entrepreneurship in the large corporation: a longitudinal …...entrepreneurship in the large...

Strategic Management JournalStrat. Mgmt. J.,22: 521–543 (2001)

DOI: 10.1002/smj.176

ENTREPRENEURSHIP IN THE LARGECORPORATION: A LONGITUDINAL STUDY OF HOWESTABLISHED FIRMS CREATE BREAKTHROUGHINVENTIONSGAUTAM AHUJA* and CURBA MORRIS LAMPERTCollege of Business Administration, University of Texas at Austin, Austin, Texas,U.S.A.

We present a model that explains how established firms create breakthrough inventions. Weidentify three organizational pathologies that inhibit breakthrough inventions: the familiaritytrap – favoring the familiar; the maturity trap – favoring the mature; and the propinquitytrap – favoring search for solutions near to existing solutions. We argue that by experimentingwith novel (i.e., technologies in which the firm lacks prior experience), emerging (technologiesthat are recent or newly developed in the industry), and pioneering (technologies that do notbuild on any existing technologies) technologies firms can overcome these traps and createbreakthrough inventions. Empirical evidence from the chemicals industry supports our model.Copyright 2001 John Wiley & Sons, Ltd.

INTRODUCTION

Radical or ‘breakthrough’ inventions lie at thecore of entrepreneurial activity and wealth cre-ation (Kirchhoff, 1991; Schumpeter, 1975). Suchinventions serve as the basis of new technologicaltrajectories and paradigms and are an importantpart of the process of creative destruction inwhich extant techniques and approaches arereplaced by new technologies and products. Mostacademic studies, as well as reports in the popularpress, have focused on the role of new firms inthe creation of such breakthroughs (Metheet al.,1997). This focus is understandable since severalstudies show that breakthrough inventions areoften likely to originate with entrants rather thanincumbents (Cooper and Schendel, 1976; Foster,1986). However, recent research suggests thatestablished firms may actually be contributing to

Key words: radical innovation; corporate entrepreneurship;organizational learning*Correspondence to: Gautam Ahuja, Management Department,College of Business Administration, 4.202, The University ofTexas at Austin, Austin, TX 78712, U.S.A.

Copyright 2001 John Wiley & Sons, Ltd.

breakthrough inventions to a far greater extentthan is generally recognized, and, in some indus-tries, may even dominate this process (Metheetal., 1997). Thus, contrary to common perceptions,it appears that at least some large firms are ableto establish routines that enable them to generatesignificant technological breakthroughs, and rein-vent themselves and retain technological leader-ship in their industry. In this study we examinethe issue of how established firms create suchbreakthrough inventions.

Understanding how large, established firms cre-ate breakthrough inventions has rich theoreticaland practical implications from the perspectives ofentrepreneurship, technology strategy and organi-zational learning. As Venkataraman (1997) notes,‘Entrepreneurship as a scholarly field seeks tounderstand how opportunities to bring into exis-tence “future” goods and services are discovered,created and exploited, by whom and with whatconsequence.’ Further, he elaborates, trueentrepreneurship entails the creation of both privatewealth and social benefit (Schumpeter, 1975; Ven-kataraman, 1997). Breakthrough inventions of thekind that are studied here possess all three of these

522 G. Ahuja and C. M. Lampert

concomitants. Almost by definition breakthroughinventions serve as the basis of ‘future’ technol-ogies, products and services. Further, research findsthat breakthrough inventions are related to thecreation of private wealth and the generation ofstreams of Schumpeterian rents for their inventors(Harhoff et al., 1999), while also enhancing socialwelfare (Trajtenberg, 1990a, 1990b).

Examining the nexus between such break-through inventions and large established corpo-rations also provides insights into the processesof corporate entrepreneurship. Although thestereotype of the solitary inventor toiling in agarage adds a memorably heroic dimension tothe breakthrough invention story, the fact remainsthat a very large proportion of R&D resourcescontinue to be expended by established, publiclyheld corporations. Identifying strategies that canhelp such corporations to improve their recordof breakthrough inventions can potentially createsignificant private and social value. Further,beyond simply wealth creation, for establishedcorporations technological breakthroughs canserve as internally generated opportunities forcorporate reinvention, business growth, and newbusiness development (Burgelman, 1983).Research suggests that the routines of the largeestablished firm that ensure reliable throughputand output also entail formalization and bureau-cratization, and sometimes even obsolescence anddeath, as the organization’s fit with a changingenvironment deteriorates (Hannan and Freeman,1989; Sorensen and Stuart, 2000). Studying entre-preneurial behavior in large corporations canpresent important insights for corporate rejuven-ation (Covin and Miles, 1999).

Exploring the determinants of breakthroughinventions is also of importance from the perspec-tive of technology strategy and organizationallearning. Breakthrough inventions represent rare,valuable, and potentially inimitable sources ofcompetitive advantage (Barney, 1991). Under-standing how large corporations create techno-logical breakthroughs and sustain their preemi-nence in an industry is of fundamental concernto strategy theorists trying to explain durable orsustained superior performance. Relatedly, organi-zational learning theorists have argued that learn-ing creates its own traps: as organizations developcapabilities that improve immediate performancethey often simultaneously reduce competence withrespect to new paradigms that may hold the

Copyright 2001 John Wiley & Sons, Ltd. Strat. Mgmt. J.,22: 521–543 (2001)

key to future performance (Levinthal and March,1993). Identifying the kinds of corporate activitiesthat help firms escape suchlearning trapsis thenof significant importance.

In the sections that follow we integrate theentrepreneurship and organizational learningliteratures to develop a theoretical model thatexplains how established firms create fundamentaltechnological breakthroughs. Our model has thefollowing key features. First, from the entrepre-neurial literature we draw on the notion thatdiversity and experimentation within the largecorporation are central to successful entrepre-neurial activity (Burgelman, 1983; Lant andMezias, 1990; McGrath and MacMillan, 2000;Mezias and Glynn, 1993). Second, from theorganizational learning literature we draw on theidea that the dynamics of established organi-zations make the provision of such diversity dif-ficult, leading organizations into learning trapsthat favor specialization and inhibit experimen-tation (Levinthal and March, 1993; March, 1991).Thus, we argue that the essential constraints tothe ability of large firms to create breakthroughinventions stem in large part from practices thatare both necessary and efficient for them. Third,we suggest that in the context of breakthroughinventions such learning traps are manifested inthree types of organizational pathologies: a ten-dency to favor the familiar over the unfamiliar;a tendency to prefer the mature over the nascent;and a tendency to search for solutions that arenear to existing solutions rather than search forcompletely de novo solutions. We call these threepathologies thefamiliarity trap, the maturity trap,and thepropinquity trap, respectively, and arguethat each of these is grounded in significantimmediate benefits for firms, but eventually con-strains their ability to create breakthrough inven-tions that hold the key to future performance.Finally, by expanding and elaborating the notionof learning traps in this fashion, we identifyspecific strategies that organizations can use tocounter these pathologies. Specifically, we sug-gest that by experimenting withnovel, emerging,and pioneering technologies firms can overcomethe liabilities of these traps and successfully cre-ate breakthrough inventions.1

1 Learning traps (Levinthal and March, 1993) are closelyrelated to another, more familiar construct: competency traps(Levitt and March, 1988: 322). Competency traps are defined

Entrepreneurship in the Large Corporation 523

Novel technologies are technologies that arenew or unfamiliar to the firm (i.e., they aretechnologies in which the firm lacks priorexperience), even though they possibly haveexisted in the industry before.Emergingtechnol-ogies are leading-edge technologies that are recentor new to the entire industry (as distinct fromolder, mature technologies).Pioneering technol-ogies are technologies that have no technologicalantecedents (i.e., they represent technologies thatdo not build on any existing technologies). Wetest our arguments with longitudinal data on theinvention output of the leading firms in thechemical industry to demonstrate that these strate-gies are predictive of a firm’s record of break-through invention.2

THEORY AND HYPOTHESES

The distinction between invention and innovationis an important one. Invention refers to the devel-opment of a new idea or an act of creation;innovation refers to the commercializing of theinvention (Hitt, Hoskisson, and Nixon, 1993: 162;Schumpeter, 1934). In this study, for clarity andfocus we restrict our attention to the creation ofthe actual inventions rather than their subsequentcommercialization. Breakthrough inventions can

to occur ‘when favorable performance with an inferior pro-cedure leads an organization to accumulate more experiencewith it, thus keeping experience with a superior procedureinadequate to make it rewarding to use’ (Levitt and March,1988: 322). Learning traps, on the other hand, embody theconflict between routines that enable the organization to per-form well in the short run but may position the organizationunfavorably for the future. Thus, while competency trapsentail choices between two procedures or routines targetedtowards the same outcome, the learning traps we discuss hereare about the implications of the same routines for twodifferent outcomes such as reliable and predictable outputsthat are necessary for immediate or short-run performance,and breakthrough inventions that may form the basis ofsuperior performance in the future.2 A relevant issue is whether these three strategies representactions to overcome learning traps, or a more complex global‘ability to learn.’ From our perspective a firm’sability to learnis a broad construct that is likely to subsume many forms oflearning (see, for instance, Huber, 1991; Lei, Hitt, and Bettis,1996). In this study we focus on the firm’s ability to overcomethree specific forms of weaknesses or traps. Although the over-coming of these traps is probably just one facet of a firm’s overallability to learn, we believe that this specific characterization isuseful in identifying specific countering strategies. The iden-tification of these countering strategies would not be possible orat least not as easy if we focused on the broader, but lesstangible, construct of a global ability to learn.


provide a unique competitive advantage andattendant rents to the inventing organization(Achilladelis, Schwarzkopf, and Cines, 1990; Har-hoff et al., 1999). Further, the capacity to createbreakthrough or radical inventions can itself beregarded as a form of meta-learning or dynamiccore competence (Leiet al., 1996; Prahalad andHamel, 1990) reflecting a firm’s unique andspecialized problem-solving capabilities. Thus,from both practical and theoretical perspectives,understanding the determinants of breakthroughinventions at the firm level is important.

Radical or breakthrough inventions can be definedalong different dimensions. At a very basic level adistinction can be made between inventions that areradical from a technological perspective vs. inven-tions that are radical from a user or market perspec-tive. In this research we focus on the technologicalimportance of an invention in classifying it asbreakthrough or radical (Trajtenberg, 1990a, 1990b;Podolny and Stuart, 1995; Rosenkopf and Nerkar,2001), and accordingly define breakthrough inven-tions as those foundational inventions that serveas the basis for many subsequent technologicaldevelopments (Trajtenberg, 1990a, 1990b). Thetechnological importance of inventions can varysignificantly. While some inventions open new pathsof technological progress and spawn many sub-sequent inventions, others are technological deadends (Dosi, 1988; Fleming, 1999; Podolny andStuart, 1995; Sahal, 1985). Inventions that serve asthe source of many subsequent inventions can beregarded as breakthrough or radical because theyhave demonstrated their utility on the path of tech-nological progress (Achilladeliset al., 1990; Flem-ing, 1999). Past research suggests that such inven-tions that open the door to many subsequentinventions have considerable technological and eco-nomic value (Trajtenberg, 1990a; Rosenkopf andNerkar, 2001). Although technologically importantinventions can also be radical inventions from theperspective of a user, neither our theory nor ourdata permit us to extend our arguments to a user-based concept of radicality. Accordingly, we limitthe domain of our theory and empirical claims toinventions that are technologically important.

Prior research on corporate entrepreneurshipand breakthrough inventions

Recent research has compellingly argued thatcorporate entrepreneurship adds value not only


by utilizing resources in new ways but also,perhaps more importantly, by creating newresources (Zahra, Jennings, and Kuratko, 1999;Floyd and Wooldridge, 1999; Greene, Brush, andHart, 1999). Prominent among these createdresources are knowledge and the various knowl-edge outcomes of the corporate entrepreneurshipprocess (Zahraet al., 1999; Hitt et al., 1999).Consistent with this theme several recent studieshave looked at knowledge-related outcomes ofcorporate entrepreneurship. However, the focusof many of these studies is on innovation, in theform of new product development, rather thaninvention, the act of creating a technologicalbreakthrough. For instance, Hittet al. (1999) haveexamined the role of cross-functional teams inthe design and development of new products,finding that elements of team context such astop management team support and organizationalpolitics have a more significant influence on teamsuccess than internal team characteristics. Simi-larly, Koberg, Uhlenbruck, and Sarason (1996)examine the moderating effect of the life cyclestage of a venture on the organizational andenvironmental determinants of product inno-vation. Relatively little research has examinedinventionsas the outcome variable of the entre-preneurial process. Yet, technologically radicalinventions can be regarded as opportunities oroptions that are subsequently exploited throughnew ventures or commercialization within theexisting businesses. Thus, research on the deter-minants of breakthrough inventions complementsthe above studies that focus on the exploitationof opportunities, by trying to establish an under-standing of thecreation of opportunities. Indeed,to the extent that without inventions there are noinnovations, improving our understanding of thestrategies that lead to breakthrough inventionsis critical.

Large-sample studies of breakthroughinven-tions are relatively rare even in the technologyarea. In common with the literature on corporateentrepreneurship, studies that have examined theissue of radical technological breakthroughs havemore often focused on the commercialization ofinventions or the introduction of new productsrather than on the actual inventions themselves(Anderson and Tushman, 1990; Mitchell, 1989;Tushman and Anderson, 1986). For instance, acommon objective across many of these studieshas been the identification of the relative impor-


tance and likelihood of incumbents and entrantsas sources of successful innovations (Cooper andSchendel, 1976; Foster, 1986; Henderson, 1993;Metheet al., 1997; Tushman and Anderson, 1986;Utterback, 1994). The few studies examining thecorporate creation of breakthrough inventionshave focused on the impact of such inventionsrather than their creation (for instance, Achillad-elis et al., 1990; Trajtenberg, 1990b). Similarly,in-depth case studies of breakthrough inventions(Foster, 1986; Kusunoki, 1997; Nayak and Ketter-ingham, 1986) or of corporate practices fosteringsuch inventions (Brown, 1991) have providedwonderfully rich and insightful characterizationsof the invention process; however, the task ofsynthesizing these individual findings into a for-mal model of breakthrough invention in largecorporations remains. In this research we attemptto address this gap.

Established firms and breakthroughinventions

The failure of large firms to create breakthroughinventions can be understood through either theirlack of motivation (the economic perspective) ortheir lack of ability (the organizationalperspective) (Henderson, 1993). In this study wecontrol for economic motivations such as theprofitability of firms in our empirical work butfocus our theory development largely on theorganizational question of why large firms mayfail to create breakthrough inventions. We usethe most influential patents (defined in greaterdetail later) in the chemicals industry over an 8-year period as our indicators of breakthroughinvention and develop a framework to explainthe strategies that led to the creation of suchinventions. We focus especially on why the verynature of being an established firm creates ten-sions with regard to the generation of break-through inventions.

Our model of breakthrough inventions in estab-lished firms begins with three basic premisesdrawn from the organizational learning literature.First, we presume that organizational behavior isbased on routines (Nelson and Winter, 1982;Levitt and March, 1988). Second, we presumethat routines are oriented to targeted outcomes(Levitt and March, 1988; Simon, 1955). Eachorganization is subject to a set of external andinternal objectives, and routines are oriented to


the accomplishment of these objectives. Third,we assume that in successful or establishedorganizations, i.e., those organizations that surviveand mature in the organizational ecology, organi-zational routines and actions are path-dependentand therefore based on interpretations and out-comes of past actions (Leiet al., 1996; Levittand March, 1988). Routines that are associatedwith success in a situation are replicated andperpetuated, while those associated with failureare discarded or modified. Over time this processof winnowing of unproductive routines and repli-cation of successful ones, combined with thesecond premise identified above, ensures that inestablished organizations routines are specializedtowards very specific outcomes.

A firm’s survival and success are eventuallybased on its ability to meet at least three sets ofobjectives. First, it must satisfy some marketdemands or needs. Without an external demandfor its outputs, the firm must eventually die.Second, in a competitive environment, to succeeda firm must attempt to develop a competitiveadvantage over other firms seeking to offer prod-ucts to the same markets. Third, a firm needs toestablish an internally consistent set of throughputprocesses that ensure that the above output andcompetitive advantage demands are met. Estab-lished firms, or firms that emerge as leaders orsurvivors in an industry, are those that have metthese three objectives in at least a minimalfashion.

Although a firm must meet several require-ments to satisfy market demands, its ability toprovide reliable high-quality outputs in anefficient and predictable fashion is likely to bekey to its success and survival (Hannan andFreeman, 1989). Similarly, from the perspectiveof obtaining a competitive advantage, a firmneeds to develop a distinctive competence, aunique capability housed in the organization thatdifferentiates it from its competitors (Hitt andIreland, 1985). Finally, from the perspective ofinternal consistency it is important that a firm’sstructure and systems conform to its strategy(Burgelman, 1985; Dess, Lumpkin, and McGee,1999). The greater the effectiveness with whicha firm can accomplish these three objectives, thehigher the likelihood of its survival, at least inthe context of a stable environment.

By developing and refining a competence, byproviding reliable outputs, and by operating a set


of internal controls and processes that ensure thefirst two outcomes, a firm can enjoy the benefitsof external and internal consistency. Interestingly,these very attributes can also serve to limit thefirm’s effectiveness at breakthrough invention. Ina set of processes that we describe in greaterdetail slightly later, we note that each of theseperformance-enhancing attributes that enable acorporation to survive and establish itself alsopotentially entail a significant dark side. Theimpetus to provide reliable and predictable so-lutions focuses a firm’s attention on maturetechnologies (Christensen and Rosenbloom, 1995;Sorensen and Stuart, 2000). The impetus todevelop a competitive advantage favors the reten-tion of routines that lead to distinctive com-petence and specialization, rather than experimen-tation (Levinthal and March, 1993). Finally, thenecessity of establishing control to accomplishthe first two objectives leads to bureaucratic pro-cedures and structures that favor searching proxi-mate domains of technology, rather than itsunknown nether regions that may hold eventuallymore effective, butex antemore uncertain andunknown, solutions. As we argue below, thisfocus on the familiar, the mature, and the proxi-mate may serve to limit the likelihood of creatinga truly breakthrough invention.

The above, prognosis however, presumes thedominance of single-loop learning among organi-zations (Argyris, 1983). The retention of routinesthat enhance reliability, specialization, and controlmay enable an organization to prosper in a steadystate. However, the reality of business indicatesthat significant innovation is likely to be animportant dimension of firm performance formany firms. In such situations, the possibility ofdouble-loop learning suggests a second dynamicthat may be operative on at least some firms(Lei et al., 1996). Recognizing that the above-described routines lead to significant deficienciesin their capabilities to create breakthrough inven-tions, some firms may consider a second loop ofactivity that enables them to counter this dysfunc-tional outcome of the primary loop. Accordingly,instead of seeking to develop just a primary setof competencies at producing output, firms maytarget the development of more dynamic capabili-ties (Lei et al., 1996; Deeds, DeCarolis, andCoombs, 1999). Prominent among such capabili-ties is the development of heuristics and insightsto define and solve complex technological prob-


lems, problems of the kind that can lead totechnological breakthroughs (Leiet al., 1996).The integration of these second-order capabilitiesof problem definition and solution, with the pri-mary capabilities at output generation describedearlier, can serve as a form of meta-learning forthese firms and provide a basis for even moresignificant and durable competitive advantages(Lei et al., 1996; Prahalad and Bettis, 1986).This incentive, of developing a highly tenablecompetitive position, can serve as a basis for thereevaluation of the firm’s existing routines andhelp the firm to identify strategies to counterthese deficiencies. In the sections that follow weinvestigate in some detail both these deficienciesand the strategies firms may follow to over-come them.

The familiarity trap and novel technologies

Received research suggests that increasing returnsto experience, or mutual positive feedbackbetween experience and competence, make therefinement of familiar technologies preferable tothe exploration of new ones (Levinthal andMarch, 1993; March, 1991). Experience with atechnology leads to enhanced absorptive capacityand increased competence with the technology(Cohen and Levinthal, 1990). Greater competencewith a technology fosters increased usage, andhence increases experience with the technology(Cohen and Levinthal, 1990). This cycle ofexperience and competence is rewarding in termsof enabling the organization to build a specializedcompetence. However, the increased ease oflearning and problem solving in specific direc-tions made possible by enhanced absorptivecapacity and competence in those areas makesthe adoption of alternate directions of develop-ment less attractive and potentially less rewarding.Since developing deeper expertise with familiarknowledge bases yields more immediate andlikely returns, it is preferred to investing in unfa-miliar technologies (Cohen and Levinthal, 1990;Levinthal and March, 1993). Unfortunately, thispath dependence increases the risks of the organi-zation falling into afamiliarity trap.

As experience and competence in a specific setof technologies accumulate, knowledge architec-tures reify (Henderson and Clark, 1990). Cogni-tive maps become increasingly rigid and existing,dominant, paradigmatic solutions are applied to


all problems (Leonard-Barton, 1992; March,1991). The reduction in experimentation and theinvocation of a dominant and familiar paradigmto address all problems reduces the probabilitythat a distinct, radically different approach tosolving a given problem will emerge (Fiol andLyles, 1985; Leiet al., 1996). However, a givenset of routines and competencies can address onlyso many problems in an effective fashion. Thelikelihood that the optimal, or even a highlyeffective, solution to a problem will be discovereddiminishes as the range of problems addressedwith a given set of competencies is increased.Concurrently, the likelihood that some principleswill be inappropriately applied rises as a con-strained set of competencies is applied to moreand varied technological problems. Withoutexposure to novel technologies and the novelmodes of reasoning and variation in cause–effectunderstandings that are associated with suchexposure, breakthrough solutions become increas-ingly unlikely. Thus, although the firm uses fa-miliar, well-understood technologies with greatcompetence, the absence of novelty and exper-imentation that are likely to help the firm craftbreakthrough solutions to upcoming problemslimits the firm’s capacity for breakthrough inven-tion.

Exploring novel technologies, i.e., technologiesthat are new to the organization, even thoughthey may have been in existence earlier, are animportant mechanism by which firms can avoidfamiliarity traps. Exploring realms of knowledgethat an organization has hitherto not exploredprovides the organization with multiple benefitsfrom the perspective of generating new, break-through solutions. First, it provides the organi-zation with the benefit of heterogeneity in itsproblem-solving arsenal (Amabile, 1988). Noveltechnologies may differ in their modes of reason-ing and problem formulation and solution.Exposure to these different approaches adds tothe repertoire that the organization can bring tobear on any new problem that it faces. Newlylearnt or observed perspectives may reflect better,more effective solutions to a given problem.Second, as new technologies are observed andstudied, the stability of existing cognitive struc-tures and cause–effect relationships is challenged(Lei et al., 1996). New world-views have to bedeveloped that account for both the known aswell as the unfamiliar, and this process can lead


to additional insights and profundity. Metaphor-ically speaking, the irritant of new, imperfectlyunderstood streams of knowledge can foster thepearls of insight that encompass both old and newknowledge. The enhanced repertoire and deeperunderstanding that are the consequence of explo-ration of new technologies can provide the basisfor breakthrough inventions.

Even though exposure to new technologies islikely to be beneficial up to a point, excessiveexploration of new technologies must eventuallybe harmful (Levinthal and March, 1993). In mod-erate quantities, the novelties that new technol-ogies draw attention to spark renewed exami-nation of causes and effects, improveunderstanding and insight, and lead to break-through inventions. In excess, the same noveltiescan become a source of confusion and infor-mation overload. As organizations are reduced to‘frenzies of experimentation’ performance suffers(Levinthal and March, 1993). Further, expendingresources on multiple new technologies si-multaneously may eventually imply diseconomiesof scale within the individual technologies. Thesearguments suggest that:

Hypothesis 1: A firm’s creation of break-through inventions is related to its explorationof novel technologies in a curvilinear(inverted-U shaped) manner.

The maturity trap and emerging technologies

Closely related but conceptually distinguishablefrom the tendency to favor familiar technologiesis the tendency to favor mature technologies.Mature technologies are technologies that havebeen in existence for some time and are relativelywell known and understood in the industry. Incontrast,emerging technologies are technologiesthat are new in chronological terms. They rep-resent the leading edge of technology and haveonly recently been developed.

Mature technologies are closely tied to theadvantages and characteristics of the establishedfirm. First, mature technologies are usually wellunderstood and offer greater reliability relative tomore recently developed and less testedapproaches. For the established firm, providingreliable performance to its constituencies is acritical element of its competitive repertoire(Hannan and Freeman, 1989). Second, mature


technologies are likely to have highly developedvalue networks and organizational and extra-organizational assets that are co-specialized withthese technologies (Christensen and Rosenbloom,1995). These co-specialized assets and networksmake subsequent innovations on these existingtechnologies easier, but may impede experimen-tation with nascent technologies that require dif-ferent sets of assets, inputs, and complements.Third, mature technologies being well known inthe industry offer the benefits of legitimacy; evenif new technologies hold the promise of superiorperformance, convincing customers to trustunproven technologies may be difficult andexpensive. For all these reasons, the establishedfirm may prefer to exercise its innovative effortsin well-developed, mature technologies whilechoosing to forgo more current alternatives. Thisfailure to explore emerging technologies may,however, lead the firm into amaturity trap. Theorganization’s assets and commitments favor thefurther development of mature technologies; how-ever, lack of exposure to immature technologiesmay reduce the likelihood of creating a break-through invention.

Experimenting with nascent oremergingtechnologies can be a mechanism by which firmscan increase their likelihood of creating a break-through invention and circumvent the maturity trap.Emerging technologies are likely to differ frommature technologies in terms of the nature oftechnical problems they pose as well as the possi-bilities of technical solutions that they present. Onboth these accounts they offer a greater potentialfor breakthrough invention. We examine these twomechanisms in some detail below.

Research suggests that the character of inven-tion and innovation changes across the life cycleof a technology (Abernathy and Utterback, 1978;Tushman and Anderson, 1986). In its earliestperiod a technology poses many significant prob-lems as its basic concepts are reduced to practice.Although these problems raise the uncertaintiesassociated with a technology, they also representsignificant opportunity for early entrants into thistechnology. Solution of the fundamental problemsof a technology can often be of a path-breakingcharacter (Dosi, 1988). In contrast, as a tech-nology matures, fewer major problems remainto be solved. Thus, the opportunity to makefundamental breakthroughs is higher in emerg-ing technologies.


The theory of recombinant invention providesa second argument for the increased likelihoodof breakthrough invention with an emerging tech-nology (Fleming, 1999; Utterback, 1994). Accord-ing to this theory inventions are very commonlythe result of combining or recombining existingelements of knowledge into new syntheses(Henderson and Clark, 1990; Kogut and Zander,1992; Tushman and Rosenkopf, 1992; Fleming1999). Every new technology that is inventedadds a new set of knowledge elements to theexisting universe of knowledge elements. Thesenew knowledge elements can themselves berecombined in various ways and thus serve asthe basis for further inventions (Fleming, 1999).

The recombination potential of any set ofknowledge elements is, however, finite in thatthere are only so many ways that existingelements of knowledge can be fruitfully recom-bined. As technologies mature, the likelihood thathigh-utility combinations of the technology’selements have not yet been tried or exploitedalready must eventually decline. Conversely,emerging technologies, technologies whose con-stituent elements are relatively new, offer signifi-cantly higher potential for breakthrough recombi-nations. Since the elements in these technologieshave been in existence for a relatively shortperiod, experimenting with such technologies canenrich the set of underexploited technologicalfactors or primitives available to the organizationand increase the potential for breakthrough inven-tions. Thus, from both the perspective of unsolvedproblems as well as the prospect of recombinatorysolutions, emerging technologies present a higherlikelihood of breakthrough invention.

Eventually, the logic of diminishing returnsmust apply to the exploration of emergingtechnologies too. Working on emerging technol-ogies is likely to demand more focus, attention,and resources. These technologies are likely to berelatively poorly understood given their recency.Further, even the infrastructure for research inthese technologies may be underdeveloped rela-tive to that of more mature technologies. Researchinputs and materials that are routinely availablefor older technologies may need to be developedafresh for these technologies. Further, the path-ways to successful innovation are more uncertainin such technologies (Sahal, 1985). Experimentingwith many emerging technologies at the sametime may fragment the organization’s efforts and


resources and may well undermine its efforts torecharge its invention potential. As with noveltechnologies, the pursuit of emerging technologiesis likely to be most rewarding when conductedin moderation. Without access to new and under-exploited technological elements, the organi-zation’s ability to create breakthrough recombi-nations may be affected. With excessiveexploration of emerging technologies the organi-zation’s focus and resources may be challenged.Accordingly, we propose:

Hypothesis 2: A firm’s creation of break-through inventions is related to its explorationof emerging technologies in a curvilinear(inverted-U shaped) manner.

The propinquity trap and pioneeringtechnologies

A third dynamic that is likely to characterize theproblem-solving behavior of established organi-zations is their propensity to search for solutionsin the neighborhood of existing solutions (Helfat,1994; Martin and Mitchell, 1998; Nelson andWinter, 1982; Stuart and Podolny, 1996).Attempting to solve technological problems is anenterprise fraught with uncertainty. In ambiguousand uncertain environments, reliance upon histori-cal experience is often the norm (March, 1988).Previously used solutions provide a base of fa-miliarity from which the problem solver can moveforward. Further, using elements or approachesthat are known to have succeeded in past searchesprovides some assurance to the problem solverthat the endeavor will not be a complete failure(Fleming, 1999). Thus, from a corporate inno-vation champion’s perspective an effort thatbuilds on technological antecedents is less riskythan one that attempts ade novosolution to aproblem (Hoskisson, Hitt, and Hill, 1993; Hoskis-son, Hitt, and Ireland, 1994). Similarly, adaptationof existing solutions to new problems conservescognitive effort and resources, both scarce inputs.

The impulse to build on existing foundationsis therefore likely to be strong in the context ofinventions in general. In the context of an estab-lished organization this impulse is likely to beheightened by the organization’s need to ensureorganizational order (Burgelman, 1983; Meziasand Glynn, 1993). In large corporations, organi-zational size and complexity demand that a struc-


tured approach be used to regulate organizationalactivity (Burgelman, 1983). Resource allocationmust follow established norms, controls, and pro-cedures (Hittet al., 1996). Projects that build onclearly specified antecedents are likely to be moreeasily justifiable before a rule-bound decision-maker than projects that rely on completelynew principles.

Although risk aversion, organizational routines,and bureaucratic convenience all mandate a pri-ority for projects that look for new solutions nearold solutions, these factors also predispose theorganization towards falling into thepropinquityor nearness trap. If the organization searchesextensively and almost universally for new so-lutions in the neighborhood of old solutions, thenlarge areas in the solution domain remain unex-plored. Yet, the history of science suggests thatmany remarkable inventions eventually emergefrom precisely these hitherto unexplored domains(Utterback, 1994). Actors from unrelated contexts,unfettered by the need to build on existing prec-edents, introduce new solutions or define prob-lems in new ways that facilitate completelyunprecedented and discontinuous solutions(Foster, 1986; Brown, 1991).

Experimentation withpioneering technologiesmay provide one mechanism for incumbents tocircumvent the dangers of the propinquity trapand preempt such attackers (Brown, 1991). Pio-neering technologies build on no existing technol-ogies. Instead of trying to modify an availablesolution, pioneering technologies focus on com-pletely de novosolutions. Indeed, the directive toresearchers from a pioneering technology perspec-tive is often to ignore all available solutions,focus instead on basic problems and their rootcauses, and step into the complete unknown insearch of a fundamental solution. Such a path isvery aptly captured in the following missive fromthe Director of the Xerox Palo Alto ResearchCenter (PARC) to an incoming employee(Brown, 1991):

Our approach to research is ‘radical’ in the senseconveyed by the word’s original Greek meaning:‘to the root’. At PARC we attempt to pose andanswer basic questions that can lead to fundamen-tal breakthroughs … If you come to work herethere will be no plotted path. The problems youwork on will be the ones that you help to invent.When you embark on a project, you will haveto be prepared to go in directions you couldn’thave predicted at the outset … That’s why fol-


lowing your instinct is so important. Only byhaving deep intuitions, being able to trust themand knowing how to run with them will you beable to keep your bearings and guide yourselfthrough uncharted territory. The ability to doresearch that gets to the root is what separatesmerely good researchers from world class ones.The former are reacting to a predictable future;the latter are enacting a qualitatively new one.

In more abstract form, the role of pioneeringstrategies in fostering radical inventions can beunderstood in terms of the research on technologi-cal progress and technological trajectories.Researchers have often described technologicalprogress as a series of continuous improvementsin fitness (a composite of all relevant performanceattributes of a technology) along a technologytrajectory, with occasional discontinuities thatemerge because of jumps to a different tech-nology trajectory (Dosi, 1988; Foster, 1986;Sahal, 1985). If we consider the fitness of atechnology as a function defined on a technologi-cal space, then technology trajectories can berepresented as the mapping of elements of thetechnology space onto several distinct, continuousfunctions. The individual distinct functions rep-resent different trajectories, and thus have discon-tinuous or very different ranges of fitness values,but within a trajectory there is continuity asfitness values of a given technology are closelyrelated to fitness values of proximate technol-ogies. In such a situation, solutions that build onexisting solutions are likely to map onto the sametechnological trajectory and consequently yieldvery similar fitness values to the ones alreadyobtained by other solutions. A pioneering tech-nology is an attempt to jump to a distinct tra-jectory in the hope that the range of fitness valuesembodied in the new trajectory is radically higher.Such an attempt is, of course, risky. There is noguarantee that the new trajectory will yield ahigher range of fitness values, and indeed mayyield extremely low ranges of fitness. However,increasing the number of experiments with suchpioneering technologies should eventually yield atleast some breakthrough inventions. Other thingsbeing equal, the larger the number of such pio-neering attempts the greater the likelihood that atleast some of them are successful.

The implications of increased experimentationwith pioneering technologies for breakthroughinventions are not as clear as those of experi-menting excessively with novel and emerging


technologies. Both novel and emerging technol-ogies directly challenge the cognitive capabilitiesand research resources of the organization. Forinstance, we noted earlier that excessive explo-ration of novel or emerging technologies canlead to information overload and diseconomies ofscale, and a consequent reduction in breakthroughinventions as the organization tries to developand integrate too many unfamiliar or nascentand underdeveloped streams of knowledge. Withrespect to pioneering technologies this effect isnot as clearly specified. On the one hand pio-neering technologies may imply an even greatercognitive task as the organization grapples withdeep and fundamental problems. On the otherhand pioneering technologies may result fromsimply ignoring conventional wisdom or fromusing broad-based creativity from people notdirectly involved in the field. For instance,Exxon’s ‘Wish’ program entailed the involvementof a large set of interested people who were notdirectly involved in the subject field to come upwith radical inventions (Berkowitz, 1996). Sincesuch pioneering efforts may involve only anunconventional approach to a problem, rather thancomplicated integration of disparate or novelknowledge bases, their costs may be reflected notin terms of cognitive overload but rather simplyin greater financial outlays or organizationalexpenditures. Thus, their negative consequencesmay show up in financial figures rather thanthrough a decline in the number of breakthroughinventions. A priori, this suggests two possibleoutcomes, the first consistent with a cognitiveoverload interpretation, while the second wouldsuggest that the negative consequences of exces-sive experimentation with pioneering technologiesmay be reflected on dimensions other than theoutput of breakthrough inventions:

Hypothesis 3a: A firm’s creation of break-through inventions is related to its explorationof pioneering technologies in a curvilinear(inverted-U shaped) manner.

Hypothesis 3b: A firm’s creation of break-through inventions is positively related to itsexploration ofpioneeringtechnologies.

It would be useful at this stage to clarify thedifferences between the three traps/strategiesidentified above. Although a firm can si-


multaneously fall into all three traps there arenevertheless important conceptual distinctionsbetween them. The familiarity trap arises onaccount of lack of variety in the firm’s conceptualrepertoire. The remedy is to introduce variety inthe form of unfamiliar or novel technologies. Incontrast, the maturity trap arises as a consequenceof the opportunity (or the lack of it) in thetechnology itself – the inventive potential of thetechnologies being used by the firm has beendiminished over time. The remedy is to be activein more leading-edge or emerging technologies.Finally, the propinquity trap arises as a functionof the search approach adopted—whether the firmattemptsde novosolutions or uses existing so-lutions as a starting point for defining andaddressing technical problems. It is thus a con-dition relating to theoriginality of the solutionapproach used. The remedy is to move away fromexisting solutions and explore the possibilities ofa radically different solution. This originalapproach could be applied to emerging or maturetechnologies, or familiar or novel ones.

Table 1 provides a set of illustrative cases offirms that fall into each of these traps. Althoughthe table highlights the cases of three pure types,in reality firms are likely to fall along a con-tinuum on each of these dimensions. Further,while there can be overlaps between novel,emerging, and pioneering technologies (forinstance, anemerging technology can also benovel to the firm), the constructs do not perfectlysubsume each other. They work through distinctmechanisms and the occurrence of one does notnecessarily imply the occurrence of the others.

RESEARCH DESIGN

Sample and data selection

Since longitudinal data on all inventions in anindustry are not generally available, in prior stud-ies scholars have been forced to examine onlythe cases of inventions that proved to be radical.The methodology of this research attempts toovercome this problem of sampling on the depen-dent variable and its attendant threats to internaland external validity (Berk, 1983). Specifically,we study a sample of firms irrespective ofwhether or not they have created breakthroughinventions. By obtaining measures of the strate-gies followed by them and incorporating a history


Table 1. The three traps

Firm description Familiarity Maturity trap Propinquitytrap trap

Firm works in a single technology, not exploring any Yes No Noothers. However, in that single technology it usually workson the leading edge and often uses a very original approachin terms of addressing the problems in that technology.

Firm explores several technologies but usually works on No Yes Nomature technologies. Within these mature technologies itsometimes adopts a very original approach to addressingthe problems in that technology.

Firm explores several technologies but usually works on No No Yesleading-edge technologies. Within these leading-edgetechnologies it usually adopts an unoriginal approach,preferring to work on problems and solutions that have wellestablished precedents.

of all their inventions, breakthrough or otherwise,we are able to present a relatively unbiased pic-ture of the association between firm strategiesand breakthrough inventions.

We tested the hypotheses on a longitudinaldata set on the patenting activities of the globalchemicals industry over the period 1980–95. Weused patent citation counts to identify break-through inventions. Several studies have shownthat patent citation counts are important indicatorsof the technical importance of an innovation(Albert et al., 1991; Narin, Noma, and Perry,1987). Further, highly cited patents represent criti-cal or path-breaking inventions (Trajtenberg,1990a, 1990b). We selected the chemicals indus-try as the setting for this research because patentsare widely regarded as a meaningful indicator ofinvention in this industry (Levinet al., 1987;Arundel and Kabla, 1998).

The data collection consisted of two phases. Inthe first phase we identified the most highly citedpatents in the chemicals industry for each yearbetween 1980 and 1989. For each successfulchemical patent application between these years,we computed the number of citations received bythe patent. Thereafter, for every year we sortedthe patents applied for in that year on the basisof their citation weights and identified the top 1percent of patents for that year as breakthroughinventions. This procedure ensures that each pa-tent is compared in its importance only to otherpatents of the same year. Since the duration for


which a patent is at risk of being cited variesfor patents of different vintages, it is importantto compare patents only with their own cohort. Asimilar procedure has been used in past research(Trajtenberg, 1990a, 1990b). A benefit of thisapproach is that breakthrough inventions are iden-tified from a universe of all inventions. Hence,sampling on the dependent variable is avoided.

In the second phase of the data collectionour task was to identify and obtain data on theestablished firms in the industry. To accomplishthis we consulted the leading trade journals(Chemical Weekand C&E News) that provideannual listings of the largest chemicals firms. Inthe lists published by these journals, subsidiarieswere often listed separately from parent firms.From an original sample of approximately 120firms, after including subsidiaries with parentfirms, a sample of 107 firms remained. For 10of these firms either patent data or covariate datacould not be reliably obtained and they weredropped from the analysis. For the remainingfirms in the sample we obtained yearly patentinghistories identifying each patent that they hadcreated over the study period. We then used thelist of breakthrough inventions in the chemicalsindustry to identify the breakthrough inventionscreated by these firms. Note that with thisapproach, since the sampling on firms is inde-pendent of whether or not they created break-through inventions, there are many firms in thesample that do not create any breakthrough inven-


tions in a given year. This approach now enablesus to have a panel data set with each firm’scomplete history of breakthrough inventionsacross the study period.

Patents applied for in later years have a smallerwindow for getting cited. Although our procedurefor counting citations only compares patents withother patents applied for in the same year, andthus keeps the ‘at risk of being cited’ periodconsistent for all compared patents, and we useyear dummy variables as suggested in the litera-ture (Trajtenberg, 1990a, 1990b), there is still theconcern that the citation patterns for the mostrecent patents might not be representative of theirfinal worth given the shorter period they wereavailable for being cited. Accordingly, we omittedall observations for years after 1989. Constructionof some of the independent variables entailedlags too, and the final panel used for regressionanalysis covers 8 years. The panel is unbalancedas some of the firms were acquired by other firmsor restructured in a fashion that made comparisondifficult beyond a particular year. Even thoughthe sample was focused on the largest firms inthe chemicals industry the inclusion of 97 firmsprovides significant depth to the sample andensures that there is considerable variety on allkey variables. For instance, the number ofemployees for firms in the sample varies from aminimum of 1100 to a maximum of more than181,000. Financial figures and personnel data onthese firms were obtained from Compustat,Worldscope, Japan Company Handbooks, DaiwaInstitute Research Guides, and trade publicationsand company annual reports. For all firms, finan-cial data were converted to constant (1985) U.S.dollars to ensure standardization within the sam-ple. A full list of the sample firms is availablefrom the authors.

We used U.S. patent data for all firms, includ-ing the foreign firms in the sample. This wasnecessary to maintain consistency, reliability, andcomparability, as patenting systems across nationsdiffer in their application of standards, system ofgranting patents, and value of protection granted.The United States represents one of the largestmarkets for chemicals, and firms desirous of com-mercializing their inventions typically patent inthe United States. Prior research using patent dataon international samples (e.g., Stuart and Podolny,1996; Patel and Pavitt, 1997), including studiesof the global chemicals industry (Achilladeliset


al., 1990), have followed a similar strategy ofusing U.S. patent data for international firms.

Variable definitions

Dependent variable

Breakthrough inventions.This was computed asthe number of breakthrough patents in the chemi-cal industry in any year that were created by thefocal firm. As noted earlier, for every year wesorted the patents applied for in that year on thebasis of their citation weights and identified thetop 1 percent of patents for that year as break-through patents. Note that the dependent variableis computed based on citations in patents that areissued after the breakthrough invention. Thesepatents are therefore different from the patentson which the independent variables are based.The independent variables, described below, arebased on the focal firm’s patenting history in theperiod before the breakthrough invention. Thus,even though both the dependent and some of theindependent variables are based on patent data,the actual patents on which they are based are dif-ferent.

Our choice of identifying the top 1 percent ofpatents (based on the number of citations receivedby the patents) as breakthrough inventions wasbased on the following rationale. Prior researchsuggests that (a) the most heavily cited patentsare the most valuable (Trajtenberg, 1990b), and(b) the value distribution of patents is very highlyskewed, a few patents are very valuable, whilemost patents have relatively low values (Griliches,1990; Harhoff et al., 1999). Studies usingcitations suggest that the distribution of citationsto a patent drops off pretty sharply, but does notprovide an indication of an exact cut-off pointthat can be used to classify truly breakthroughinventions (Trajtenberg, 1990b). Therefore, weexamined the actual pattern of citations receivedby the top 1 percent, top 2 percent, and top 5percent of patents. These indicated a fairly sharpdrop-off in average number of citations receivedbetween these three categories of patents. Forinstance, of all chemicals patents applied for in1981, the top 1 percent received 37 citations onaverage, the top 2 percent received 30 citationson average, while the top 5 percent received 22citations on average. Since our interest is inidentifying truly path-breaking inventions we


decided to use the top 1 percent as our indicatorsof breakthrough inventions. However, for sensi-tivity we also repeated the analyses using the top2 percent (as we report in the Results section,the findings are substantively the same witheither measure).

Independent variables

Novel technologies. For this variable we neededto develop a measure that taps into the degree towhich a firm experiments with technologies thatit has not used previously. We computed thisvariable using the technology classification pro-vided by the U.S. patent system. The U.S. patentsystem classifies the technology domain into 400broad classes and several hundred thousand sub-classes nested within the classes. Based on afirm’s prior patenting history we computed thisvariable as the number of new technology classesthat were entered by a firm in the previous 3years. A firm was considered to have entered anew technology class when it first applies for apatent in a class in which it had not patented inthe previous 4 years. The presumption is that ifa firm has not patented in a technology in theprevious 4 years, then that technology representsan unfamiliar technology for the firm. For robust-ness, we also computed this measure using a 5-year interval instead of a 4-year interval. Sincetechnical knowledge tends to depreciate or obso-lesce over time, not participating in a technologyfor an extended period of time is likely to sig-nificantly reduce a firm’s stock of viable knowl-edge in that technology. Prior research in tech-nology-intensive industries has used a 4- to 5-year window as the appropriate time frame forassessing the validity of a knowledge base in agiven technology (Stuart and Podolny, 1996;Ahuja, 2000). The choice of a 4- to 5-year periodfor knowledge relevance is also consistent withstudies of R&D depreciation (Griliches, 1984).

Emerging technologies.For this variable weneeded a measure that would capture the degreeto which a firm experiments with leading-edgeor nascent technologies. We based this measureon the average age of the patents cited by a firm.Every patent is required by law to disclose allprior art, the previous patents that served as thefoundation for the current patent. If a firm isworking primarily on old technologies then the


average age of the patents it cites is likely to behigh. On the other hand, if a firm cites veryrecent patents then it can be said to be workingon very current technologies. We computed thevariable as the number of a firm’s patents thatcite technology that is on average less than 3years old. As an alternate measure (forrobustness), we also computed this variable asthe number of a firm’s patents that cite tech-nology that is on average less than 2 years old.The choice of a 2- to 3-year time period tosignify an emerging technology was related tothe currency of knowledge issue raised in thecontext of the previous variable. Since priorresearch has used a 4- to 5-year period as onerepresenting the most viable life of a technology,the earlier part of this period would be appropri-ate for measuring emerging technologies. Hence,we use 2- and 3-year cut-offs.

Pioneering technologies.For this variable weneeded a measure that would capture the degreeto which a firm experiments with technologiesthat build on no prior technologies. We computedthis variable as the number of a firm’s patentsthat cite no other patents. As noted earlier, patentsmust indicate their prior technological lineage byciting all patents that they build on (Podolny andStuart, 1995; Stuart and Podolny, 1996; Jaffe,Trajtenberg, and Henderson, 1993; Trajtenberg,Henderson, and Jaffe, 1992). Patents that cite noother patents indicate that they have no discern-ible technological antecedents. Past research hasused the relative lack of prior art citations in apatent as an indicator of theoriginality and crea-tivity of that patent (Trajtenberget al., 1992).Accordingly, we suggest that the creation of manysuch patents by a firm reflects its willingness toadopt a pioneering or unprecedented approach inits innovation strategy. Thus, firms that createmany patents that cite no other patents are firmsthat can be regarded as willing to explore tech-nology spaces that have not been explored before.

Control variables. Prior research suggests thatthe incentives of a firm to introduce breakthroughinventions may vary with its profitability(Henderson, 1993). Accordingly, we included thevariable Net Income as a control variable. Wealso included several other control variablesincluding R&D expenditures (R&D), firm size asmeasured by natural log of number of employees


(Logemployees), and firm diversification(Diversification/Entropy) as calculated using theentropy measure (Palepu, 1985). In all modelswe included the unobserved heterogeneity controlvariable Prior Breakthrough Inventions (the sumof breakthrough inventions created by a firm inthe 3 years prior to the firm’s entry into thesample) and dummy variables for firm nationality(U.S. and Japanese–European being the basecategory) and calendar year. Finally, it is possiblethat the frequency of patenting or citations orbreakthrough inventions varies across the UnitedStates Patent Office (USPTO) technology classes(Trajtenberg et al., 1992). To account for thepossibility of technology class effects we createda set of 80 dummy variables to reflect the 81classes that cover the chemicals sector. The mostcommonly occurring class (Class 428) was treatedas the omitted category. For each observationthese dummy variables reflect a firm’s partici-pation or nonparticipation in that particular tech-nology class in that year.

Model specification and econometric issues

The dependent variable of the study, Break-through Inventions, is a count variable and takesonly nonnegative integer values. A Poissonregression approach is appropriate for such data(Hausman, Hall, and Griliches, 1984; Hendersonand Cockburn, 1996). Accordingly, we specifiedthe following Poisson regression model:

Pit = eXit-1γ+Ait-1β (1)

where Pit is the number of breakthrough inven-tions obtained by firmi in year t, Xit-1 is a vectorof control variables affectingPit, and Ait-1 is avector of variables representing the hypothe-sized effects.

The above specification does not account forunobserved heterogeneity, the possibility thatobservationally equivalent firms may differ onunmeasured characteristics. For instance, firmsmay enter the sample with inherently differentbreakthrough invention-generating capabilities(Ahuja and Katila, 2001). To address this possi-bility we used the Presample Panel Poissonapproach (Blundell, Griffith, and Van Reenen,1995) and included the Presample variabledescribed earlier, Prior Breakthrough Inventions,as a measure of the unobserved differences


between the sample firms in their ability to createbreakthrough inventions.

The presample approach presumes that theinfluences valid in the presample period continueto be valid in the study period. To ensure thatour results were robust to this assumption we didtwo things. First, as an alternate measure we usedlagged values of the dependent variable as analternate measure of unobserved heterogeneity.Second, in addition to using conventional Poissonestimation of the Presample Panel Poisson model(Blundell et al., 1995) we also estimated themodels using the GEE (General EstimatingEquation) approach for modeling longitudinalPoisson data (Liang and Zeger, 1986). Sinceunobserved heterogeneity that is affecting thedependent variable should be reflected in corre-lation between the residuals of the same firm, usingan approach that models serial correlation into theestimation procedure accounts for any remainingcorrelation. Finally, we report all results with‘robust’ or empirical standard errors (SAS, 1997).In case of model misspecification or overdispersion,the model-based standard errors for a Poissonregression can be incorrect. Using robust standarderrors guards against this possibility.

RESULTS

Table 2 provides descriptive statistics and corre-lations for all variables. Table 3 presents theresults of the hypothesis testing. We originallyattempted to run the GEE regressions with thefull set of 80 USPTO class dummies. However,the full models with 80 class dummies and 21other covariates proved to be nonestimable usingGEE methodology. Accordingly, we first esti-mated a series of regular Presample Poisson mod-els with all 80 dummies. From these estimationswe identified all USPTO classes that indicated asignificant class effect. To be conservative weincluded all classes that were significant even atp < 0.10. Then, we estimated the GEE modelswith this smaller subset of 38 class dummies,treating the remaining classes as a single class.The results of these estimations are reported inthe GEE Models 1–8 in Table 3.

Model 1 in Table 3 presents the results for thecontrol variables. Model 2 adds the variables forthe three hypothesized effects, with no squaredterms. Model 3 adds all three squared terms for

En

trep

ren

eu

rship

inth

eL

arg

eC

orp

ora

tion

535

Table 2. Descriptive statistics and zero-order correlations

Variable Mean S.D. Min. Max. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

1 Breakthrough 0.42 1.11 0 9 1.00Inventions

2 Novel Technologiesit-1 6.89 3.59 0 27 0.03 1.003 Emerging 7.39 11.64 0 79 0.54 0.05 1.00

Technologiesit-14 Pioneering 1.64 3.44 0 24 0.39 0.02 0.82 1.00

Technologiesit-15 R&Dit-1 86.53 166.36 0.10 1081.2 0.60 0.04 0.82 0.72 1.006 Firm Size 2.40 1.22 0.09 5.20 0.44 0.12 0.60 0.53 0.66 1.00

(LogEmployees)it-17 Net Incomeit-1 120.12 254.39−647 2312.4 0.64 0.10 0.58 0.48 0.81 0.59 1.008 Diversification- 1.30 0.34 0.20 2.19 0.23 0.05 0.40 0.40 0.41 0.46 0.30 1.00

Entropyit-19 Prior Breakthrough 1.32 2.54 0 16 0.53 0.02 0.64 0.49 0.62 0.53 0.53 0.24 1.00

Inventionsit-110 U.S.A. 0.26 0.44 0 1 0.26 0.04 0.06−0.04 0.12 0.26 0.24 −0.12 0.24 1.0011 Japan 0.43 0.50 0 1 −0.19 −0.03 −0.20 −0.19 −0.32 −0.68 −0.32 −0.14 −0.21 −0.52 1.0012 Year 1982 0.13 0.33 0 1 0.01−0.06 −0.01 −0.02 −0.05 0.00 −0.12 0.02 0.00 −0.01 −0.00 1.0013 Year 1983 0.13 0.33 0 1 −0.02 −0.02 −0.04 −0.02 −0.04 0.00 −0.07 0.01 0.00 0.00 −0.00 −0.14 1.0014 Year 1984 0.13 0.33 0 1 0.01−0.03 −0.03 −0.01 −0.04 −0.01 −0.01 −0.01 −0.00 0.01 −0.01 −0.15 −0.15 1.0015 Year 1985 0.13 0.34 0 1 −0.02 −0.01 −0.02 −0.01 −0.02 0.01 −0.08 −0.00 −0.00 0.00 −0.01 −0.15 −0.15 −0.15 1.0016 Year 1986 0.13 0.33 0 1 0.00 0.01−0.02 0.02 0.03 0.00 0.01 0.00−0.01 −0.00 −0.01 −0.15 −0.15 −0.15 −0.15 1.0017 Year 1987 0.12 0.33 0 1 −0.01 0.08 0.03 0.00 0.07 −0.01 0.10 −0.01 −0.00 −0.00 0.00 −0.14 −0.14 −0.14 −0.14 −0.14 1.0018 Year 1988 0.12 0.33 0 1 0.04 0.10 0.06 0.07 0.10−0.01 0.18 −0.02 0.00 −0.00 0.00 −0.14 −0.14 −0.14 −0.14 −0.14 −0.14 1.00

N = 721 observationsAll correlations with magnitude >|0.07| are significant at the 0.05 level

Copyright

2001John

Wiley

&S

ons,Ltd.

Stra

t.M

gm

t.J.,22:

521–543(2001)

536G

.A

hu

jaa

nd

C.

M.

La

mp

ert

Table 3. GEE/Poisson regressions of the impact of novel technologies, emerging technologies and pioneering technologies on breakthrough inventions

Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10GEE GEE GEE GEE GEE GEE GEE GEE (2%) Poisson Poisson

(2%)

Constant −4.7787∗∗∗ −4.8255∗∗∗ −5.1820∗∗∗ 4.7206∗∗∗ −5.2875∗∗∗ −5.1613∗∗∗ −5.0607∗∗∗ −3.3905∗∗∗ −5.7640∗∗∗ −3.7188∗∗∗[0.6869] [0.7129] [0.7109] [0.7133] [0.7092] [0.7110] [0.7159] (0.5247] [0.6299] [0.4808]

Novel Technologiesit-1 0.0059 0.2074∗∗∗ 0.0147 0.1993∗∗∗ 0.2057∗∗∗ 0.2053∗∗∗ 0.1603∗∗∗ 0.1974∗∗ 0.1642∗∗

[0.0216] [0.0618] [0.0208] [0.0633] [0.0609] [0.0620] [0.0417] [0.0694] [0.0535]Novel Technologies Squaredit-1 −0.0105∗∗∗ −0.0104∗∗∗ −0.0105∗∗∗ −0.0103∗∗∗ −0.0074∗∗∗ −0.0104∗∗ −0.0075∗∗

[0.0032] [0.0033] [0.0032] [0.0032] [0.0020] [0.0038] [0.0028]Emerging Technologiesit-1 0.0078 0.0440∗ 0.0399+ 0.0106 0.0465∗ 0.0447∗ 0.0582∗∗∗ 0.0358+ 0.0456∗

[0.0086] [0.0240] [0.0247] [0.0084] [0.0238] [0.0241] [0.0168] [0.0232] [0.0184]Emerging Technologies Squaredit-1 −0.0005∗ −0.0004+ −0.0005∗ −0.0005∗ −0.0006∗∗ −0.0004+ −0.0005∗

[0.0003] [0.0003] [0.0003] [0.0003] [0.0002] [0.0003] [0.0002]Pioneering Technologiesit-1 0.0279∗ 0.0659∗ 0.0586∗ 0.0811∗ 0.0425∗∗ 0.0399∗∗ 0.0403∗∗ 0.0430∗ 0.0423∗

[0.0151] [0.0327] [0.0333] [0.0353] [0.0144] [0.0139] [0.0141] [0.0222] [0.0180]Pioneering Technologies Squaredit-1 −0.0012 −0.0012 −0.0024+

[0.0016] [0.0015] [0.0017]R&Dit-1 −0.0000 −0.0005 −0.0003 −0.0003 −0.0004 −0.0003 −0.0005 −0.0004 0.0000 −0.0002

[0.0006] [0.0007] [0.0007] [0.0007] [0.0007] [0.0007] [0.0007] [0.0004] [0.0008] [0.0007]Firm Size (LogEmployees)it-1 0.3145∗ 0.2717+ 0.1583 0.2190 0.2173 0.1521 0.1081 0.1768+ 0.2200+ 0.1573

[0.1410] [0.1395] [0.1372] [0.1453] [0.1342] [0.1377] [0.1432] [0.0990] [0.1467] [0.1179]Net Incomeit-1 0.0010∗∗ 0.0012∗∗ 0.0012∗∗ 0.0011∗∗ 0.0012∗∗∗ 0.0012∗∗∗ 0.0012∗∗∗ 0.0007∗∗∗ 0.0013∗∗∗ 0.0008∗∗∗

[0.0003] [0.0004] [0.0004] [0.0004] [0.0004] [0.0003] [0.0003] [0.0002] [0.0003] [0.0002]Diversification-Entropyit-1 −0.0776 −0.0945 −0.1134 −0.0911 −0.1230 −0.1136 −0.0330 −0.4857∗ −0.0153 −0.4419∗

[0.3180] [0.3114] [0.3042] [0.3155] [0.3007] [0.3049] [0.3056] [0.2309] [0.2759] [0.2054]Prior Breakthrough Inventionsit-1 −0.0259 −0.0286 0.0313 −0.0293 −0.0291 −0.0319 0.0086 −0.0362 −0.0013

[0.0224] [0.0213] [0.0206] [0.0207] [0.0205] [0.0206] [0.0106] [0.0345] [0.0167]

Copyright

2001John

Wiley

&S

ons,Ltd.

Stra

t.M

gm

t.J.,22:

521–543(2001)

En

trep

ren

eu

rship

inth

eL

arg

eC

orp

ora

tion

537

Table 3. Continued

Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10GEE GEE GEE GEE GEE GEE GEE GEE (2%) Poisson Poisson

(2%)

Lagged Breakthrough Inventions 0.0335[0.0332]

U.S.A. 1.1246∗∗∗ 1.2424∗∗∗ 1.2515∗∗∗ 1.2773∗∗∗ 1.2182∗∗∗ 1.2450∗∗∗ 1.1409∗∗∗ 1.0461∗∗∗ 1.5072∗∗∗ 1.1267∗∗∗

[0.2172] [0.2200] [0.2041] [0.2050] [0.2084] [0.2067] [0.1979] [0.1723] [0.2593] [0.1982]Japan 1.1954∗∗ 1.2207∗∗ 0.9629∗∗ 1.1386∗∗ 1.0429∗∗ 0.9547∗∗ 0.8076∗ 0.8276∗∗∗ 1.3003∗∗∗ 0.9166∗∗∗

[0.3793] [0.3888] [0.3590] [0.3805] [0.3649] [0.3627] [0.3534] [0.2370] [0.3575] [0.2774]Year 1982 0.3778 0.4729+ 0.4616+ 0.4567 0.4867+ 0.4580+ 0.4727 0.0954 0.4087+ 0.1328

[0.2564] [0.2797] [0.2801] [0.2808] [0.2818] [0.2780] [0.2940] [0.1587] [0.2405] [0.2014]Year 1983 −0.0014 0.0635 −0.0033 0.0081 0.0209 0.0085 −0.0122 −0.0339 0.1127 0.0640

[0.2742] [0.3029] [0.3007] [0.3025] [0.3049] [0.2982] [0.3134] [0.1871] [0.2724] [0.2188]Year 1984 0.3036 0.3602 0.3071∗ 0.3459 0.2935 0.3175 0.3221 0.3174+ 0.3255 0.4223∗

[0.2785] [0.2951] [0.2790] [0.2923] [0.2812] [0.2801] [0.2781] [0.1691] [0.2399] [0.1926]Year 1985 0.0589 0.1380 0.1093 0.1354 0.1023 0.1138 0.1330 0.0336−0.0590 −0.0777

[0.2963] [0.3236] [0.3142] [0.3225] [0.3162] [0.3134] [0.3257] [0.1901] [0.2549] [0.2071]Year 1986 −0.2935 −0.2289 −0.3095 −0.2549 −0.3158 −0.2950 −0.2288 −0.0635 −0.2646 −0.0581

[0.2842] [0.2988] [0.2878] [0.3009] [0.2906] [0.2857] [0.2915] [0.1867] [0.2664] [0.2098]Year 1987 −0.5115 −0.4397 −0.5562+ −0.4568 −0.5636+ −0.5449 −0.5197 −0.1925 −0.4703+ −0.0800

[0.3270] [0.3394] [0.3330] [0.3430] [0.3300] [0.3330] [0.3637] [0.1896] [0.2818] [0.2154]Year 1988 −0.5332 −0.4968 −0.6499 −0.4935 −0.6607∗ −0.6451∗ −0.5687+ −0.3316 −0.7269∗ −0.3818+

[0.3296] [0.3400] [0.3262∗] [0.3426] [0.3269] [0.3260] [0.3280] [0.2094] [0.2941] [0.2302]

USPTO technology class dummies 38 38 38 38 38 38 38 38 80 80dummies dummies dummies dummies dummies dummies dummies dummies dummies dummies

Scale 0.7492 0.7476 0.7379 0.7461 0.7395 0.7375 0.7383 0.9161 0.7366 0.9187

N 721 721 721 721 721 721 721 721 721 721Deviance 374.98 371.72 360.42 369.04 362.56 360.62 361.36 556.39 336.94 524.16

+p < 0.10; ∗p < 0.05; ∗∗p


the three hypothesized effects to account for thepossibility of curvilinearity. Model 3 indicatesthat the coefficients for Novel Technologies,Novel Technologies Squared, Emerging Technol-ogies and Emerging Technologies Squared, andPioneering Technologies are statistically signifi-cant with the predicted sign. However, the coef-ficient for Pioneering Technologies Squared,while negative, is not statistically significant. Wealso ran a series of models (Models 4, 5, and 6)omitting the squared terms for each of the threehypothesized effects, one at a time. The deviancestatistics, which indicate overall model fit, con-firm the same results as they indicate that omit-ting the square term for either Novel or EmergingTechnologies leads to a worsening of model fit(although for Emerging Technologies the worsen-ing was only marginally statistically significant,p< 0.10), but omitting the Pioneering TechnologiesSquared term does not worsen the model fit.Thus, Model 6 that omits the Pioneering Technol-ogies Squared term is the best-fitting specification.

Model 6 indicates some support for all threehypotheses. In Hypothesis 1 we had predicted thatexperimenting with Novel Technologies shouldincrease the likelihood of breakthrough inventionsup to a point and then lead to a diminution (aninverted U). The coefficient of Novel Technol-ogies is positive and statistically significant, whilethe coefficient of its squared term is negative andstatistically significant, indicating support for theprediction. In Hypothesis 2 we had predictedthat experimenting with Emerging Technologiesshould increase the likelihood of breakthroughinventions up to a point and then lead to adiminution (an inverted U). The coefficient ofEmerging Technologies is positive and sta-tistically significant, while the coefficient of itssquared term is negative and statistically signifi-cant, again indicating support for Hypothesis 2.The estimated coefficients for these variables indi-cate that the turning point of the curve lies wellwithin the observed range of data for both thesevariables (for Novel Technologies the point ofinflection is at 0.2057/(2∗0.0105) = 10, wherethe observed range is 0–27, and for EmergingTechnologies it occurs at 0.0465/(2∗0.0005) =46.5, where the observed range is 0–79), thusfurther confirming the downward component ofthe curve. In the case of Hypothesis 3 we haddeveloped two competing predictions. Hypothesis3a predicted an inverted U relationship between


Pioneering Technologies and Breakthrough Inven-tions, while Hypothesis 3b predicted a positiveeffect of Pioneering Technologies on Break-through Inventions. The coefficient of the Pio-neering Technologies variable was positive andstatistically significant, but the coefficient for itssquared term was not statistically significant, andindeed was dropped from the final model. Thus,the results do not support Hypothesis 3a, but dosupport the competing hypothesis, Hypothesis 3b.It appears that Pioneering Technologies have apositive impact on breakthrough inventions butthe diminishing returns to experimentation evidentwith the strategies of Novel and Emerging Tech-nologies are not statistically visible with Pio-neering Technologies. We explore this issue inthe Discussion section.

Since the Emerging and Pioneering Technol-ogies variables were highly correlated (0.82) adanger of multicollinearity arises. In general, thesymptoms of multicollinearity include (a) verylarge standard errors for the affected variablesand therefore even true effects show up as nonsig-nificant or (b) extreme sensitivity of results suchthat coefficients flip signs after even minorchanges in the specification or sample size andomitting even a few observations affects theresults materially (Greene, 1997). In the contextof the reported results, neither of these symptomswas observed. Indeed, the results are very robustand stable across many changes in specification,sample, dependent variables, independent vari-ables, and estimation method. We reran the analy-ses with a one-period lagged value of the depen-dent variable Breakthrough Inventions as aregressor in place of the presample variable PriorBreakthrough Inventions (Model 7). We also rees-timated the models after defining our dependentvariable Breakthrough Inventions as including allpatents in the top 2 percent of cited patents(rather than top 1%) (Model 8). Finally, we esti-mated the models using regular Poisson maximumlikelihood estimation instead of the GEE approachand included all 80 class dummies, for both thetop 1 percent and the top 2 percent definitionsof the dependent variable (Models 9 and 10respectively). The results are again consistent withthe overall pattern though the Emerging Technol-ogies coefficients in Model 9 are only marginallystatistically significant (p < 0.07 andp < 0.06,for the variable and its squared term,respectively). Among other sensitivity tests, not


presented here but for which results are availabledirectly from the authors, we used alternate defi-nitions of Novel Technologies (we recomputedthe variable after defining novel technologies astechnologies in which the firm had not patentedin, in the previous 5 years instead of 4) andEmerging Technologies (we recomputed the vari-able as the number of patents citing technologyless than 2 years old instead of 3 years old).Finally, as another alternate estimation approachwe aggregated all 34 patent classes that wereinvolved in less than 10 percent of the obser-vations into a single class and estimated a GEEmodel using the resulting 46 class dummies. Theresults were robust to these and many other speci-fications of these models.

DISCUSSION AND CONCLUSIONS

Researchers have suggested that the pursuit ofcorporate entrepreneurship requires establishedcompanies to strike a delicate balance betweenengaging in activities that use what they alreadyknow, while at the same time challenging them-selves to embark upon new activities and oppor-tunities to rejuvenate themselves (Floyd andWooldridge, 1999; Hannan and Freeman, 1989;Huff, Huff, and Thomas, 1992). Leonard-Bartonhas aptly termed this conflict as a ‘capability–rigidity paradox, where existing capabilities pro-vide the basis for a firm’s current competitiveposition, [but] without renewal, these same capa-bilities become rigidities constraining the firm’sfuture ability to compete’ (Leonard-Barton, 1992).Interest in resolving this apparent paradox hasled researchers to examine the processes by whichcorporations have attempted to ‘redefine, renewand remake themselves’ (Covin and Slevin, 1991;Guth and Ginsberg, 1990; Zahra, Jennings, andKuratko, 1999). In this study we explored thiscapability–rigidity paradox in the context of onemechanism through which established firms caninitiate the process of renewal: breakthroughinventions. After identifying three organizationalpathologies that could hinder the creation ofbreakthrough inventions in established corpo-rations we suggested three strategies that couldhelp firms to counter these problems. Our empiri-cal results provided support for our argumentsthat experimenting with novel, emerging, and pio-neering technologies may be ways for organi-


zations to overcome the traps of familiarity,maturity, and propinquity. Specifically, we foundthat exploration of novel and emerging technol-ogies is curvilinearly associated with subsequentbreakthrough inventions, first increasing and thendecreasing a firm’s likelihood of creating a break-through invention. However, the downward slop-ing part of the curve was not identifiable in thecase of pioneering technologies.

Two possibilities could be consistent with thisnonfinding. First, as we argued in the hypothesisdevelopment for Hypothesis 3b, it could be thatPioneering Technologies differ in their cognitivedemands and likelihood of information overloadfrom both Novel Technologies and EmergingTechnologies. Entry into Novel and EmergingTechnologies entails study of new areas from thefirm’s perspective and the absorption of a body ofknowledge already in existence or being created.Pioneering Technologies on the other hand canoften imply an attempt to move away from exist-ing bodies of knowledge and give rein to creativesolutions. Thus, it is not so much the possibilityof information overload that is problematic butsimply that the attempted solution may yield noresults. In such a circumstance, although excess-ive experimentation with pioneering technologieswill have significant costs, these costs may not beeventually reflected in a decline in breakthroughinvention; instead they would appear as largerresource outlays or monetary costs. Anotherexplanation for this finding could be that notenough firms reached the level of experimentationon this variable that was sufficiently high forthe negative effects of resource fragmentation tobecome statistically significant. These two expla-nations are not mutually exclusive; however,future research may help to further clarify thismatter. We now briefly touch upon the impli-cations of our study for theory, research, andpractice.

Implications for theory, research, andpractice

Theoretically, the arguments and conclusions ofthis study make contributions to the literatureon entrepreneurship, strategy, and organizationallearning. From the perspective of organizationallearning this study’s identification of three trapsthat hinder breakthrough invention in the largecorporation is important in and of itself; however,


it is also important in terms of drawing attentionto the fact that the constraints to breakthroughinvention in the established firm do not necessar-ily stem from some dysfunctional traits of theorganization, but emerge as a very natural conse-quence of some fairly functional traits. As ourmodel argues, emphasizing the familiar andmature, and building on existing developments,are all efficient responses from the standpoint ofthe large firm. They may, however, not be effec-tive when the outcome variable of interest isbreakthrough invention. Characterizing the prob-lem in this fashion is useful when compared witha potential alternative characterization: that largecorporations fail at breakthrough inventionsbecause they are inept or incompetent. There isno antidote to incompetence. However, thedynamics we identify can be productively arrestedas our identification of three countering strate-gies demonstrates.

In drawing the relationship between rationalstrategies and unintended negative consequenceswe follow the lead of organizational theoriststhat have argued for more refined analyses oforganizational processes accounting for both first-order and second-order effects (March, 1991;Levinthal and March, 1993). However, we alsotake their work a few steps further. Prior theoriz-ing suggests that firms often get caught inlearn-ing traps (Levinthal and March, 1993). Here webuild on the learning trap arguments in threeways. First, we suggest that in the context ofbreakthrough inventions learning traps are mani-fested in three types of organizational pathologies:traps of familiarity, traps of maturity, and trapsof propinquity. Factoring learning traps in thisfashion permits us to identify specific strategiesthat organizations can use to counter these patho-logies. Second, we extend the learning trap argu-ments by including the possibility of second-orderor double-loop learning (Argyris, 1983; Huber,1991). Double-loop learning requires the reexam-ination and change of the governing values ofthe organization from the perspective of long-range outcomes (Argyris, 1983; Huber, 1991). Iffirms regard breakthrough inventions as importantto their future, organizational routines that fail toproduce such inventions are likely to be thesubject of reappraisal and reformulation (Huber,1991). Thus, by incorporating the possibility thatfirms will counteract such pathologies we deepenand condition the learning trap arguments from a


conceptual standpoint. Finally, we provide empiri-cal evidence in support of our arguments andthus contribute to filling what has been identifiedas a significant gap in the organizational learningliterature – the absence of systematic large-sampleempirical studies to supplement the rich butpotentially idiosyncratic case studies (Huber,1991).

This study found that firms varied in their useof entrepreneurial strategies and that using theseentrepreneurial strategies led to superior inventionperformance. However, the variation in firms’strategies on the three exploration variables andthe identification of a statistically discernibleimpact of these variations on the firm’s outputof breakthrough inventions raises a natural ques-tion: why do firms vary in their adoption ofentrepreneurial strategies? Why do some firmspursue novel, emerging, and pioneering technol-ogies more than others? Although this variationacross the entrepreneurial behavior of firms hasbeen noted earlier (Lant and Mezias, 1990;Mezias and Glynn, 1993), less research explainswhy this variation occurs.

We believe that our work, when contrastedwith prior contributions in the entrepreneurshipliterature, draws attention to one possible expla-nation of this variation that could be addressedin future research: the existence of a ‘virtuouscircle of corporate entrepreneurship.’ The virtuouscircle argument would suggest that although manyfirms would like to pursue such strategies theyare unable to do so. This inability stems fromtheir exclusion from a virtuous cycle in which(a) the pursuit of novel, emerging, and pioneeringtechnologies leads to breakthrough inventions, (b)breakthrough inventions when they occur, createwealth and surplus resources, and (c) these sur-plus resources fund the next cycle of entrepre-neurial experimentation, which in turn leads tomore breakthrough inventions. Although the fullinvestigation of this virtuous circle goes beyondthe scope of this study, prior research and thetheory developed in this study provide at leastsome evidence in support of this explanation.This study’s results provide direct support for thefirst leg of this cycle.

However, the study also provides some indi-cations in support of the second leg. For instance,the descriptions of the three forms of explorationidentified here suggest that the pursuit of novel,emerging, and pioneering technologies is likely


to require considerable slack resources. To go offin search of the unknown (a.k.a. pioneering) orexperiment with novel technologies is likely todemand extensive resources. Without slack, thesestrategies may be attractive but beyond reach.Indeed, both prior theory (Burgelman, 1983) andanecdotal evidence suggest that cash-rich corpo-rations can far more easily afford certain kindsof speculative and experimental ventures (Styros,1997). Similarly, in support of the third leg ofthe cycle, prior research on the wealth-creatingimpact of breakthrough inventions suggests thattechnologically important inventions can generatevery significant returns (Hall, Jaffe, and Trajten-berg, 1998; Harhoffet al., 1999).

entrepreneurship in the large corporation: a longitudinal …...entrepreneurship in the large...

Documents