cockburn, henderson and stern 2000 smj - untangling competitive advantage

Strategic Management JournalStrat. Mgmt. J.,21: 1123–1145 (2000)

UNTANGLING THE ORIGINS OF COMPETITIVEADVANTAGE

IAIN M. COCKBURN1, REBECCA M. HENDERSON2* and SCOTT STERN2

1School of Management, Boston University, Chestnut Hill, Massachusetts, U.S.A.and National Bureau of Economic Research, Cambridge, Massachusetts, U.S.A.

2Sloan School of Management, Massachusetts Institute of Technology, Cambridge,Massachusetts, U.S.A. and National Bureau of Economic Research, Cambridge,Massachusetts, U.S.A.

This paper begins to reconcile competing perspectives on the origins of competitive advantageby examining the adoption of ‘science-driven’ drug discovery, a performance-enhancing organi-zational practice. Science-driven drug discovery diffused slowly, allowing us to disentanglealternative theories of organizational heterogeneity. Adoption is driven by initial conditions,time-varying internal and external environmental conditions, and convergence (firms positionedleast favorably adopt most aggressively). While accounting for initial conditions is critical,managers are sensitive to idiosyncratic environmental cues. The origins of competitive advantagemay therefore lie in the ability to identify and respond to environmental cues well in advanceof observing performance-oriented pay-offs.Copyright 2000 John Wiley & Sons, Ltd.

INTRODUCTION

Strategy studies … is systematic, like the systemof touts at the race track … (Stinchcombe, 2000)

What are the origins of competitive advantage?Although this question is fundamental to strategyresearch, it is one to which we lack a clearanswer. As strategy researchers we believe thatsome firms consistently outperform others, andwe have some evidence consistent with this belief(Rumelt, 1991; McGahan and Porter, 1997). Wealso have a number of well-developed theoriesas to why, at any given moment, it is possiblefor some firms (and some industries) to earnsupranormal returns. As of yet, however, we haveno generally accepted theory—and certainly nosystematic evidence—as to the origins or thedynamics of such differences in performance. We

Key words: competitive advantage; pharmaceuticals;R&D; diffusion*Correspondence to: Rebecca M. Henderson, Sloan Schoolof Management, Massachusetts Institute of Technology, 50Memorial Drive, Cambridge, MA 02139, U.S.A.

Copyright 2000 John Wiley & Sons, Ltd.

know, for example, why high barriers to entrycoupled with a differentiated product positioningobtained through unique organizational com-petencies may provide a firm with competitiveadvantage. But we know much less about howbarriers to entry are built: about why this firmand not that one developed the competencies thatunderlie advantage, and about the dynamic pro-cess out of which competitive advantage firstarises and then erodes over time.

This conceptual ambiguity has always beenproblematic for many economists, who havetended to view persistent differences in perfor-mance as a function of ‘unobserved heterogeneity’(Mundlak, 1961; Griliches, 1986). For example,empirical work in industrial organization routinelycontrols for ‘firm fixed effects.’ These are usuallystatistically significant and often account for asubstantial fraction of the total variation in firmproductivity or performance. Whereas strategyresearchers tend to emphasize the degree to whichthese kinds of results offer support for the impor-tance of ‘capabilities’ or ‘positioning’ (Rumelt,1991; Henderson and Cockburn, 1994; McGahanand Porter, 1997; Lieberman and Dhawan, 2000),

1124 I. M. Cockburn, R. M. Henderson and S. Stern

economists tend to emphasize the possibility thatfixed effects are simply controlling for a seriesof much more mundane measurement problems,ranging from the difficulty of computing appropri-ately depreciated capital stocks and of measuringfirm-specific input and output price schedules, tothe problem of controlling for difficult-to-observefactors such as worker effort or worker quality.In short, the evidence which strategy researchersview as the motivation for their intellectualagenda are interpreted by many economists interms of ‘nuisance’ parameters—things whichmust be controlled for but which are not ofintrinsic interest.

This implicit critique has been reinforced bytheoretical and empirical research in the traditionof population ecology (see, for example, Hannanand Freeman, 1989). In summarizing the contri-butions of this literature and its application tostrategy, Stinchcombe (2000) charges that thepreponderance of strategy scholars have simplyfailed to understand (and certainly to systemati-cally account for) the implications of populationdynamics for performance heterogeneity. Stinch-combe suggests that if superior performancearises from the degree to which a firm’s resourcesand/or strategy ‘match’ the competitive environ-ment, and if resources are randomly distributedat ‘birth’ (or if the environment which firms faceat the time strategies are chosen and resourceinvestments are made is sufficiently uncertain),then performance heterogeneity simply reflectsthe fact that the realized competitive environmentfavors some strategies and some resource bundlesover others. Such a critique implies that the caseswhich motivate so much of strategy research, andindeed even some of our theoretical frameworks,are roughly equivalent toex postaccounts of theway in which a winning gambler chose to puther money on red rather than on black at theroulette table.

In this paper we argue that grappling with thisproblem should be of central concern to strategyresearchers: that while many of us are aware ofthe issue that Stinchcombe raises, without a moredetailed understanding of the origin and dynamicsof the development of competitive advantage werun the grave risk of meriting Stinchcombe’staunt that we are indeed ‘touts at the racetrack.’We suggest that empirical strategy researchersneed to move beyond studies of differential per-formance to more integrated studies which not

Copyright 2000 John Wiley & Sons, Ltd. Strat. Mgmt. J.,21: 1123–1145 (2000)

only identify those factors that are correlated withsuperior performance but also attempt to explorethe origins and the dynamics of their adoption.

We begin the paper with a brief literaturereview. By and large, Stinchcombe’s critique—and population ecology more generally—suggeststhat most performance differentials, particularlydifferences in the probability of survivorship, canbe explained by differences in a firm’s initialconditions, and, moreover, that differences ininitial conditions are largely the result of difficult-to-explain (and even harder-to-measure) differ-ences in each firm’s initial allocation of resourcesand capabilities. In contrast, strategy is centrallyconcerned with the process of how firms andmanagersrespond to and exploit environmentalsignals. For example, in the last 20 years, therehas been an explosion of powerful frameworksfor evaluating the determinants of differential per-formance, from Porter’s five forces framework tothe resource-based view to transaction-cost eco-nomics. While each of these frameworks offers asomewhat different explanation for heterogeneousperformance, all share two assumptions: that com-petitive advantage arises through earlier or morefavorable access to resources, markets, or organi-zational opportunities; and that exploiting suchopportunities reflects some degree of active inter-pretation of internal and external environmentalsignals by managers. Indeed, we argue that theseliteratures often share the implicit view that theorigins of competitive advantage lie in theunusual foresight or ability of the firm’s man-agers.

Of course, this characterization of the twoschools of thought is unrealistically stark: popu-lation ecologists recognize the possibility thatfirms may be able to adapt to their environmentand their competitive experience (Barnett andHansen, 1996), and strategy researchers under-stand that firm strategy and capabilities are sub-ject to powerful inertial forces (Christensen andBower, 1996). But as we discuss in some detailin the next section, substantial differences infocus do exist: while population ecology is princi-pally focused on exploring the performance impli-cations of strong organizational inertia, one ofthe core agendas of strategy is understandingwhich organizational structures allow firms to firstidentify and then exploit opportunities offered bytheir environment and so (potentially) overcomeorganizational constraints.

Origins of Competitive Advantage 1125

We then turn to a discussion of an empiricalmethodology that might enable us to explorethe relative importance of initial conditions andstrategic choice in shaping competitive advantage.Our goal is to lay out an empirical frameworkthat might allow us to compare and contrast thedistinctive implications of the strategic perspec-tive alongside the predictions of the populationecology literature, and thus to both offer a pre-liminary response to Stinchcombe’s critique andto begin disentangling the origins of competi-tive advantage.

We begin by distinguishing between differ-ences among firm in terms of their ‘initial con-ditions’ vs. differences in the rate at which theyadopt a particular performance-enhancing practice(or strategy) that has been linked to superiorperformance. Using Stinchcombe’s analysis, wewould expect the primary determinant of eachfirm’s degree of adoption at any particular pointin time to be the initial condition or foundingstate of that firm. This is not to imply thatStinchcombe’s analysis suggests that firms cannotchange, only that we believe that in general hewould argue that change will not be systemati-cally tied to the kinds of environmental cueswhich would be readily amenable to empiricalanalysis by strategy researchers.

We contrast this hypothesis with two expla-nations for diffusion which are consistent with astrategic perspective, and are more difficult toreconcile with Stinchcombe’s hypothesis. First, astrategic orientation would suggest that, in theabsence of organizational failure, those firmswhose early history places them in a particularlyunfavorable position will tend to be the firms thatrespond most aggressively in terms of the rate atwhich they adopt particular performance-enhancing practices or strategies. We label thisthe ‘convergence’ hypothesis, under which thekey empirical question is not so much whetherdifferences between firms will persist but howlong they take to erode. From this perspective,the dynamics of competitive advantage are drivenby two distinct processes: the exploitation ofparticularly favorable combinations of practicesand/or market positions by firms whose initialpositions ‘match’ their environment, and the ero-sion of these rents as competitors ‘catch up’ bymimicking the successful strategies of marketleaders.

Second, we suggest that, beyond initial con-


ditions and the process of convergence, adoptionrates may be associated with environmental cueswhich are more intensively experienced by somefirms rather than others. Firm strategy may thusbe responsive to factors which provide infor-mation about a distinctive opportunity to investin resources or strategies which will ultimatelybe associated with competitive advantage. Forexample, if firms are in different ‘niche’ marketsduring the early stage of an industry, then theinformation these firms will possess about thefuture evolution of the industry may be differentand their strategy should in principle respond tothese signals. From an empirical perspective, thishypothesis suggests that strategy will be a func-tion of the firm’s current or recentenvironment;of course, a second issue exists as to whether weshould focus on environmental cues which tendto be associated with internal (organizational)factors, external (market) factors, or both.

To illustrate these ideas, we apply this empiri-cal framework to an evaluation of the adoptionand diffusion of one particular strategy—the useof science-driven drug discovery—in the contextof the worldwide pharmaceutical industry. Webegin by briefly motivating the study of the adop-tion of science-driven drug discovery as a usefulcontext in which to think about the origins ofcompetitive advantage, and then turn to a dis-cussion of our data sources and variable construc-tion. We explore the measurement of both initialconditions and convergence, and then use ourqualitative knowledge of the industry to identifyfive distinct ‘environmental cues’ that might drive‘strategic’ adoption: a firm’s distance from publicscience, whether or not the CEO is a scientist,its accumulated stock of scientific knowledge,its market position, and the composition of itssales portfolio.

Our analysis is by no means definitive. Weoffer it as a preliminary descriptive analysis of acomplex phenomenon that raises as many ques-tions as it answers. However, we believe thatit illustrates concretely one approach to takingStinchcombe’s hypothesis seriously while alsoevaluating the origins of competitive advantagein a systematic manner.

Our findings are consistent with a perspectivein which both population ecology and strategyhave an important role to play in explainingpatterns of organizational heterogeneity. On theone hand, ‘initial conditions’—or the position of


each firm at the beginning of the period coveredby our data—play an important role. Firms differvery significantly in the degree to which theyhad adopted the techniques of science-driven drugdiscovery in the first years of our period, andthese differences persisted for many years. Butthere is also evidence for a powerful convergenceeffect, with the firms which were furthest frombest practice at the beginning of the period mov-ing most aggressively to adopt it. Finally, aftercontrolling for initial conditions and convergence,additional time-varying characteristics of the firmplay a modest role in explaining patterns of dif-fusion: both the composition of the firm’s salesportfolio and its market share are significantlycorrelated with the rate at which the practice ofscience-driven drug discovery is adopted by thefirm. Our results therefore provide substantialsupport for both Stinchcombe’s view of thesources of competitive advantage and for a moretraditional, ‘strategic’ view, the implications ofwhich are addressed in the conclusion.

THE ORIGINS OF COMPETITIVEADVANTAGE

Early studies of competitive advantage wererooted firmly in historical analyses and carefulqualitative research. This work could be inter-preted as suggesting that competitive advantagewas a complex phenomenon, that depended cru-cially on the active presence of superior leader-ship (Andrews, 1971; Selznick, 1957; Chandler,1962). For example, Chandler’s early work canbe read as implying that those firms who adoptedthe new M-form before their competitors gaineda strategic advantage, and, moreover, that thechoice to adopt the new organizational formreflected the structure and leadership qualities ofa company’s top management. Through the 1960sand 1970s, the study of ‘strategy’ was thus thestudy of what general managers or ‘leaders’should do and it was generally assumed thatdoing these things would make a difference: firmswith better leaders would make better choices andwould ultimately do better than their competitors.

Porter turned this paradigm on its head (Porter,1980). In transforming the study of ‘imperfectcompetition’ into the analysis of ‘competitiveadvantage,’ Porter shifted the focus of strategyresearchoutward, towards the analysis of the


firm’s microeconomic environment. Porter’sapproach yielded sharply defined tools for under-standing exactly why some firms (and industries)were likely to be more profitable than others. A‘five forces’ analysis is essentially a structuralmap of the underlying economics of an industry:a map of the degree to which competitors,entrants, substitutes, and vertical bargainingpower exert pressure on the margins of a firm ina particular industry. A firm operating in anindustry in which there are substantial returns toscale coupled with opportunities to differentiate,that buys from and sells to perfectly competitivemarkets and that produces a product for whichsubstitutes are very unsatisfactory (e.g., the U.S.soft drink in the 1980s), is likely to be muchmore profitable than one operating in an industrywith few barriers to entry, and a large numberof similarly sized firms who are reliant on a fewlarge suppliers and who are selling commodityproducts to a few large buyers (e.g., the globalsemiconductor memory market).1

Structural analysis is a powerful tool for under-standing why a particular strategic action (e.g.,branding or investment in complementary productareas) may be associated with supranormalreturns, but in and of itself says nothing aboutthe role of senior management—or the processof strategic choice—in determining profitability.Consider the case of Crown Cork and Seal.2 Thisclassic HBS case describes the metal can industry:an industry that has a classically unfavorablestructure and in which, in consequence, the vastmajority of firms are relatively unprofitable. Astructural analysis provides concrete insight intowhy it is so difficult for most firms to makesupranormal returns in steel can production. Inaddition, the case can be read as suggesting thatCrown itself earns supranormal returns becauseit has developed unique capabilities that haveallowed it to differentiate its product in a waythat is difficult for competitors to imitate. Doesthe case therefore imply that Crown was partic-ularly well managed or that it had chosen a‘good’ strategy? Analogously, does it also implythat those metal can producers that are not per-

1 Structural analysis has, of course, moved much beyondPorter’s original book, and we do not even attempt to summa-rize this literature here. For some recent contributions, seeBrandenberger and Nalebuff (1998) and Besanko, Dranove,and Shanley (2000).2 HBS #: 9-378-024.


forming well are managed by ‘bad’ managerswho have chosen ‘bad’ strategies? Certainly thisis the way that the case is often taught: as if thevision and determination of John Connelley, whodeveloped Crown’s strategy, was a critical deter-minant of its success. To take Stinchcombe’scritique seriously, however, is to wonder whetherCrown was simply lucky: whether it happened tobe managed by Connelley, who happened to beobsessed with customer service. Stinchcombeforces us to ask: would Connelley have been assuccessful, as ‘visionary,’ in choosing a strategyin another environment?

Porter’s analysis was fundamentally agnosticon this point, but a simplistic interpretation ofhis work seemed to imply that structural analysiscould be used prospectively: that doing strategywas about ‘choosing good industries’ or‘rebuilding industry structure.’ The literature filledup with ‘five force analyses’, and it was temptingto use them prescriptively: build these kinds ofbarriers to entry, structure rivalry along theselines, and your firm and perhaps your industrywould become more profitable. Notice that at itsroots this stream of work returns to the foundingassumption of the field: good strategy is aboutleadership, about foresight. Managers who aresmart enough to understand the implications ofstructural analysis and to make the commitmentsthat it requires are likely to outperform thosewho do not (Ghemawat, 1991; Shapiro andVarian, 1998). Certainly it is the case that mostteaching in strategy (at least implicitly) assumesthis to be true. We act as if we believe that ifwe teach our students to analyze industry struc-ture they will be better positioned to allocateresources into the ‘right’ industries, or to influ-ence industry structure in favorable directions.But we should recognize that while a variety offield studies certainly suggest that this is the case,there is, at least to our knowledge, no compellingquantitative evidence supporting this view: nobroad-based statistical study showing that firms inwhich senior management actively used analyticaltools to understand industry structure out-performed those that did not.

It was against this background that the‘resource-based view’ of the firm emerged(Wernerfelt, 1984; Barney, 1991; Peteraf, 1993).At one level, the resource-based view (hereafter‘RBV’) is simply a reinterpretation of theenvironmental perspective. Where the latter


describes analytically why a differentiated posi-tion within an industry coupled with high entrybarriers can lead to profitability, the formerredirects attention towards the underlying hetero-geneity making such a position sustainable. Forexample, while early environmental analysesseemed to suggest that competitive advantagearose from purely technological factors (such aseconomies of scale) or from unique assets (suchas a brand name reputation), the RBV emphasizedthe idea that these technological or market posi-tions reflect internal organizational capabilities,such as the ability to develop new products rap-idly, to understand customer needs profoundly,or take advantage of new technologies cheaply.Proponents of the RBV suggested that strategicinvestments directed towards theseinternal activi-ties might be of equal (or even greater) impor-tance in generating supranormal returns.

More importantly, however, the RBV deepenedthe discussion of causality by focusing attentionon two key insights about the sources of competi-tive advantage. First, in many cases, an industry’s‘structural’ features are the result of the organi-zational capabilities of its constituent firms: apowerful brand name, for example, may reflectyears of successful new product introduction andsuperb (and unique) marketing skills. Second,there are good reasons for thinking that the mar-ket for organizational capabilities may be imper-fect in exactly the kinds of ways likely to leadto the existence of supranormal returns. In partbecause of these two insights (which are implicitbut not always manifest in environmentalanalyses), the RBV is often positioned as an‘alternative’ to the environmental perspective. Inour view, such a positioning reflects a significantmisconception, since the RBV and the environ-mental perspective are complementary in manyimportant respects. Each proposes a model ofwhy firms may sustain superior performance, butthe two models are not mutually exclusive, atleast in terms of their empirical predictions: whilethe environmental view focuses attention onexternal industry structure, the RBV directs ustowards the fact that internal capabilities andinvestments provide the instruments and tools toshape this external environment.3

3 For example, from an environmental perspective, thepharmaceutical industry has historically been profitablebecause buyers and suppliers are weak, entry is costly and


Moreover, both literatures offer a similar theoryabout the process of strategic choice. In the caseof environmental analysis, while economics isused to explain what kinds of strategic positionsare likely to be most profitable, the theory isessentially agnostic as to how firms come toassume them. Firms can be lucky, they can befast, or they can be far sighted: as long as theydeal with the ‘five forces’ they will be profitable,and the power of the tools lies in explainingexactly what these forces are and what kinds ofmechanisms will help a firm deal with them. Atleast at some level, many (perhaps most) empiri-cal treatments within the RBV tradition follow asimilar logic, with the caveat that rather thanemphasizing the fact that one firm rather thananother ‘chose’ a particular market position orproduction technology, the analyses focus on thefact that one firm ‘chose’ to develop a certainset of internal capabilities or unique organi-zational assets (see, for example, Henderson andClark, 1990; Clark and Fujimoto, 1991; Eisen-hardt and Tabrizi, 1995).

At a more subtle level, however, the RBVliterature begins to address causality and the ulti-mate origins of competitive advantage moredeeply. It has implicit within it a significantlydifferent view of the dynamics of strategic advan-tage, and, in particular, of exactly what managerscan and cannot do. While the canonical referencefor the RBV literature is usually taken to bePenrose (1959), the RBV perspective on the stra-tegy process seems to be influenced more byStinchcombe (1965), on the one hand, and Nelsonand Winter (1982), on the other. Specifically,RBV scholars often seem to suggest that organi-zations are fundamentally different from eachother for reasons that may have very little to dowith any kind of ‘strategic logic,’ and that theycan only change through limited, local search.To the extent that competencies are built onorganizational routines that are only tacitly under-stood—indeed if tacit understanding and com-

difficult, and substitutes and rivalry are quite muted. At leastin part, this structure reflects external forces, such as the factthat the industry’s products can be securely protected throughpatents, and so is not the result of a specific organizationalcompetence on the part of a particular firm. But the industry’sattractiveness also reflects the unique competencies developedby the larger pharmaceutical firms, including, among otherfactors, their years of investment in sophisticated researchcapabilities, their knowledge of regulatory systems around theworld, and their extensive distribution and physician networks.


plexity are a prerequisite for any competence tobe a source of competitive advantage—then theremay exist a fundamental tension between the factthat competencies lie at the heart of competitiveadvantage and the use of this insight to guidestrategy choice (Leonard-Barton, 1998). From thisperspective, the simple observation that com-petencies may lead to advantage is only halfthe battle: the other half is understanding wherecompetencies come from.

The theoretical RBV literature thus makesexplicit a view of ‘strategy’ that has long beenlatent—the sense that strategy is not all, or notonly, about the cognitive ability of the seniormanagement and their ability to make the ‘right’decisions, but also about their ability to workcreatively with the raw material presented bytheir firm and their environment (Quinn, 1978;Mintzberg, 1987); to respond appropriately whentheir firm’s organizational structure finds ‘good’strategies (Burgelman, 1994); and to createdecision structures and procedures that allow afirm to respond to its environment adaptively(Bower, 1974; Levinthal, 1997). In short, infocusing on the dynamics of competence andresource creation, the RBV is centrally concernedwith the degree to which successful firms areindeed ‘lucky’—since it suggests that many ofthe competencies underlying advantage are theresult of investments made under a heavy cloudof uncertainty and that they are subject to localbut not globally adaptive evolution.

This approach promises a bridge between theinsights of those who stress ‘luck’ or ‘initialheterogeneity’ in shaping firm performance andthe central insight of the strategy field: that whatmanagers do matters. But we believe that theseimplications have not been fully worked out.Most importantly, the RBV has not generated thekinds of empirical studies of adoption that arecrucial both to fleshing out a full response toStinchcombe’s critique and to building a richerunderstanding of the origins of competitiveadvantage. While there are studies suggesting thatthe possession of unique organizational com-petencies is correlated with superior performance(Henderson and Cockburn, 1994; Powell, Koput,and Smith-Doerr, 1996), and others suggestingthat competitive advantage may be heavily influ-enced by conditions at a firm’s founding(Holbrook et al., 2000; Eisenhardt, 1988; Eisen-hardt and Schoonhoven, 1990), so far as we are


aware there are few careful studies of thehypothesis that successful strategy is the success-ful management of the evolution of organizationalskills and its changing environment.4 Indeed,some of the most elegant and widely cited studiesof the diffusion of organizational innovation donot control for firm founding conditions (see,for example, Davis, 1991). At its worst, just asoverenthusiastic readings of Porter led to theprescription ‘choose a good industry,’ overenthu-siastic readings of the RBV have had the flavorof ‘build the right resources/competencies.’ EvenBarney’s original piece has something of thistone, as for example when he suggests that:

To be a first mover by implementing a strategybefore any competing firms, a particular firmmust have insights about the opportunitiesassociated with implementing a strategy that arenot possessed by other firms in the industry, or bypotentially entering firms … (Barney, 1991: 104)

Such an analysis clearly places the source ofcompetitive advantage in the insight of the firm’smanagers, and leaves unanswered the criticalempirical question:how does one know?Is thesuccess of Sony in introducing the Playstationthe result of a strategic competence possessed bySony’s senior managers? Or is it the result of acomplex history, an evolutionary process bywhich Sony was ‘lucky’ enough to be in theright place at the right time? In short, what arethe concrete and distinct implications of Stinch-combe’s and Nelson and Winter’s insights forour understanding of strategy?

EMPIRICAL METHODOLOGY

In the remainder of this paper, we begin to addressthe implications of these questions for empiricalresearch in strategy, at least in a preliminary way.Distinguishing between alternative theoretical per-spectives in evaluating the origins of competitiveadvantage is a massive and difficult task. Theempirical work in the remainder of this paper

4 There is of course a powerful literature demonstrating thatchanges in top management team compensation are correlatedwith major strategic reorientations (see, for example, Fletcherand Huff, 1990, and work by Tushman and his collaborators,including Murmann and Tushman, 1997). While this literatureis suggestive, in general it leaves the question of causalityunaddressed.


should thus be interpreted as outlining one typeof empirical approach which might yield fruitfulanalysis of the origins of competitive advantagerather than as a definitive test of one particulartheory in the specific context that we examine.

Our empirical analysis is designed to teaseapart the issues that we identified above. Weidentified two core hypotheses: on the one hand,the view that competitive advantage is largelydetermined by factors put in place at the organi-zation’s founding; and, on the other, the viewthat competitive advantage results from a firm’s‘strategic’ response to changes in this environ-ment or to new information about profit oppor-tunities. We believe that one way in which onemight plausibly begin to separate these twoeffects is to identify an organizational practicethat is closely associated with competitive advan-tage, and then to explore the determinants ofthe diffusion of this practice across a populationof firms.5

To do this, we build on a long tradition ofmodeling the process of diffusion of new technol-ogies. Our analysis consists of the estimation andinterpretation of a ‘diffusion equation’yj,t = f(I j,0,Zj,t, t; Ω), where yj,t, our dependent variable, isa measure of the extent to which firmj hasadopted the practice as of timet, and the explana-tory variables are chosen to illuminate the com-peting hypotheses.Ij,0 is a measure of the initial,or founding conditions of firmj, while Zj,t is avector of variables which reflect the opportunitiesand environmental conditions affecting firmj attime t, and thus highlight the existence of poten-tial sources of differences across firms. Forexample,Zj,t might include measures of the natureof firm decision making, of absorptive capacity orother strategies associated with seeking externalsources of knowledge, of market position, or ofthe talent or focus of managers in the firm. Time,t, plays a critical role as an explanatory variablehere, since it allows us to capture therate ofdiffusion of the practice, and so allows us to testwhether firms in alternative environments tend tohave faster or slower rates of adoption of thepractice in question.Ω is the vector of parametersto be estimated.6

5 Note that our focus on diffusion contrasts with the traditionalemphasis on explainingperformanceas a function of initialconditions, environmental factors, or strategic choices.6 In the remainder of this discussion we suppress the firm-specific subscriptj on all the variables.


Within this general framework, choices aboutfunctional form or about which of these explana-tory variables to include correspond to differentperspectives on the sources of competitive advan-tage. Three possibilities are of particular interest.First, perhaps the simplest model one can imaginewould be to regress yt on a favored set ofZt’s,from which one might conclude that the levelof adoption of the practice was satisfactorily‘explained’ by the Z’s, providing support for a‘strategic response’ perspective.

Alternatively a second approach would be toregressyt on I0 by itself, perhaps to test theconclusion that ‘it’s all initial conditions.’ Finally,our focus on diffusion allows us to evaluate athird (somewhat more subtle) implication of the‘environmental’ perspective, namely that firmswho find themselves to have chosen strategiesor organizational practices which are particularlyunfavorablewill be likely to have a higher rateof adoption than firms who are near the ‘frontier’of profitability. In other words, even if the levelof yt is positively associated with I0 (there is‘persistence’ in the impact of initial conditions),a strategy-oriented perspective suggests that yt

will be negativelyassociated withI0 3 t. Thatis, firms which begin at the ‘lowest’ level willhave the highest rate of increase. This strategicresponse, if it occurs, will manifest itself in thedata in the form ofconvergenceover time among(surviving) organizations in terms of the organi-zational practices which they exhibit—the lag-gards will ‘catch up’ with firms who were ‘lucky’enough to have chosen favorable practices at theirfounding (or at the beginning of the observedsample period).

To assess the relative salience of ‘initial con-ditions’ and environmental heterogeneity in driv-ing patterns of adoption of a performance-enhancing practice, we require an empirical modelof the dynamics of diffusion which allows forboth effects to work factor simultaneously. Wetherefore specify a model which includes a firm’sinitial conditions, I0, a firm’s environment andopportunity set at a given point in time, Zt, andinteractions between each of these elements andtime, t.

yt = a + bt + gI0 + dI0·t + wZt + rZt·t (1)

The parameters of this model nest a number ofkey hypotheses about the determinants of yt


which can be formulated as restrictions on theparameters.7 For example, a naive reading ofpopulation ecology would suggest that, beyond afirm’s initial conditions, there should be no sys-tematic pattern to diffusion relating to the firm’schanging environmental circumstance: some firmswill be able to execute the practice as a functionof their founding conditions and others will not.Thus for a given set of firms, current levels ofadoption should be a function only of initialconditions, implying thatw = ρ = δ = 0.

Conversely, the simple environmental or ‘stra-tegic’ hypothesis implies that yt should also bedriven by Zt: firms may change their behavior inresponse to changes in the environment with lev-els of adoption not wholly determined by a firm’sstarting position. This would implyw Þ 0 and/orr Þ 0. For example, we believe that one couldinterpret a finding that yt was correlated with thecurrent market position of the firm as suggestingit is indeed reacting to exogenous shocks in itsenvironment, independent of its founding con-ditions. Similarly a finding thatyt was correlatedwith the experience of the firm’s CEO wouldsuggest that the firm is changing its strategy asit develops new capabilities.

Finally, the ‘convergence’ hypothesis suggeststhat, in addition to a firm’s initial conditions, andits responsiveness to objective measures of itsenvironment, firms who face unfavorable initialconditions may be those who are most likely tohave the highest rate of adoption, that is,δ < 0.This would imply that initial conditions areimportant in shaping strategy, though hardly inthe way that Stinchcombe hypothesized.

Before turning to the data, two further model-ing issues must be addressed. The first is that offunctional form: if we interpret yt as ‘fractionadopted’ then it would be appropriate to followthe diffusion literature in recognizing that the

7 Note that this is very much a ‘reduced form’ model ofadoption. It is not our goal (at least in this paper) to developor estimate a fully specified structural model of optimizingadoption behavior. Indeed, given the diverse set of plausiblehypotheses about the drivers of adoption behavior, it is notclear that estimating such a model is currently feasible withoutpreliminary empirical work, which substantially narrows downthe potential range of theories to be accommodated. Conse-quently, we confine ourselves here to identifying the principalcovariates of adoption, recognizing that adoption is a dynamicprocess. Distinguishing the relative empirical salience of com-peting hypotheses may provide an initial guide to specifyinga more fully articulated structural model.


diffusion of most practices or technologies fol-lows an S-shaped (or sigmoid) pattern throughtime. There are a variety of functional formswhich allow for this type of nonlinear time path,but the simplest transformation involves takingthe log-odds ratio (log(yt/(1 2 yt)) of the depen-dent variable measuring the extent of adoption(Griliches, 1957). On the other hand, ifyt is amore general indicator of the use of a practice,then theory offers us little guidance as to func-tional form.8

A second issue relates to the measurement ofinitial conditions, I0. Two possibilities presentthemselves: either to take on the formidable taskof identifying and explicitly measuring all therelevant aspects of heterogeneity in firm capabili-ties, or, more straightforwardly, to capture theirempirical content withy0—the presample valueof yt. This second option is attractive not justbecause it is easy to implement, but also becauseit allows us to be agnostic about the precisenature of the initial characteristics of firms whichdrive their subsequent evolution.9 Actual esti-mation of these models therefore requires dataonly on y and Z. We now turn to our specificapplication of this framework, the diffusion ofscience-driven drug discovery in the worldwidepharmaceutical industry.

DATA AND VARIABLECONSTRUCTION

Competitive advantage from science-drivendrug discovery

We focus here on the adoption of ‘science-drivendrug discovery’ as a primary strategy in themanagement of pharmaceutical research. As wehave laid out in a number of papers, prior to thelate 1970s pharmaceutical research was conductedlargely through a process of so-called ‘random’

8 Our empirical results are surprisingly robust to variation inthe functional form imposed on the adoption-intensity variable.Our robustness checks (not presented here, but available onrequest) included using the level of adoption intensity as wellas the natural logarithm.9 Conditioning on the starting value of the dependent variable(or some function of it) is a well-understood procedure inthe econometrics literature on dynamic panel data models(see, for example, Arellano and Bond, 1991). This also allowsus to understand the implications of our results in the frame-work the heterogeneity/state dependence distinction used byeconometricians (see Heckman, 1991).


search (see, for example, Henderson, Orsenigo,and Pisano, 1999, and papers referencedtherein).10 From the late 1970s on, firms beganto respond to the acceleration in the growth ofpublicly available biological knowledge by adopt-ing a new mode of so-called ‘science-driven’drug discovery. This move appears to have beentightly associated with the creation of competitiveadvantage: those firms adopting the new tech-niques appear to have been significantly moreproductive (Gambardella, 1995; Henderson andCockburn, 1994) in an industry in which theintroduction of blockbuster drugs is a source oftremendous value (Grabowski and Vernon,1990).11

Despite the power of this strategy, its diffusionacross the industry was surprisingly slow. As lateas 1991, some firms continued to view scientificpublication and participation in the broader com-munity of public science—a key requirement forthe full adoption of science-driven drug dis-covery—as a diversion from the more ‘important’business of finding drugs. As the research man-ager at one of these companies put it:

Why should I let my people publish? It’s just awaste of time that could be spent in the searchfor new drugs. (Quoted in Henderson, 1994)

Here we argue that exploring the factors thatdrove these differences in the rate of adoptionacross firms not only concretely illustrates thedifficulty of distinguishing between explanations,but also serves as a springboard for thinkingthrough the relationship between theories of adop-

10 Notice that this terminology can be deceptive. The majorpharmaceutical firms began investing in fundamental scienceafter the First World War, and by the 1950s many of thememployed significant numbers of well-trained scientists(Parascandola, 1985). ‘Random’ drug discovery was ‘random’in that while research was often guided or informed byfundamental science, the precise mechanism of action of manydrugs was not understood, and new compounds were screenedagainst animal models rather than against particular diseasepathways (Henderson, 1994; Henderson, Orsenigo, and Pis-ano, 1999).11 The question of justwhy ‘science-driven drug discovery’should be a major source of competitive advantage is afascinating one. The interested reader is referred to our relatedwork (Henderson, Orsenigo, and Pisano, 1999; Cockburn andHenderson, 1998; Cockburn, Henderson and Stern, 1999a,1999b; Stern, 1999). There are, of course, other sources ofcompetitive advantage in the pharmaceutical industry, in-cluding the possession of strong regulatory capabilities, anextensive distribution network or well-developed marketingcapabilities.


tion and theories of the source of strategic advan-tage. The remainder of this section describes thesources of the data from which we construct ourempirical tests, our alternative measures of thepractice itself,yt, and measures associated eachfirm’s environment at a given point in time,Zt.

Data sources and sample construction

We draw on two primary data sets. The firstcontains detailed qualitative data and quantitativedata for 10 major pharmaceutical firms for theperiod 1965–90. These data formed the basis forour previous work and are described in detailthere (Henderson and Cockburn, 1994, 1996). Thesecond data set consists of a sample of 16 largeresearch-oriented pharmaceutical firms for the pe-riod 1980–97.12 Overall the average firm in bothsamples is somewhat larger than the average pub-licly traded pharmaceutical firm. While they donot constitute a random sample in a statisticalsense, they are reasonably representative of thelarger pharmaceutical firms who invest substan-tially in original research and comprise a substan-tial portion of the industry (the firms in thesample account for approximately 50 percent ofU.S. pharmaceutical sales in 1993). While wewere able to obtain detailed, internal data for thefirst sample, the second relies on comprehensivedata collected from secondary sources, includinginformation on the composition and backgroundof senior management, firms’ geographicallocation and patenting activity, scientific papers,and product sales compiled at the therapeuticclass level (e.g., cardiovascular therapies, anti-infectives, or cancer). Data about senior man-agement were collected from annual reports,while geographical data were derived from thepublication record of each firm. Our source forpatent data is Derwent Inc.’sWorld Patent Index;scientific publications are drawn from ISI’sScience Citation Indexand Web of Science. Salesdata are from various publications of IMS Amer-ica, a market research firm.

12 The firms are: Abbott, Bristol-Myers Squibb, Burroughs-Wellcome, Ciba-Geigy, Glaxo, Fujisawa, Hoechst, HoffmanLa-Roche, Lilly, Merck, Pfizer, Sandoz, Searle/Monsanto,SmithKline Beecham, Takeda, and Upjohn. Nine of thesefirms are also included in the earlier sample: the remainderwere selected to include the industry’s leading R&D per-formers and to obtain worldwide geographical representation.


Measuring the adoption of ‘science-drivendrug discovery’ (yt)

A quantitative analysis of the adoption of science-driven drug discovery requires the constructionof a reliable measure of the extent to which firmsare engaged in the practice.

Our initial measure of this practice, PROPUB,identifies the degree to which researchers weregiven incentives to establish standing in the scien-tific public rank hierarchy. In prior work(Cockburn and Henderson, 1998; Cockburn,Henderson, and Stern, 1999b), we have arguedthat the extent to which a firm has adopted thispractice is an indicator of how thoroughly theyhave embraced the techniques of science-drivendrug discovery.13 This measure was derived fromdetailed qualitative interviews conducted within10 major pharmaceutical firms. While it has thegreat virtue of being a direct measure of theorganizational practice in which we are interested,it unfortunately suffers from two important limi-tations. First, it was derived from qualitative inter-views conducted by a single researcher, whichmay raise questions as to its reliability and rep-licability. Second, PROPUB is not currently avail-able beyond 1990 or for the 11 firms in thesample that were not included in our earlier datacollection efforts.

We therefore also evaluate the diffusion ofscience-based drug discovery using three quanti-tative measures derived from public sources.These variables all draw on bibliographic infor-mation collected for every paper published in thejournals indexed in the Institute for ScientificInformation’s Science Citation Indexbetween1980 and 1994 for which at least one author’saddress was at one of our sample of firms. Themost straightforward, PAPERS, is a simple publi-cation count: the number of papers published inany given year for which at least one author wasaffiliated with our sample firms. As prior researchhas established, pharmaceutical companies pub-lish heavily, with annual counts of papers compa-

13 In the ideal case we would like to be able to use a measurebased on the actual nature of the science conducted withineach firm over time: in the absence of reliable longitudinalsurvey data about research methodology this is impossible toconstruct from secondary sources. Instead we use here anumber of measures derived from the publication behaviorand incentives provided by research-oriented pharmaceuticalfirms.


rable to the output of similarly sized universitiesand research institutes (Koenig, 1982; Hicks,1995). Publication counts are clearly an importantindicator of research activity, and have been pre-viously interpreted as capturing the level ofinvestment in ‘basic science’ (Gambardella,1995). Of course, since the volume of publicationactivity may simply reflect the scale of the firm’sresearch effort, when we use any of these meas-ures as a dependent variable we control for sizeby including total U.S. pharmaceutical sales inthe regression.14

While PAPERS has several attractive propertiesin terms of capturing the degree to which firmshave adopted the science-based drug discoverytechniques, PAPERS is also fundamentally anoutput measure, and may therefore also measurefactors such as the success of the firm’s researchprogram: the discovery of interesting compoundsalso generates interesting papers. Using the samedata, we therefore constructed AUTHORS: a vari-able which identifies the number of science-oriented researchers at each firm. AUTHORS isa count of the number of distinct individualspublishing at the firm in any given year. Wehypothesize that at those firms in which the scien-tists face high-powered incentives to raise theirstanding in the public eye the marginal incentivesto publish will be higher and the ‘marginal’scientist will make more of an effort to get theirname on the papers to which they have contrib-uted.15

Although counting authors may give us a bettermeasure of the adoption of the new practice thana raw publication count, it remains an outputmeasure. Our preferred measure of the degree towhich the firm has adopted propublication incen-tives is PUBFRAC: the fraction of individualswhose names appear on a patent who also appear

14 Total spending on research would clearly be preferable, butfor the complete sample we have only data on total R&Dexpenditures. This measure is very noisy in that it includesdevelopment expenditures in addition to discovery expendi-tures, and we know from our prior work that the ratio ofdiscovery to development spending is not constant acrossfirms. It is a particularly noisy measure when the firm isdiversified beyond the pharmaceutical industry.15 Another possibility would be to count ‘stars’: the numberof individuals in a firm who publish prolifically. Zucker,Darby, and Brewer (1998) have suggested that this ‘higherstandard’ is critical for understanding successful performancein, at least, biotechnology firms. In preliminary work we havefound that this variable performs much like AUTHORS.


as an author on papers published within 2 yearsof the patent application. By explicitly tying pub-lishing and patenting together, this measureincorporates the degree to which a firm is encour-aging those researchers who are directly involvedin the firm’s drug discovery process to participatein scientific publication. In addition, since this ismeasured as a share rather than an absolute num-ber of papers or authors, in principle it capturesthe propensityto publish, independent of the scaleof the firm, either in terms of sales or numberof employed scientists.16

Table 1 provides summary statistics and thecorrelation coefficients for these four variables. Inaddition to the relatively high levels of objectivemeasures such as PAPERS and AUTHORS, it isworth noting that all four of the measures are

Table 1. Descriptive statistics for alternative depen-dent variables(a) Summary statistics

N Mean S.D.

PROPUB 83 3.6 1.3AUTHORS 153 854 550PAPERS 153 299 189PUBFRAC 153 0.65 0.14

(b) Correlation coefficients

PROPUB PUBFRAC AUTHORS

PUBFRAC 0.11AUTHORS 0.56 0.42PAPERS 0.55 0.38 0.35

16 Constructing PUBFRAC was far from straightforward. Foreach firm in the sample, we first identified all papers (a)which were published between 1980 and 1994 in journalsindexed by theScience Citation Indexand (b) in which thename of the firm, or one of its subsidiaries, appears in atleast one of the authors’ addresses. We had then to attemptto match the names of many thousands of individuals acrosstwo different data sets and consistently apply rules for ambigu-ous cases. Much of this matching was accomplished straight-forwardly in software, but a number of difficulties did inducesome measurement error into the process. These includedtypographical errors in the source data, and differences acrosspapers by the same author in the use of initials, surnames,and given names; extensive hand-coding was necessary tocomplete the task. A consistent matching procedure wasapplied to all firms, and so we are reasonably confident thatbias in the measurement of PUBFRAC is limited to differ-ences across firms in the severity of these problems.


quite closely correlated. The time path of thesevariables indicates the gradual diffusion of thepractice across the industry over time. Table 2shows the evolution of both the mean values andthe coefficient of variation (standard deviation/mean) for each of our four measures of science-driven drug discovery. Not only does the meanlevel of each measure rise, but the coefficient ofvariation steadily falls, suggesting that there wasconsiderable ‘convergence’ towards a commonlevel of intensity of use of the practice at thesame time that its average level increased. Notethat this diffusion is not driven primarily by theexit of firms from the sample.17 In itself, thisimmediately raises questions about the extremeform of Stinchcombe’s hypothesis: the firms inthe sample appear to be actively adopting thepractice of science-driven drug discovery over theperiod covered by our data, an observation whichsuggests that there is more going on than a simpleinitial random distribution of capabilities followedby an exogenous environmental shock. Indeed, itis consistent with a ‘strategic’ view in whichmanagers gradually recognize the opportunities

Table 2. Means and coefficients of variation over time, alternate dependent variables

PROPUB PAPERS AUTHORS PUBFRAC

Year Mean C.V. Mean C.V. Mean C.V. Mean C.V.

1976 2.40 0.711977 2.40 0.711978 2.50 0.661979 2.70 0.551980 2.70 0.551981 3.10 0.51 182.4 0.63 393.6 0.60 0.54 0.301982 3.30 0.43 197.2 0.65 431.7 0.61 0.58 0.241983 3.30 0.43 210.1 0.67 478.3 0.64 0.53 0.311984 3.40 0.42 199.3 0.69 489.9 0.66 0.56 0.291985 3.40 0.42 252.0 0.74 615.8 0.68 0.58 0.251986 3.50 0.41 257.9 0.67 656.3 0.62 0.66 0.181987 3.50 0.41 283.9 0.67 746.3 0.55 0.68 0.171988 3.60 0.37 279.8 0.68 800.1 0.59 0.67 0.211989 3.67 0.36 309.1 0.50 913.9 0.46 0.64 0.161990 4.00 0.23 370.6 0.46 1128.8 0.38 0.68 0.131991 376.8 0.47 1230.1 0.41 0.68 0.161992 452.5 0.46 1436.0 0.45 0.73 0.121993 460.1 0.49 1553.9 0.45 0.80 0.13

17 Of our original sample of 10 firms, two exited throughmerger. Of our later sample of 19, only two have (at thetime of writing) exited through merger.


presented by the new techniques and move toadopt them. These simple statistics highlight theempirical salience of the convergence hypothesis:that initial conditions engender persistence overthe medium term, but that laggard firms aggres-sively move to adopt these techniques, erasingthe importance of the past in the long term.

Measuring initial conditions, I 0

As discussed above, rather than attempt to charac-terize initial conditions directly, we use y0, or thefirm’s initial level of adoption of the practice ofscience-driven drug discovery, as the summarystatistic for its initial condition. This is calculatedfrom average level ofy in the 2 years precedingthe sample period.

Defining Z: measures of firm heterogeneity

We draw on the literature and on our qualitativedata to construct six distinct measures of hetero-geneity in the environment and opportunity setsof each firm. Some of these focus on general


Table 3. Descriptive statistics for independent variables

Variable Mean S.D. Min. Max.

Cardiovascular patents share of firm 12.47 4.40 5.18 27.98patents (%)Cardiovascular therapies as share of 18.54 19.50 0 73.78firm sales (%)Share of firm in U.S. cardiovascular 5.27 8.71 0 44.92market (%)Total firm sales ($ million) 938.8 701.1 99.9 3653.7Scientist–CEO 0.065 0.248 0 1Distance from science 0.363 0.515 0.001 2.290

(150 observations)

characteristics of the firm: others focus on therole of cardiovascular therapies—a leading appli-cation for science-driven drug discovery—in thefirm’s research and sales portfolio and on thefirm’s share of the market for cardiovascular ther-apies. This list of variables includes both factorsthat might ‘rationally’ drive differences in thediffusion of science-driven drug discovery—including the firm’s closeness to public scienceand its technological experience and market posi-tion—and factors that might shape the attentionof senior managers, including whether the firm’sCEO has a scientific background and the compo-sition of its sales portfolio. Our discussion herefocuses largely on the qualitative and theoreticalevidence that motivates each variable: full detailsof the ways in which each was operationalizedare included in the Appendix. Table 3 presentsdescriptive statistics for these variables.

(a) Scientist–CEO

Qualitative interviews with senior industry parti-cipants often explained variations across firms inthe rate at which they moved to science-drivendrug discovery in terms that resonate deeply witha ‘strategic’ picture of adoption.18 Our informantssuggested that differences in ‘leadership’ or‘vision’ across firms had a very significant effect

18 In order to assist us in framing our hypotheses, betweenSeptember 1998 and April 1999 we conducted interviews atseven pharmaceutical firms. All but one were based in theUnited States. The interviews were loosely structured dis-cussions that explored, first, whether it was plausible that theadoption of science-driven drug discovery had significanteffects on research productivity and, second, why it mighthave been the case that despite their plausible impact onproductivity many firms were slow to adopt them.


on the decision to adopt the new mode ofresearch. One senior researcher remembered:

We spent most of the seventies going to Tgroups. It was fun, but we didn’t get muchscience done … The managers of the firm werelargely focused on using the cash generated bythe pharmaceuticals group to look for diversifi-cation opportunities …

Managers in firms that failed to adopt the newtechniques looked back on their failure as anembarrassment:

By now you will have worked out that we didn’treally do any real research in the eighties. Wetalked about it, but X didn’t understand what itrequired. We kept doing what we had alwaysdone …

The decision to invest in the science-drivenapproach was often identified with the hiring ofkey individuals:

Oh, that’s when we hired Y. He built a com-pletely new research facility and started to hireaggressively. He had a fundamental belief thatif we did leading edge science we would findbreakthrough drugs…

Capturing this insight empirically is a challengingtask. One option is to explore the correlationbetween changes in research strategy and changesin senior management personnel. Unfortunatelywe have not yet been able to collect this infor-mation in rich enough detail for our full sample.Here we use a binary variable (Scientist–CEO)measuring whether or not the firm’s CEO has ascientific background as a first attempt at captur-ing the idea. We hypothesize that firms with


scientifically trained CEOs will be more likely,all else equal, to adopt the techniques of science-driven drug discovery. While the classicalenvironmental approach implicitly suggested thattop managers who can deeply understand thevalue associated with novel strategies isimportant, the RBV literature has placed muchgreater emphasis on these kinds of factors.

(b) Cardiovasculars as a share of total firmsales

Our qualitative evidence suggests that the speedwith which the new techniques were adopted wasa function of the balance of power within thefirm. Those firms whose sales portfolios weredominated by therapeutic classes in which thenew techniques were likely to be particularlyimportant—particularly by sales of cardiovasculardrugs—appear to have become convinced that thenew techniques were likely to be important muchfaster than managers in those firms in whichsales were much more heavily concentrated intherapeutic areas for which the new techniqueswere much less useful. Thus we hypothesize thatfirms whose sales are dominated by cardiovascu-lar therapies will adopt the new techniques fasterthan their rivals.

Notice that because of the very long lead timescharacteristic of the pharmaceutical industry,share of the sales portfolio isnot highly correlatedwith share of the patent portfolio. Thus we inter-pret any effect of the share of sales portfolio onthe diffusion of science-driven drug discovery asa reflection of the focus of attention of seniormanagement attention, rather than as a responseto any difference in the firm’s research base.19 Itmay also, of course, reflect the presence of mar-keting and sales assets.

(c) Distance from public science

Our qualitative evidence suggests that differencesin geographic location may also have had system-atic effects on the decision to adopt the new

19 Share of the sales portfolio might also shape the choice ofresearch technique if firms have specialized marketing ordistribution assets. These assets certainly exist—Eli Lilly, forexample, has historically had a very strong position in thedistribution and marketing of treatments for diabetes. How-ever, we believe that in the majority of cases they arerelatively fungible.


research techniques since the benefits of adoptingthe new approach may well have been signifi-cantly higher (and the costs significantly lower)for those firms whose research laboratories werelocated in reasonably close proximity to largecommunities of publicly funded researchers.Close proximity not only made it easier to attractthe best-quality researchers, but probably alsomade it much easier to maintain close connectionsto the public sector once these researchers hadjoined the firm (Jaffe, 1986; Zuckeret al., 1998).Indeed, both the classical environmental perspec-tive and the RBV have emphasized the role ofgeographic location in shaping the ability of firmsto more aggressively adopt practices associatedwith superior performance. We have developed ameasure, DISTANCE, of the geographical posi-tion of a firm’s research activities using the publi-cations data from ISI.

(d) Knowledge capital: cardiovascular patentsas a percentage of total firm patents

One of the key claims of the RBV is that a firm’sparticular experience with specific technologies ortechnological trajectories provides it with a firm-specific capital stock which it can both exploitas well as providing a mechanism to draw uponoutside knowledge. In the current context, thetechniques of science-driven drug discovery werenot immediately applicable to all areas of drugresearch. In the late 1970s and early 1980s, forexample, many scientists believed that the newtechniques would be much more useful in thesearch for cardiovascular therapies than theywould be in the search for new anti-infectivedrugs. Scientific understanding of cardiovasculardisease was much further advanced, and it wasthought that this understanding would provide aparticularly rich basis for drug discovery. Firmswith more experience in cardiovascular therapiesmight therefore have expected to benefit dispro-portionately from adoption of the new techniques.Thus, in the RBV tradition, we hypothesize thatadoption will be positively correlated with experi-ence in cardiovascular disease.

(e) Market position: share of thecardiovascular market

Following the classical environmental perspective,we also expect a firm’s position in the product


market to have an effect on its decision to adoptthe new techniques. A large literature in economicslinks the degree to which a firm possesses mo-nopoly power in the product market to its decisionto adopt an innovation (Gilbert and Newbery,1982; Reinganum, 1981; Fudenberg and Tirole,1985; Henderson, 1993; Karshenas and Stoneman,1993). Unfortunately this literature yields no veryclear predictions—under some circumstances mon-opolists appear to have incentives to adopt inno-vationsbeforeentrants; under others they have anincentive to wait. Moreover, it has not directlyaddressed the question of the drivers of adoptionof a research technique, as opposed to the adoptionof a new product. Nonetheless, we believe that inthis context the most plausible interpretation ofthis literature is that firms with monopoly powerin the relevant product market will have higherincentives to adopt the new techniques, since theyare more likely to benefit from them. Thus, consist-ent with an externally oriented environmental per-spective, we evaluate whether firms that have ahigh share of the cardiovascular market will adoptthe new techniques faster than others.

(f) Firm sales

Finally, total U.S. sales is included to control forthe fact that there may be economies of scale inthe adoption of the new techniques. Zucker andDarby (1997) found evidence for economies ofscale in the adoption of the techniques of biotech-nology, and to the degree that ‘science-driven’drug discovery also takes advantage of scientificresources that can be maintained centrally andexploited throughout the organization one wouldexpect economies of scale to be important in thiscase as well. Indeed, one of the key tenets ofmost strategy perspectives (either the classicalenvironmental perspective or RBV) is thatscalematters, if only in terms of having sufficientliquidity or slack resources to aggressively adoptperformance-enhancing organizational practices.

Other potential sources of heterogeneity

This list of variables could be expanded almostindefinitely. We do not explore, for example, thedegree to which either a firm’s position in itsnetwork or wider institutional context determinesadoption. Determining the network position of amajor pharmaceutical firm turns out to be an excep-


tionally difficult undertaking20 and, while we aresure that the wider institutional context plays a rolein shaping the adoption of science-driven drugdiscovery, distinguishing between the effects ofinstitutional isomorphism or management fashionand the simple effects of time and convergenceturns out to be very difficult. We also do notaddress the extent to which governance structuresand incentive systems at the most senior levels ofthe firm affect adoption decisions. While this is anintriguing hypothesis, preliminary analysis high-lighted both the difficulty of constructing consistentmeasures of governance problems across nationalboundaries and the fact that they appeared to beonly very weakly correlated, if at all, with thefirm’s choice of research technique.

RESULTS

In this section we discuss the results from twosets of regression models. The first set focuseson the ‘initial conditions’ and convergence model.The second set brings in the measures of environ-mental heterogeneity.

Our analysis begins by considering a simplemodel of the adoption of science-driven drugdiscovery which focuses exclusively on thedynamic impact of ‘initial conditions’:

yt = a + bt + gy0 + dy0·t + et

This equation is estimated using each of ouralternate measures of the extent of diffusion ofscience-driven drug discovery (with the appropri-ate functional forms for each measure). Table 4presents the results, which we find quite striking.In the first place, they show that at any point intime the degree to which a firm has adopted thetechniques of science-driven drug discovery is to

20 To begin with, there is the question of which network tofocus on. In the context of a single industry study, it doesnot make sense to focus on the degree to which there isinterlock across the board of directors, for example, sincewithin a single industry directors are forbidden from servingon multiple boards to avoid potential conflicts of interest. Weexplored the degree to which we could use patterns of jointpublication to construct a plausible network, but discoveredthat constructing such a network requires the collection (andcoding) of data about hundreds of thousands of papers andwould result in an intractably large network containing thou-sands of nodes. We are currently exploring whether we canusefully construct such a network by tracing the movementof senior executives across the industry.


Table 4. Initial conditions and convergence at determinants of the adoption of science driven drug discovery

Log-odds Ln Ln Log-odds(PROPUB) (PAPERS) (AUTHORS) (PUBFRAC)

Dependent variable: N 135 150 150 150

PROPUB0 1.00∗∗

(0.08)PROPUB0

∗ time 20.04∗∗

(0.01)Ln(PAPERS0) 1.02∗∗

(0.07)Ln(PAPERS0) ∗ time 20.05∗∗

(0.01)Ln(AUTHORS0) 1.04∗∗

(0.05)Ln(AUTHORS0) ∗ time 20.04∗∗

(0.01)PUBFRAC0 3.57∗∗

(0.71)PUBFRAC0

∗ time 20.29∗∗

(0.11)Intercept 23.87∗∗ 21.03∗ 20.85∗ 22.45∗∗

(0.61) (0.04) (0.40) (0.60)Time 0.18∗∗ 0.32∗∗ 0.35∗∗ 0.26∗∗

(0.03) (0.06) (0.05) (0.06)Ln (Fmsales) 0.21∗ 0.15∗∗ 0.10∗∗ 0.09

(0.10) (0.05) (0.03) (0.08)

Adj. R2 0.75 0.82 0.91 0.38

∗p,0.05; ∗∗p,0.01

a very great extent a function of the firm’s posi-tion at the beginning of the period. The estimatedcoefficient on the initial value of the dependentvariable is large and very significant in all of theregressions. Second, the highly significantnega-tive coefficient estimated for the interaction ofy0 with time provides evidence in favor of theconvergence hypothesis. The firms that were‘furthest away’ from best practice (with lowerinitial levels of the dependent variable) movedfastest to adopt. However, while convergenceseems to be an important aspect of these data,the coefficient estimates imply a relatively slowdiffusion process: 10 years or more for the lag-gard firms to catch up with the leaders.

These results are important. They illustrate con-cretely a model of competitive advantage that isimplicit in much of our teaching and researchbut that is rarely demonstrated. Understanding theorigins of competitive advantage requires a focuson two distinct processes:creation, the meanswhereby first adopters develop and exploit newtechniques or strategies; andimitation, whereby


laggard firms respond to their unfavorable posi-tions and move to imitate market leaders.

Note that at one level these results are consis-tent with a Stinchcombian view of the sourcesof competitive advantage. Early movers, who hap-pen to be endowed with the ‘right’ set of organi-zational practices, reap the returns from being inthe right place at the right time. Laggards manageto change, but changing does not create advan-tage, it merely re-levels the playing field. Atanother level, though, convergence provides evi-dence for a more ‘strategic’ perspective on thedynamics of competitive advantage. Competitiveadvantage is as much about responding to unfav-orable positioning as it is about exploiting thoseopportunities which present themselves. As such,aggressive imitation by those who are initiallypositioned most unfavorably is consistent witha process whereby thoughtful managers activelyinterpret their environment and use their discre-tion to implement astrategic response.

Table 5 builds on this analysis and developsevidence relating to the full diffusion model fol-


Table 5. Initial conditions and contemporaneous heterogeneity. Dependent variable: log-odds (PUBFRAC), 150observations throughout

(5.1) (5.2) (5.3) (5.4) (5.5)

PUBFRAC0 3.57∗∗ 7.11∗∗

(0.71) (0.92)

PUBFRAC0∗ time 20.29∗∗ 20.79∗∗

(0.11) (0.15)

Scientist–CEO(t21) 0.84 0.87 0.33(0.53) (0.54) (0.46)

Scientist–CEO(t21)∗ time 20.07 20.07 20.01

(0.13) (0.13) (0.11)

Share of cardiovasculars in firm 20.00 0.00 0.05∗∗

sales(t21) (0.01) (0.01) (0.01)

Share of cardiovasculars in firm 0.00 0.00 20.006∗∗

sales(t21)∗ time (0.00) (0.00) (0.001)

Distance from science(t21) 0.20 0.29 20.60∗

(0.26) (0.27) (0.25)

Distance from science(t21)∗ time 20.01 20.02 0.08

(0.04) (0.04) (0.04)

Share of cardiovasculars in firm 0.01 0.00 20.03patents(t21) (0.03) (0.03) (0.03)

Share of cardiovasculars in firm 0.00 0.00 0.01patents(t21)

∗ time (0.01) (0.01) (0.01)

Share of U.S. cardiovascular 20.03 20.03 20.12∗∗

market(t21) (0.02) (0.02) (0.02)

Share of US cardiovascular 0.004∗ 0.01∗ 0.02∗∗

market(t21)∗ time (0.002) (0.00) (0.00)

Intercept 22.45∗∗ 21.22∗∗ 21.43∗ 21.52 24.99∗∗

(0.60) (0.56) (0.74) (0.83) (0.83)

Time 0.26∗∗ 0.08∗∗ 0.04 0.03∗∗ 0.44∗∗

(0.06) (0.02) (0.07) (0.07) (0.10)

Ln(total firm salest) 0.09 0.21∗ 0.23∗ 0.24∗ 0.22∗

(0.08) (0.09) (0.11) (0.12) (0.10)

Adj. R2 0.38 0.28 0.28 0.33 0.51

∗p,0.05; ∗∗p,0.01

lowing Equation 1. Each column includes either(a) only variables associated with initial con-ditions (5.1), (b) only variables associated withtime-varying environmental heterogeneity ((5.2)through (5.4)), or (c) both types of variables(5.5). For each regression, the dependent variableis PUBFRAC.21 The environmental variables are

21 We have experimented extensively both with PUBFRACas well as our other measures of the intensity of science-driven drug discovery (PUBFRAC, PAPERS, and


lagged by one period to allow time for firms toadjust their level of PUBFRAC, and eachregression includes both a time trend and the logof total firm sales as control variables.

After replicating the ‘initial conditions plusconvergence’ result in (5.1) for easy reference,

AUTHORS). Overall, the key results as they relate to dis-tinguishing between alternative sources of competitive advan-tage are robust.


we present our initial analysis of the impactof environmental heterogeneity by dividing theexplanatory variables into two groups correspond-ing to ‘behavioral’ and ‘economic’ hypothesesabout the drivers of adoption. The two ‘behavior-al’ variables are Scientist–CEO and the within-firm share of cardiovascular drugs, which areincluded by themselves in (5.2). The ‘economic’variables are those measuring the firm’s distancefrom science, its knowledge capital, and its com-petitive positioning, which are included by them-selves in (5.3). (5.4) includes both the behavioraland economic variables together but still does notinclude initial conditions. These results are quiteweak. Among the 11 explanatory variables(including firm sales), the only (marginally) sig-nificant coefficients are associated with firm salesand the interaction between the firm’s marketposition in cardiovasculars and TIME; the smallpositive parameter suggests faster ‘catch-up’ forfirms with a larger share of the cardiovascularmarket.

However, we believe that each of the first fourregressions in Table 5 are likely misspecified.For example, by excluding the impact of initialconditions in (5.2) through (5.4), we introduce agreat deal of noise into the regression, likelylimiting our ability to distinguish the ‘true’ impactof the environmental drivers. Model (5.5) there-fore includes both the initial conditions andenvironmental heterogeneity variables, with a dra-matic impact on the explanatory power of theregression and on the coefficient estimates. Forexample, the regressions suggest that, after con-trolling for initial conditions and convergence (thecoefficients for which remain large and significantas in earlier regressions), those firms whose saleswere concentrated in cardiovascular therapieswere associated with higher levels of PUBFRAC(though their rate of adoption was slower) andfirms more distant from scientific centers areassociated with lower levels of PUBFRAC. Thosefirms that had a larger share of the cardiovascularmarket were also faster to adopt the new tech-niques (though the impact on the level is negative(in contrast to our initial hypotheses)). Finally,the measures of Scientist–CEO and knowledgecapital (share of firm patents associated with car-diovascular therapies) remain insignificant, per-haps due to the noisy nature of each of theseproxy measures and their imperfect correlationwith the underlying concepts they are associated


with. More generally, (5.5) suggests that theinclusion of variables to incorporate the impactof initial conditions and convergence allows fora more precise and systematic evaluation of theimpact of environmental heterogeneity on the dif-fusion process, as well as providing an importantconfirmation for our earlier results about theimpact of initial conditions and convergence.

The alternative specifications offered in Table5 also provide some evidence about the relativeimportance of initial conditions vs. environmentalheterogeneity, at least in terms of goodness-of-fit. Of course evaluating the differences in R2

from each regression is of only limited valuesince it cannot provide evidence for how a changein one of the explanatory variables changes theexpected value of the dependent variable. How-ever, this comparison does suggest two things inthe current context. First, the regressions whichfocus on the firm’s initial conditions explain asignificantly higher share of the overall variancethan the regressions which evaluate environmentalheterogeneity by itself. Second, and perhaps moreimportantly, initial conditions and environmentalheterogeneity are complementary in their explana-tory power—there is over a 33 percent increasein adjusted R2 in the combined model (5.5) com-pared with either (5.1) or (5.4).

CONCLUSIONS AND IMPLICATIONSFOR FURTHER RESEARCH

These results are intriguing. At face value, Table5 appears to offer support for Stinchcombe’scritique, in that initial conditions play a veryimportant role in explaining the diffusion ofscience-driven drug discovery.22 However, bydeveloping a dynamic model of adoption whichallows for both convergence and changeassociated with local environmental cues, ourresults suggest that conscious strategic adjustmentis also quite important for understanding the

22 One possibility, of course, is that y0 is capturing not thecharacteristics of the firm’s founding, but a complex(unobserved) history of strategic adjustment. As a very pre-liminary cut at whether this is plausibly the case, we comparedthe usefulness of our measures of heterogeneity as measuredat the beginning of the period,Z0, as explanatory variablesfor y. The regression performed quite poorly, with an evenlower proportion of the variance in the dependent variableexplained than in the regressions that included contempor-aneous measures of heterogeneity.


dynamics of competitive advantage: convergenceis a key factor in explaining the overall patternof adoption, many of the ‘strategic’ variables aresignificant after controlling for initial conditions,and the explanatory power of the diffusion modelsubstantially improves when all of the effects areincluded simultaneously.

In other words, our results imply that—at leastin this context—Stinchcombe’s hypothesis ispowerful but not all encompassing. One of themost striking features of our data isconvergence:firms that were initially ‘behind’ at the beginningof the period move more rapidly to ‘catch up’ totheir more advanced competitors. Whileassociated with the same ‘initial conditions’ thatStinchcombe emphasizes, the process of conver-gence and imitation is nonetheless interestinglydifferent from a simple model of competition inwhich competitive advantage reflects the failureor exit of those firms that are not initially wellpositioned. Our results also suggest that, aftercontrolling for initial conditions, variables tra-ditionally associated with ‘strategic intent’ areimportant determinants of the level and rate ofadoption. While Stinchcombe’s hypothesis is wellworth taking seriously (and we would encouragefuture researchers to be careful to control for it),it seems to us that there is sufficient evidencehere to suggest that firms can and do change inpurposeful ways in response to exogenous shocks.For example, convergence suggests that managerswho are initially associated with poorly positionedfirms may be the most proactive in altering theirpractices and strategies. In other words, thereseems to be a clear role for understanding theprocesses of strategic adjustment which occur inresponse to the internal and external environmen-tal cues. In this sense, our framework and resultsprovide evidence consistentboth with Stinch-combe’s hypothesis that competitive advantage isimportantly driven by exogenous initial variationand with the more ‘rational’ or ‘strategic’ per-spective that views firms as responding with fore-sight to changes in their environment.

Of course, as we have emphasized several times,our empirical results are quite preliminary, andperhaps raise as many questions as they answer.Indeed the statistical analysis is largely descriptive,and we have tried to highlight throughout thepaper some of the difficult and interesting statisticalquestions that are raised by the attempt to dis-tinguish among the central hypotheses. Our paper


also highlights the need for richer longitudinal data.One could plausibly argue, for example, that thesignificance of initial conditions in our results issimply a consequence of ‘left truncation:’ that ifwe knew more about the history of the firms inour sample—about the origins of each firm’s ‘initialconditions’— we might be able to uncoveradditional evidence for the importance of mana-gerial vision and action in shaping competitiveadvantage. For example, we know from a numberof early histories of the industry that pharmaceuticalfirms differed enormously in the extent to whichthey invested in basic or scientific research in the1940s and 1950s. Some firms, such as Parke-Davis, Lilly, and Merck, made early and extensivecommitments to pure research, while other firmsdelayed making these investments for many years(Parascandola, 1985). Perhaps extending our empiri-cal study of this industry further back in timewould allow us to conclude that the adoption ofscience-driven drug discovery stemmed from far-sighted decisions made decades before the practicebecame widespread.

This suggestion brings us back to the discussionwith which the paper began. During the 1980s,firms that used science-driven drug discovery tech-niques outperformed those that did not. As such,one might be tempted to conclude that adoptingthis practice was a ‘good’ strategy that led, throughthe creation of unique capabilities, to significantcompetitive advantage in the sense of superiorability to generate new drug products. We have,ourselves, recommended to pharmaceutical firmsthat they adopt these techniques and the managersof many firms have by and large agreed with us.But one of the key conclusions to be drawn fromour analysis here is that positively responding tosuch a recommendation is only half the battle.Incorporating best practice is necessary, but imi-tation based on the experience of rivals can usuallyonly level the playing field, and in this case, atleast, imitation took many years. While key aspectsof the adoption process—in particular firms’ rateof convergence to best practice—reflected purposive‘strategic’ responses to a changing environment,these responses were greatly constrained by factorsthat were in place long before many of these firmseven began to think seriously about the issue. TheRBV has emphasized the importance of long-lived‘sticky’ assets, and the powerful influence of his-torical circumstances seen here indicates that thesecan be very long lived and very sticky indeed.


The advent of science-driven drug discoveryhad a dramatic impact on the innovation processin the pharmaceutical industry. The basis of com-petitive advantage was changed in important ways,and firms in the industry responded to this shockby making appropriate changes to their internalstructure, and by investing in the necessary capa-bilities and resources. But some firms entered thisperiod of transition much better equipped thanothers to exploit the potential of the new tech-niques. Were they lucky or were they smart? Thedifficulty of answering this question empiricallypoints to a deeper research agenda. We believethat a fuller appreciation of the origins of competi-tive advantage is to be found in a better analyticaland empirical understanding of the types of mana-gerial processes which allow some firms to beahead, and stay ahead, of the game.Ex post, it isclear that some firms actively identify, interpret,and act upon early signals from their internal andexternal environment, and so position themselvesto effectively exploit these opportunities well inadvance of others’ demonstration of the pay-offfrom the strategies which emerge later on as ‘bestpractice.’ These firms arecreating new sources ofcompetitive advantage. Understanding how theyorganize ex ante to do this is, in our view, acentral question for strategy research.

ACKNOWLEDGEMENTS

This study was funded by POPI, the Program forthe Study of the Pharmaceutical Industry at MITand by the MIT Center for Innovation in ProductDevelopment under NSF Cooperative AgreementNumber EEC-9529140. Stern completed this pa-per while a Health and Aging Fellow at theNational Bureau of Economic Research. Theirsupport is gratefully acknowledged. Jeff Furmanand Pierre Azoulay provided outstanding researchassistance, and we would also like to thank BobGibbons, Connie Helfat, David Hounshell, ZviGriliches, and participants at the Dartmouth con-ference focused on this special issue for theirmany helpful comments.

REFERENCES

Andrews KR. 1971.The Concept of Corporate Strategy.Dow–Jones Irwin: Homewood, IL.


Arellano M, Bond S. 1991. Some tests of specificationfor panel data: Monte Carlo evidence and an appli-cation to employment equations.Review of Eco-nomic Studies58(194): 277–320.

Barnett WP, Hansen MT. 1996. The Red Queen inorganizational evolution. Strategic ManagementJournal. 17: 139–158.

Barney J. 1991. Firm resources and sustained competi-tive advantage.Journal of Management17(1): 99–120.

Besanko D, Dranove D, Shanley M. 2000.Economicsof Strategy. 2nd edn, Wiley: New York.

Bower JL. 1974.Managing the Resource AllocationProcess. Harvard Business School Press: Boston,MA.

Brandenberger A, Nalebuff B. 1998.Co-opetition. Har-vard Business School Press: Boston, MA.

Burgelman RA. 1994. Fading memories: a processstudy of strategic business exit in dynamic environ-ments.Administrative Science Quarterly39(1): 24–56.

Chandler A. 1962.Strategy and Structure. MIT Press:Cambridge, MA (reprinted January 1990).

Christensen CM, Bower JL. 1996. Customer power,strategic investment, and the failure of leading firms.Strategic Management Journal17(3): 197–218.

Clark KB, Fujimoto T. 1991.Product Development inthe World Automobile Industry. Harvard BusinessSchool Press: Boston, MA.

Cockburn IM, Henderson RM. 1998. Absorptivecapacity, coauthoring behavior, and the organizationof research in drug discovery.Journal of IndustrialEconomics46(2): 157–182.

Cockburn IM, Henderson RM, Stern S. 1999a. Balanc-ing incentives: the tension between basic and appliedresearch. NBER working paper no. 6882, January.

Cockburn IM, Henderson RM, Stern S. 1999b. Thediffusion of science driven drug discovery: organi-zational change in pharmaceutical research. NBERworking paper no. 7359, December.

Davis G. 1991. Agents without principles? The spreadof the poison pill through intercorporate networks.Administrative Science Quarterly36: 583–613.

Eisenhardt KM. 1988. Agency- and institutional-theoryexplanations: the case of retail sales compensation.American Management Journal31(3): 488–511.

Eisenhardt KM, Schoonhoven CB. 1990. Organizationalgrowth: linking founding team, strategy, environ-ment, and growth among U.S. semiconductor ven-tures, 1978–1988.Administrative Science Quarterly35: 504–529.

Eisenhardt KM, Tabrizi BN. 1995. Accelerating adap-tive processes.Administrative Science Quarterly40(1): 84–110.

Fletcher K, Huff AS. 1990. Strategic argument map-ping: a study of strategy reformulation at AT&T.In Mapping Strategic Thought, Huff AS (ed). Wiley:New York; 165–194.

Fudenberg D, Tirole J. 1985. Pre-emption and rentequalization in the adoption of new technology.Review of Economic Studies52: 383–401.

Gambardella A. 1995.Science and Innovation in the


US Pharmaceutical Industry. Cambridge UniversityPress: Cambridge, UK.

Ghemawat P. 1991.Commitment: The Dynamics ofStrategy. Free Press: New York.

Gilbert RJ, Newbery DM. 1982. Pre-emptive patentingand the persistence of monopoly.American Eco-nomic Review, 514–595.

Grabowski HG, Vernon JM. 1990. A new look at thereturns and risks to pharmaceutical R&D.Manage-ment Science36(7): 804–821.

Griliches Z. 1957. Hybrid corn: an exploration in theeconomics of technical change.Econometrica25(4):501–522.

Griliches Z. 1986. Data problems in econometrics. InHandbook of Econometrics, Intriligator M, GrilichesZ (eds). North-Holland: Amsterdam; 1466–1514.

Hannan MT, Freeman JH. 1989.Organizational Ecol-ogy. Harvard University Press: Cambridge, MA.

Heckman JJ. 1991. Identifying the hand of the past:distinguishing state dependence from heterogeneity.American Economic Review81(2): 75–79.

Henderson RM. 1993. Underinvestment and incompe-tence as responses to radical innovation: evidencefrom the semiconductor photolithographic alignmentequipment industry.Rand Journal of Economics24(2): 248–270.

Henderson RM. 1994. The evolution of integrativecompetence: innovation in cardiovascular drug dis-covery. Industrial and Corporate Change3(3):607–630.

Henderson RM, Clark KB. 1990. Architectural inno-vation: the reconfiguration of existing producttechnologies and the failure of established firms.Administrative Science Quarterly35: 9–30.

Henderson RM, Cockburn IM. 1994. Measuring com-petence? Exploring firm effects in pharmaceuticalresearch. Strategic Management Journal, WinterSpecial Issue15: 63–84.

Henderson RM, Cockburn IM. 1996. Scale, scope andspillovers: the determinants of research productivityin drug discovery.Rand Journal of Economics27(1):32–59.

Henderson RM, Orsenigo L, Pisano G. 1999. Thepharmaceutical industry and the revolution in mo-lecular biology: interactions among scientific, insti-tutional and organizational change. InThe Sourcesof Industrial Leadership, Mowery D, Nelson R (eds).Cambridge University Press: New York; 267–311.

Hicks D. 1995. Published papers, tacit competenciesand corporate management of the public/privatecharacter of knowledge.Industrial and CorporateChange4(2): 401–424.

Holbrook K, Cohen WM, Hounshell DA, Klepper S.2000. The nature, sources and consequences of theearly history of the semiconductor industry.StrategicManagement Journal. 21(10–11): 1017–1041.

Jaffe AB. 1986. Technological opportunity and spillo-vers of R&D: evidence from firms’ patents, profitsand market value.American Economic Review76(5):984–1001.

Karshenas M, Stoneman P. 1993. Rank, stock, orderand epidemic effects in the diffusion of new processtechnologies: an empirical model.Rand Journal of


Economics24(4): 503–528.Koenig M. 1982. A bibliometric analysis of pharmaceu-

tical research.Research Policy12: 15–36.Lieberman MB, Dhawan R. 2000. Assessing the

resource base of U.S. and Japanese auto producers:a stochastic production frontier approach. Mimeo,UCLA Anderson School of Management, February.

Leonard-Barton D. 1998.Wellsprings of Knowledge:Building and Sustaining the Sources of Innovation.Harvard Business School Press: Boston, MA.

Levinthal DA. 1997. Adaptation on rugged landscapes.Management Science43: 934–950.

McGahan AM, Porter ME. 1997. How much doesindustry matter, really?Strategic Management Jour-nal, Summer Special Issue18: 15–30.

Mintzberg H. 1987. Crafting strategy.Harvard BusinessReview65(4): 66–75.

Mundlak Y. 1961. Empirical production function freeof managerial bias.Journal of Farm Economics43:44–56.

Murmann P, Tushman ML. 1997. Impacts of executiveteam characteristics and organization context onorganization responsiveness to environmental shock.In Technological Innovation: Oversights and Fore-sights, Garud R, Nayyar P, Shapira Z (eds). Cam-bridge University Press: New York; 260–278.

Nelson RR, Winter SG. 1982.An Evolutionary Theoryof Economic Change. Harvard University Press:Cambridge, MA.

Parascandola J. 1985. Industry research comes of age:the American pharmaceutical industry, 1920–1940.Pharmacy in History27(1): 12–21.

Penrose ET. 1959.The Theory of the Growth of theFirm. Wiley: New York.

Peteraf MA. 1993. The cornerstones of competitiveadvantage: a resource-based view.Strategic Manage-ment Journal14(3): 179–191.

Porter ME. 1980.Competitive Strategy. Free Press:New York.

Powell WW, Koput KW, Smith-Doerr L. 1996.Interorganizational collaboration and the locus ofinnovation: networks of learning in biotechnology.Administrative Science Quarterly41(1): 116–145.

Quinn JB. 1978. Strategic change: logical incremen-talism. Sloan Management Review20(1): 7–23.

Reinganum JF. 1981. Market structure and the diffusionof new technology.Bell Journal of Economics12:618–624.

Rumelt RP. 1991. How much does industry matter?Strategic Management Journal12(3): 167–185.

Selznick P. 1957.Leadership in Administration: ASociological Interpretation. Harper & Row: NewYork.

Shapiro C, Varian HR. 1998.Information Rules. Har-vard Business School Press: Boston, MA.

Stern S. 1999. Do scientists pay to be scientists? NBERworking paper no. 7410, October.

Stinchcombe AL. 1965. Social structure and organi-zations. In Handbook of Organizations, March J(ed). Rand-McNally: Chicago, IL; 142–193.

Stinchcombe AL. 2000. Social structure and organi-zations: a comment. InEconomics Meets Sociologyin Strategic Management: Advances in Strategic


Management, Baum J, Dobbin F (eds). JAI Press:Greenwich, CT.

Wernerfelt B. 1984. A resource-based view of the firm.Strategic Management Journal5(2): 171–180.

Zucker LG, Darby MR. 1997. Present at the revolution:transformation of technical identity for a largeincumbent pharmaceutical firm after the biotechnol-ogical breakthrough.Research Policy26(4): 429–447.

Zucker LG, Darby MR, Brewer M. 1998. Intellectualhuman capital and the birth of U.S. biotechnologyenterprises. American Economic Review88(1):290–306.

APPENDIX A

Data appendix: construction of our measuresof Z

Scientist–CEO

Scientist–CEO is a binary variable set to 1 ifthe firm’s CEO has a scientific background. Weconstructed this variable by systematically check-ing the Annual Report of each company in oursample for the title or background of the CEO.Those CEOs who were listed as having either aPhD or an MD were coded as scientists.

Distance from science

For any single paper, ISI lists the addresses ofeach author. From these data we constructed acount of the paper authorships by city for eachfirm for each year. We designated every city thatconstituted more than 2 percent of all the firm’sauthorships over the entire period a ‘research node’for the company, and obtained its longitude andlatitude. We then weighted each node by the frac-tion of authorships it hosted relative to all author-ships attributed to the company that year. Thesedata gave us a good sense of the changing geo-graphical location of the firm’s research over time.

In order to measure the degree to which thefirm was ‘close’ to publicly funded science weneeded to construct a weighted measure of thelocation of publicly funded research. This is dif-ficult to do without full data on public spendingby geographical location across the world, anddata for Europe and Japan are relatively difficultto obtain. As a first approximation we constructeda data set consisting of the addresses of allthose authors who coauthored papers with NIHresearchers over our period. We designated eachcity listed in these data that constituted over 2


percent of total NIH coauthorships an NIH node,and obtained its latitude and longitude. Weweighted these nodes using two schemes: as thefraction of NIH coauthorships that they obtainedin a given year and as a rank from 1 to 10 onthe basis of the scientific reputation of each node.

A firm’s ‘closeness to science’ was then calcu-lated as: Sijwi ∗ wj ∗ 1/dij, where dij is thedistance from each firm nodei to NIH nodej, wi

is the weight assigned to the company node andwj to each node.

Firm sales

Firm sales are measured as total pharmaceuticalsales in the United States, as reported to IMS.We excluded sales of OTC or ‘over-the-counter’medications that can be obtained without a pre-scription. Note that the United States is onlyabout 50 percent of the world pharmaceuticalmarket, and that many of the firms in our samplehave substantial global sales. U.S. sales are thusa rather noisy measure of total sales.

Knowledge capital: share of patents incardiovascular therapies

We used Derwent patent data to construct ameasure of the firm’s experience or ‘knowledgecapital’ in cardiovasculars. For all of the firmsin our data set, Derwent generously donated tous complete data about each patent family grantedto each firm over our period. Each patent familyincludes complete information about the patentsgranted to the firm across the entire world and acomplete set of Derwent ‘manual codes’.23 Foreach patent that had been granted in two out ofthree major world markets (Japan, the UnitedStates, and Europe) we used the ‘manual codes’to assign patents to particular therapeutic areas.We then calculated a ‘knowledge stock’ in thestandard manner assuming a 20 percent dep-recation rate.

23 The U.S. patent office classifies pharmaceutical patents largelyon the basis of chemical structure: a classification that containsvery little information about therapeutic intent. The Derwentmanual codes are assigned by specialists in the field who classifyeach patent on the basis of its therapeutic implications.


Market position: share of the cardiovascularmarket

We measure share of sales using detailed data atthe product level obtained from IMS America.Note that the pharmaceutical market is a globalone. The U.S. market represents roughly 50 per-cent of the world market, and nearly every firmin our sample has extensive global operations.Unfortunately our share measures will be dis-torted by the fact that the distribution of sales


across therapeutic classes isnot uniform acrossthe world.

Managerial attention: cardiovascular therapiesas a share of the sales portfolio

We measure cardiovascular sales as a share oftotal firm sales using IMS America data. Notethat it is subject to the same source of error asour measure of market position, above.

cockburn, henderson and stern 2000 smj - untangling competitive advantage

Documents

rms initial conditions

rms managers

performance differentials

heterogeneous performance

superior performance

convergence rms

measure differences

persistent differences