A is likely to be combined with othersubclasses without B being present.
The measure of interdependence K was com-puted in several steps. In the first step, I tabulatedco-occurrence frequencies for all subclass com-binations and also created a table of occurrencefrequency for each subclass. Then, by selectingentries from the tables, I computed the interdepen-dence Ki for each focal component (subclass) ofpatent l :
Interdependence of subclass i ≡ Ki
=∑
j∈l−i
count of patents in subclasses i and j
count of patents in subclass i
(1)
where j belongs to all subclasses except i . Themeasure K for patent l is calculated as:
Interdependence of patent l ≡ Kl
= 1
count of subclasses of patent l
∑
i∈l
Ki
(2)
E.g., when calculating the interdependence of thefirst subclass (a focal i ), the interdependencebetween the first and the third subclasses isthe number of patents in which the first andthird subclasses appear together, divided by thenumber of patents in which only the first subclassappears.
Using a focal industry dataset to derive this mea-sure relies on assuming stability in the nature ofthe interdependencies between the functional com-ponents of an innovation over time in the industry.The variable Ki thus captures the interdependencebetween functions A and B in general and notinterdependence that is ‘patent-specific’. In otherwords, the inventions are assumed to consist ofcomponents that have a certain level of interde-pendence associated with each pair of functionsrepresented by observable components. If func-tions A and B appear on two patents, one in thebeginning of the sample (in terms of calendartime) and another at the end, the interdependencebetween them would be the same. The assumptionof the stability of interdependencies between thesubclasses (‘building blocks’) is not entirely real-istic, but assuming stability within an industry anda certain time frame is a necessary simplification.The measure of K is scaled consistently with theNK model since it is in the interval [0, N – 1].
As has been done in prior studies (Fleming andSorenson, 2001, 2004; Sorenson et al ., 2006), Ioperationalized the total number of components Nby the number of patent subclasses. Following thecited studies, I obtained the complexity measureby dividing the number of interdependencies Kby the number of components N .8 To obtain thefinal measure of knowledge complexity for a giveninventor in a given year, I averaged the K/Nfor all patents awarded to the inventor in thatyear.
To verify the validity of the measure, I inter-viewed two industry experts. One is a professorof electrical engineering and a leading authorityin semiconductor design at a top research institu-tion, and the other is a senior designer holdinga doctoral degree and multiple semiconductorpatents. The experts were asked the followingquestion:
How would you describe the typical inven-tion in a given patent class in terms of itscomplexity? I define inventions with lowcomplexity as those that are composed ofstandardized components that are selected tooptimize a given problem. There are fewinterdependencies (choice of one componentaffects performance of few other compo-nents) between components of these prob-lems. I define inventions with high complex-ity as those that are composed of uniquecomponents that are selected or designedto optimize a given problem. There aremany interdependencies (design of one com-ponent affects performance of many othercomponents) between components of theseproblems.
The respondents answered ‘high,’ ‘medium,’or ‘low’ for each of a series of patent classesthat I identified. Then I aggregated the patents inmy data into main classes and calculated averagecomplexity based on the measure described above.Table 1 shows the correspondence between themeasure and the expert opinions. This validation(a crude one, owing to the aggregation into main
8 Alternatively, one could specify the model using N , K , K /Nand their squared terms (Fleming and Sorenson, 2001). However,using only K /N parsimoniously captures the effect of the fullset of variables, and the robustness checks showed that a fullyspecified model yielded identical results.
class domains) shows that the correspondenceis relatively good, with a correlation of 0.63.9
Both experts agreed with the general idea behindthe measure, while mentioning that with complexproblems, ‘everything talks to everything on thechip’ and ‘you need a close collaboration betweenthe team members .’ Problems are simpler when‘things are standardized ’ and ‘you can drawboundaries around things .’
To check the robustness of the results further, Ideveloped an alternative specification of the mea-sure. A possible concern was that the averagingproduced biases in the measure. To address thisconcern, in Equation 1, I replaced the summa-tion with Max(count of patents in subclasses i andj/count of patents in i) and then, in Equation 2,I replaced 1/(number of subclasses) with Max(.),yielding the most frequently co-occurring subclasspair over all subclasses and patents for the focalinventor in a focal year as a proxy for innovationcomplexity.
Control variables
Beyond the firm-year or patent class-year fixedeffects, all models included a set of controlvariables. To control for individual heterogeneity,I introduced variables capturing inventor quality orother differences that might affect an individual’spropensity to engage in mobility or employeeentrepreneurship and correlate with knowledgecomplexity. I calculated an inventor’s patentingproductivity as the log of the number of patentsthe focal inventor applied for at the source firmdivided by the tenure at the source firm andpatenting quality as the number of citations thefocal inventor received within the next five yearsdivided by the number of patents at the sourcefirm. To supplement the individual quality controlsand to capture gender and race differences inpropensity to exit focal firms (Kim and Marschke,2005), I created the variable female, which iscoded as 1 if the first name on the patentapplication sounds female and 0 otherwise, andthe variable nonwhite, which is set to 1 if the firstand last names on the patent application do notsound of Anglo-Saxon origin, and 0 otherwise.To control for whether the inventor works in
9 The crudeness of the aggregation into main classes does notallow using the expert ranking in the estimation.
Table 1. Knowledge complexity: measure versus questionnaire
# Patents aggregated by main class (domain)Complexity measure
(mean)
Questionnaire (average over tworespondents, complexity is low: 1,
medium: 2, high: 3)
365 Static information storage 0.084 1.5711 Memory 0.092 1323 Power supply 0.109 1.5371 Error detection (circuits and process) 0.112 1327 Non-linear circuits 0.114 2713 Digital processing support 0.122 1.5712 Processors 0.124 1326 Digital logic 0.124 2438 Semiconductor device manufacturing process 0.127 2.5257 Active solid state device 0.129 1710 Input/output data processing design 0.131 2330 Amplifiers 0.131 3395 Processing system organization 0.133 2345 Graphics processing 0.158 1331 Oscillators 0.176 3375 Pulse and digital communication 0.179 2.5324 Measure and testing circuits 0.180 2360 Magnetic storage circuits 0.182 3348 TV circuits 0.218 2702 Data processing calibration systems 0.237 3250 Radiant energy (photocells) 0.238 2.5379 Telephonic communication circuits 0.287 3
Correlation = 0.63(Both experts assigned the same score to 15 out of 22 categories)
the core versus niche technological area withinthe firm, I included the variable proximity tofirm core. It is calculated as the angular distance(Jaffe, 1989) between the ‘technology’ vectors ofthe focal inventor and all other inventors in theparent firm in the focal year. Each dimension ofthe vectors is calculated as the proportion of thepatenting in a focal main class over the focalyear. Further, I included the variable co-inventorsby calculating the log of the average number ofpatent co-inventors at the source firm in a givenyear for the inventor and patenting breadth bycalculating the log of the average number of patentmain classes for the inventor in the focal year tocapture an inventor’s specialization. The variabletenure, measured as the log of the differencebetween the focal year minus the application yearof the first patent within the given parent firmplus one, proxies for the intra-firm experience ofthe inventor. It is also possible that differencesin the opportunity space, both for mobility andemployee entrepreneurship, vary with knowledgecomplexity. Employees may exit to pursue generalopportunities in a given area rather than to exploit
their own complex knowledge. To control for thesedifferences, I introduced variables that rely onthe firm entry and exit rates into a particularcomplexity segment. First, the complexity variablewas split into ten equal-sized bins. The variabledomain attractiveness was calculated as the firmentry rate within the same bin as the focal inventorand year. It is a ratio between the number ofnew firms entering with patents for which thecomplexity is on average in the same bin as thefocal inventor’s patents in the focal year andthe total number of firms with patents appliedfor in the focal year in the same bin. Similarly,the variable domain default risk was calculatedas the firm exit rate within the same bin asthe focal inventor in the focal year. Only actualbankruptcies are considered as exits. It is a ratiobetween the number of firms failing with patentsfor which the complexity is on average in the samebin as the focal inventor’s patents in the focalyear and the total number of firms with patentsapplied for in the focal year in the same bin.Table 2 provides descriptive statistics, includingcorrelations.
In Table 3, models 1–4 show the results ofthe analysis testing Hypotheses 1 and 2. Thesignificant coefficients on the controls indicate thatmore productive employees and nonwhites aremore likely to move to a rival firm, and femaleinventors are less likely to do so. The coefficientson the number of coinventors and patentingbreadth are negative and significant in the mobilityregression, suggesting that inventors embeddedin collaborative networks and generalists are lesslikely to move. Further, tenure strongly predictsmobility. The variable ‘nonwhite’ is negativelyassociated with team mobility while the variable‘co-inventors’ is positively associated with teammobility.
Supporting Hypothesis 1, in model 1 the coef-ficient on complexity is negative and significant.Interpreting the coefficient indicates that a onestandard deviation increase in knowledge com-plexity causes the likelihood of employee mobilityrelative to staying to decrease by 13 percent. Con-sistent with Hypothesis 2, complexity significantlyincreases the likelihood of team mobility (models2–4) relative to individual moves. An increase ofcomplexity by one standard deviation increases thelikelihood of team mobility as opposed to individ-ual mobility by about 40 percent.
In model 5, tenure predicts employeeentrepreneurship while the domain defaultrisk is a negative and significant predictor. Thevariables ‘nonwhite’ and ‘female’ are positivelyassociated with team entrepreneurship (models7–8). In keeping with Hypothesis 3, the coeffi-cient on knowledge complexity is positive andsignificant in model 5. One standard deviationincrease in innovation complexity predicts a28 percent increase in the likelihood of employeeentrepreneurship rather than employee mobility.Model 6 provides an additional test of Hypothesis3 by comparing employee entrepreneurship withstaying at the existing firm and also shows apositive and significant relationship.10 One stan-dard deviation increase in knowledge complexity
10 We would need additional theoretical assumptions to predictthis alternative test. The unconditional version of Hypothesis3 can only hold if the benefits associated with exploitation ofcomplex recipes through entrepreneurship outweigh the transferdifficulties as argued in Hypothesis 1. I thank an anonymousreviewer for this insight.
predicts a 17 percent increase in the likelihood ofemployee entrepreneurship relative to staying.
Consistent with Hypothesis 4a, knowledge com-plexity significantly increases the likelihood ofteam entrepreneurship (models 7–8) relative toindividual entrepreneurship. An increase of com-plexity by one standard deviation increases thelikelihood of team founding by about 55 percent.Using a direct t-test and the coefficient ratio test(Hoetker, 2007) to compare the coefficients ofthe conditional Logit models (model 2 was testedagainst model 7, and model 3 was tested againstmodel 8) reveals that the coefficients on complex-ity for team entrepreneurship and team mobilityare significantly different at the 5 percent level,supporting Hypothesis 4b.11,12
Tables S1 and S2 in the on-line appendix showthe results of the robustness tests. I re-estimated allmodels using an alternative measure of knowledgecomplexity as described above and the main patentclass-year fixed effect. The findings remain robust(Table S1 in the on-line appendix), at least at the10 percent significance level. The coefficients andthe coefficient ratios of the respective models com-paring employee entrepreneurship and employeemobility in Table S2 in the on-line appendix remainstatistically different at the 5 percent level.
DISCUSSION
Employee mobility is a vibrant channel for knowl-edge transfer. Similarly, employee entrepreneur-ship is widely heralded as an important driver ofinnovation, firm formation, and industry growth.Far less is known about how the knowledge con-text affects an employee’s propensity to engageeither in employee entrepreneurship or mobility. Iinvestigated how knowledge complexity affects therelative likelihoods of these outcomes and exam-ined the additional factor of team movements. In
11 Comparing coefficient magnitudes across groups in Logitassumes equal unobserved variance (Hoetker, 2007). Explicittests of the equality of the unobserved variances is not availablewith non-nested samples or conditional Logit. A coefficient ratiotest (Hoetker, 2007) between innovation complexity and moreprecisely estimated controls (nonwhite, co-inventors) was usedand showed consistent results.12 Some compromises were necessary in the conditional Logitmodels of team entrepreneurship and mobility. To mitigate asubstantial loss of observations some models include firm-periodinstead of the firm-year fixed effect. I.e. each combination of afirm and 5-year period was modeled with a fixed effect.
doing so, I shed new light on a theoretical mecha-nism that has received little attention, even thoughlimited evidence and a significant body of mod-eling literature suggest its importance. The studyhighlights that the knowledge context may havewider implications for knowledge flows and indus-try structure.
Drawing on a uniquely rich database of emp-loyee entrepreneurship and mobility events andfirm patenting in the U.S. semiconductor industryover three decades, I found that inventors’ movesto rival firms decrease with the complexity of thework they have done (supporting Hypothesis 1).The finding is consistent with prior modeling andempirical literature suggesting that complexityinhibits knowledge diffusion (Rivkin, 2000; Soren-son et al ., 2006; Williams, 2007). In keeping withHypothesis 3, however, I found that the complex-ity of an inventor’s prior patents positively affectsthe inventor’s propensity to engage in employeeentrepreneurship relative to both mobility toanother firm and staying at the original firm. Onone hand, transferring complex knowledge acrossorganizational boundaries through employeemobility is problematic. On the other hand,exploitation of complex knowledge is more likelyto occur within the context of a newly establishedfirm. Transferring such knowledge to a new firm iseasier and the individuals with complex knowledgemay carry ideas that were not implemented in theirprior organizations. Complexity also dramaticallyincreases the likelihood that employees leave as ateam, with this effect being stronger for employeeentrepreneurship (Hypotheses 2 and 4a, b). Con-sequently, the departure of entire teams solvingcomplex technological problems presents a seriousmisappropriation and competitive threat for incum-bent firms (Campbell et al ., 2012; Wezel et al .,2006). The employee retention strategies thatincumbent firms employ against possible compet-itive entry of their employees are then particularlyrelevant when teams solve complex technologicalproblems.
Further, the results of the study may partlyexplain the ‘start-up phenomenon’—that in somesettings, start-ups are more innovative, betterperformers than established firms (Agarwal et al .,2004; Ganco and Agarwal, 2009; Khessina andCarroll, 2008). At the same time, the findingssuggest a new explanation for why inventorsexiting to start their own firms are likely to have amore negative impact on source firm performance
than inventors exiting to join rival firms (Campbellet al ., 2012).
Limitations and alternative explanations
Both the limitations and the findings of the studypresent avenues for future research. Although thesemiconductor industry represents a canonical con-text for examining my research questions, thesingle-industry focus may limit the generalizabilityof findings. In theory, I would expect knowledgecomplexity to be an important driver of employeeentrepreneurship and mobility patterns in sectorscharacterized by high technological intensity andhigh innovation rates. Following this logic, thefindings should generalize to other knowledge-intensive sectors. However, the results could beless generalizable to settings where the comple-mentary assets are not easily transferrable acrossorganizations (Campbell et al ., 2012; Mitchell,1991).
Since my empirical analysis hinges on the useof patent data to identify employee moves, theobservations are necessarily restricted to instancesin which an inventor was identified on a patentassigned to a source firm and identified as founder(employee entrepreneurship) or appeared as aninventor both in a source and a recipient firm(mobility). Missing from the sample, thus, areinstances in which an individual may have hadinvolvement with or general awareness of adeveloping technology, but no patent. Similarly,technologies that were in the initial stages of devel-opment but not patented prior to an employee’sdeparture are not captured in the study. However, apriori , there is no reason to expect that knowledgecomplexity would differentially affect the behav-ior of inventors who are involved in technologydevelopment without being documented in patents.
The validity of the results hinges on theability to rule out alternative explanations. Thestringent empirical approach, control variables,and a multitude of robustness checks were usedto isolate the effect of knowledge complexityfrom potential confounding factors like individualinventor quality or the heterogeneity in theopportunity space. For instance, higher-qualityinventors may be more likely to solve morecomplex problems. To address this concern,I included multiple individual-level controlscapturing inventor quality and characteristics.Prior research has shown that inter-firm mobility
increases with the inventor’s quality (Hoisl, 2007),which is consistent with my estimates on thepatenting productivity. The fact that knowledgecomplexity is negatively associated with mobilityshould further alleviate the concern that com-plexity is a simple proxy for inventor’s quality.One could also argue that future entrepreneursself-select into complex domains in anticipation ofentrepreneurial opportunities. Although I cannotcompletely rule out this conjecture, it impliessignificant foresight by the future entrepreneurs.They self-select into technological domains withmore frictions and fewer outside mobility options.Existing studies (Garvin, 1983; Klepper andThompson, 2010) have argued that inventors tendto disclose ideas to their employers and leave onlywhen they are rejected. Such findings providefurther evidence against the claim that futureentrepreneurs self-select into complex domainsbecause they anticipate entrepreneurship.13
CONTRIBUTIONS AND CONCLUSION
These limitations notwithstanding, the study makesseveral contributions. In the context of researchon employee entrepreneurship (Agarwal et al .,2004; Klepper, 2007; Klepper and Sleeper, 2005), Idevelop a theoretical mechanism connecting com-plexity with the under-exploitation of knowledgeby existing firms and the subsequent transition toentrepreneurship. The mechanism not only high-lights an additional friction operating within exist-ing firms but also helps to illuminate a key ques-tion in the study of entrepreneurship: ‘why, when,and how some people and not others discover andexploit [entrepreneurial] opportunities’ (Shane andVenkataraman, 2000: 218). Since entrepreneurialopportunities are more likely to reside in com-plex knowledge domains, employees working withsuch knowledge are better positioned to discoverthe opportunities. By focusing on the type of
13 I do not disentangle employee entrepreneurship that occurswith the approval of a parent firm from that which occurs withoutit. The arguments developed here should apply in both cases.Knowledge complexity increases the likelihood that individualsmay discover recipes that will not be used by their parentfirms—whether they leave with or without the firm’s approval.One way of examining the two possibilities is to look at thevariation as driven by the non-compete regimes (e.g. Marx et al .,2009), which I leave for future work. I thank an anonymousreviewer for this suggestion.
knowledge that inventors acquire while solvingtechnological problems, I contribute to buildinga theory of entrepreneurship emphasizing thatentrepreneurial decisions may be driven by knowl-edge or by organizational context (Agarwal et al .,2004; Aldrich and Fiol, 1994; Elfenbein et al .,2010; Sørensen, 2007; Sorenson and Audia, 2000).Such a theory implies that entrepreneurial propen-sities could be actively influenced by the assign-ment of tasks and the management of knowledgeacquisition.
The study contributes to the literature on knowl-edge spillovers (Rosenkopf and Almeida, 2003;Singh and Agrawal, 2011; Sorenson et al ., 2006).I show that, even though complex knowledge maynot be readily imitated by other firms, it maystill flow to startups through entrepreneurship. Thestudy thus implies that the transfer of complexknowledge that is inhibited by the mechanismsoperating within existing recipient firms can beovercome in appropriate organizational settings.
Further, I contribute to the recent literatureon innovative teams (Singh and Fleming, 2010;Wuchty et al ., 2007) by showing that retainingcollaborative teams allows the transfer of acquiredcomplex knowledge across organizational bound-aries. It constitutes a step toward a theory in whichcollaborative work is an integral part in the discov-ery, exploitation, and transfer of knowledge.
Unique to this study is the combination ofemployee entrepreneurship, employee mobility,and complexity. Employee entrepreneurship andmobility are phenomena that have been typicallystudied in isolation. Examining them jointly asboth driven by knowledge complexity allows forthe gaining of insights about the mechanisms ofknowledge exploitation and transfer. Extending therecent work that empirically tested the insightsgained in agent-based models (Lenox, Rockart,and Lewin, 2010; Sorenson et al ., 2006), I alsocontribute to the complexity literature by showingan empirical application of the modeling insightswithin a new context of entrepreneurship.
The study also has implications for theknowledge-based view of the firm (Grant, 1996;Kogut and Zander, 1996; Nickerson and Zenger,2004). It highlights that mechanisms associatedwith knowledge coordination and transfer withinoriginal firms may lead to the generation ofknowledge that falls outside of the boundaries ofthe firm when the problems are complex.
In summary, I theorized and found evidencethat the complexity of knowledge acquired whilesolving technological problems is an importantdriver of mobility and entrepreneurship decisions.The study sheds new light on an importantcontingency while revealing promising pathwaysfor continued research.
ACKNOWLEDGEMENTS
I am grateful to Editor Will Mitchell and twoanonymous reviewers for their invaluable guid-ance. This project would not have been possi-ble without the generous financial support of theEwing Marion Kauffman Foundation through itsDissertation Fellowship Program. The manuscripthas benefited significantly from the comments ofRajshree Agarwal, Ajay Agrawal, Thomas Aste-bro, Hari Bapuji, Janet Bercovitz, Oana Branzei,Ben Campbell, Seth Carnahan, April Franco, Shra-van Gaonkar, Glenn Hoetker, Aseem Kaul MarvinLieberman, Joe Mahoney, Steve Michael, MylesShaver, Shawn Riley, Harry Sapienza, NareshShanbhag, Deepak Somaya, Olav Sorenson, PKToh, Shaker Zahra and Rosemarie Ziedonis. Allremaining errors are my own.
REFERENCES
Agarwal R, Echambadi R, Franco AM, Sarkar M.2004. Knowledge transfer through inheritance: spin-out generation, development, and survival. Academyof Management Journal 47(4): 501–522.
Agarwal R, Ganco M, Ziedonis RH. 2009. Reputationsfor toughness in patent enforcement: implications forknowledge spillovers via inventor mobility. StrategicManagement Journal 30(13): 1349–1374.
Agrawal A. 2006. Engaging the inventor: exploringlicensing strategies for university inventions andthe role of latent knowledge. Strategic ManagementJournal 27(1): 63–79.
Aldrich HE, Fiol CM. 1994. Fools rush in? Theinstitutional context of industry creation. Academy ofManagement Review 19(4): 645–670.
Almeida P, Kogut B. 1999. Localization of knowledgeand the mobility of engineers in regional networks.Management Science 45(7): 905–917.
Anton JJ, Yao DA. 1995. Start-ups, spin-offs, and internalprojects. Journal of Law, Economics and Organization11(2): 362–378.
Brittain J, Freeman JH. 1980. Organizational proliferationand density dependent selection. In OrganizationalLife Cycles , Kimberly J, Miles M (eds). Jessey-Bass:San Francisco, CA; 291–338.
Busenitz LW, Barney JB. 1997. Differences betweenentrepreneurs and managers in large organizations:biases and heuristics in strategic decision-making.Journal of Business Venturing 12(1): 9–30.
Campbell BA, Ganco M, Franco A, Agarwal R. 2012.Who leaves, where to, and why worry? Employeemobility, employee entrepreneurship, and effects onsource firm performance. Strategic Management Jour-nal 33(1): 65–87.
Cassiman B, Ueda M. 2006. Optimal project rejectionand new firm start-ups. Management Science 52(2):262–275.
Chatterji AK. 2009. Spawned with a silver spoon?Entrepreneurial performance and innovation in themedical device industry. Strategic Management Jour-nal 30(2): 185–206.
Computer History Museum 2009. http://www.computerhistory.org (accessed 5 May 2009)
Corporate History 2009. http://www.garmin.com(accessed 12 May 2009).
Cypress Semiconductor 2009. MOS is metal oxidesemiconductor. http://www.cypress.com (accessed 3May 2009).
Dosi G, Grazzi M. 2010. On the nature of technolo-gies: knowledge, procedures, artifacts and produc-tion inputs. Cambridge Journal of Economics 34:173–184.
Dushnitsky G, Shapira Z. 2010. Entrepreneurial financemeets corporate reality: comparing investment prac-tices and performance of corporate and indepen-dent venture capitalists. Strategic Management Jour-nal 31(9): 990–1017.
Eisenhardt KM, Schoonhoven CB. 1990. Organizationalgrowth: linking founding team, strategy, environ-ment, and growth among US semiconductor ventures,1978–1988. Administrative Science Quarterly 35(3):504–529.
Elfenbein DW, Hamilton BH, Zenger TR. 2010. Thesmall firm effect and the entrepreneurial spawning ofscientists and engineers. Management Science 56(4):659–681.
Fleming L, Sorenson O. 2001. Technology as a complexadaptive system: evidence from patent data. ResearchPolicy 30(7): 1019–1039.
Fleming L, Sorenson O. 2004. Science as a map intechnological search. Strategic Management Journal25(8–9): 909–928.
Fortune A, Mitchell W. 2012. Unpacking firm exit at thefirm and industry levels: the adaptation and selectionof firm capabilities. Strategic Management Journal33(7): 794–819.
Franco AM, Filson D. 2006. Spin-outs: knowledge diffu-sion through employee mobility. The Rand Journal ofEconomics 37(4): 841–860.
Freeman J. 1986. Entrepreneurs as organizational prod-ucts: semiconductor firms and venture capital firms.In Advances in the Study of Entrepreneurship, Inno-vation, and Economic Growth , Libecap GD (ed). JAIPress: Greenwich, CT; 33–52.
Ganco M, Agarwal R. 2009. Performance differen-tials between diversifying entrants and entrepreneurial
start-ups: a complexity approach. Academy of Man-agement Review 34(2): 228–252.
Garvin DA. 1983. Spin-offs and the new firm forma-tion process. California Management Review 25(2):3–20.
Gavetti G, Levinthal D. 2000. Looking forward andlooking backward: cognitive and experiential search.Administrative Science Quarterly 45: 113–137.
Grant RM. 1996. Toward a knowledge-based theory ofthe firm. Strategic Management Journal 17: 109–122.
Hall BH, Ziedonis RH. 2001. The patent paradoxrevisited: an empirical study of patenting in theUS semiconductor industry, 1979–1995. The RandJournal of Economics 32(1): 101–128.
Helfat CE, Lieberman MB. 2002. The birth of capabili-ties: market entry and the importance of pre-history.Industrial and Corporate Change 11(4): 725–760.
Hellman T. 2007. When do employees becomeentrepreneurs? Management Science 53(6): 919–933.
Hoetker G. 2007. The use of Logit and Probit modelsin strategic management research: critical issues.Strategic Management Journal 28(4): 331–343.
Hoetker G, Agarwal R. 2007. Death hurts, but it isn’tfatal: the postexit diffusion of knowledge createdby innovative companies. Academy of ManagementJournal 50(2): 446–467.
Hoisl K. 2007. Tracing mobile inventors—the causalitybetween inventor mobility and inventor productivity.Research Policy 36(5): 619–636.
Jaffe AB. 1989. Characterizing the ‘technological posi-tion’ of firms, with application to quantifying techno-logical opportunity and research spillovers. ResearchPolicy 18(2): 87–97.
Kauffman SA. 1993. The Origins of Order: Self Organi-zation and Selection in Evolution . Oxford UniversityPress: New York.
Kenney M, Florida R. 2002. Venture capital in SiliconValley: fueling new firm formation. In UnderstandingSilicon Valley , Kenney M (ed). Stanford UniversityPress: Stanford, CA; 98–123.
Khessina OM, Carroll GR. 2008. Product demography ofde novo and de alio firms in the optical disk driveindustry, 1983–1999. Organization Science 19(1):25–38.
Kim J, Marschke G. 2005. Labor mobility of scientists,technological diffusion, and the firm’s patentingdecision. The Rand Journal of Economics 36(2):298–317.
Kirzner IM. 1997. Entrepreneurial discovery and thecompetitive market process: an Austrian approach.Journal of Economic Literature 35(1): 60–85.
Klepper S. 2007. Disagreements, spinoffs, and the evo-lution of Detroit as the capital of the U.S. automobileindustry. Management Science 53(4): 616–631.
Klepper S, Sleeper S. 2005. Entry by spinoffs. Manage-ment Science 51(8): 1291–1306.
Klepper S, Thompson P. 2010. Disagreements and intra-industry spinoffs. International Journal of IndustrialOrganization 28(5): 526–538.
Kogut B, Zander U. 1996. What firms do? Coordination,identity, and learning. Organization Science 7(5):502–518.
Lenox MJ, Rockart SF, Lewin AY. 2010. Does inter-dependency affect firm and industry profitability? Anempirical test. Strategic Management Journal 31:121–139.
Levinthal DA. 1997. Adaptation on rugged landscapes.Management Science 43(7): 934–950.
Lowe RA. 2002. Invention, innovation and entrepreneur-ship: the commercialization of university research byinventor-founded firms. Unpublished diss., Universityof California: Berkeley, CA.
Macher JT, Mowery DC, Hodges DA. 1998. Reversal offortune? The recovery of the US semiconductor indus-try. California Management Review 41(1): 107–136.
Marx M, Strumsky D, Fleming L. 2009. Mobility,skills, and the Michigan non-compete experiment.Management Science 55(6): 875–889.
Mitchell W. 1991. Dual clocks: entry order influences onincumbent and newcomer market share and survivalwhen specialized assets retain their value. StrategicManagement Journal 12(2): 85–100.
National Inventors Hall of Fame Foundation, Inc 2009.http://www.invent.org/hall_of_fame/55.html (accessed15 May 2009).
Nelson RR, Winter SG. 1982. An Evolutionary Theory ofEconomic Change. Harvard University Press: Cam-bridge, MA.
Nickerson JA, Zenger TR. 2004. A knowledge-basedtheory of the firm—the problem-solving perspective.Organization Science 15(6): 617–632.
Pitta J. 1997. Federico Faggin: the electronic brain.Forbes.com , 7 July. http://www.forbes.com/forbes/1997/0707/6001312a.html (accessed 17 April 2009).
Phillips DJ. 2002. A genealogical approach to organiza-tional life chances: the parent-progeny transfer amongSilicon Valley law firms, 1946–1996. AdministrativeScience Quarterly 47(3): 474–506.
Polanyi M. 1983. The Tacit Dimension . Reprinted byPeter Smith: Gloucester, MA.
QuickLogic 2009. http://www.quicklogic.com (acces-sed 21 March 2009).
Rivkin JW. 2000. Imitation of complex strategies.Management Science 46(6): 824–844.
Rivkin JW, Siggelkow N. 2002. Organizational stickingpoints on NK landscapes. Complexity 7: 31–43.
Rosenkopf L, Almeida P. 2003. Overcoming local searchthrough alliances and mobility. Management Science49(6): 751–766.
Sarasvathy SD. 2001. Causation and effectuation: towarda theoretical shift from economic inevitability toentrepreneurial contingency. Academy of ManagementReview 26(2): 243–263.
Sarasvathy D, Simon HA, Lave L. 1998. Perceivingand managing business risks: differences betweenentrepreneurs and bankers. Journal of EconomicBehavior and Organization 33(2): 207–225.
Shane S. 2000. Prior knowledge and the discoveryof entrepreneurial opportunities. Organization Science11(4): 448–469.
Shane S, Venkataraman S. 2000. The promise ofentrepreneurship as a field of research. Academy ofManagement Review 25(1): 217–226.
Singh J, Agrawal AK. 2011. Recruiting for ideas: howfirms exploit the prior inventions of new hires.Management Science 57(1): 129–150.
Singh J, Fleming L. 2010. Lone inventors as sources ofbreakthroughs: myth or reality? Management Science56(1): 41–56.
Sørensen JB. 2007. Bureaucracy and entrepreneurship:workplace effects on entrepreneurial entry. Adminis-trative Science Quarterly 52(3): 387–412.
Sorenson O, Audia PG. 2000. The social structure ofentrepreneurial activity: geographic concentration offootwear production in the United States, 1940–1989.American Journal of Sociology 106(2): 424–462.
Sorenson O, Rivkin JW, Fleming L. 2006. Complexity,networks and knowledge flow. Research Policy 35(7):994–1017.
Stuart TE, Sorenson O. 2003. Liquidity events andthe geographic distribution of entrepreneurial activity.Administrative Science Quarterly 48(2): 175–201.
Teece DJ. 2003. Expert talent and the design of(professional services) firms. Industrial and CorporateChange 12(4): 895–916.
Wezel FC, Cattani G, Pennings JM. 2006. Competitiveimplications of interfirm mobility. Organization Sci-ence 17(6): 691–709.
Williams C. 2007. Transfer in context: replication andadaptation in knowledge transfer relationships. Strate-gic Management Journal 28(9): 867–889.
Wuchty S, Jones BF, Uzzi B. 2007. The increasingdominance of teams in production of knowledge.Science 316(5827): 1036–1040.
SUPPORTING INFORMATION
Additional supporting information may be foundin the online version of this article:
![HIGH COURT OF AUSTRALIA - mcmullansolicitors.commcmullansolicitors.com/.../Westport-Insurance-Corporation-v-Gordian... · Westport Insurance Corporation v Gordian Runoff Limited [2011]](https://static.documents.pub/doc/80x56/5b313b6d7f8b9a2c328c56cb/high-court-of-australia-westport-insurance-corporation-v-gordian-runoff-limited.jpg)
