DEATH HURTS, BUT IT ISNÕT FATAL: THE POSTEXIT DIFFUSION …terpconnect.umd.edu/~rajshree/research/20 Hoetker... · logical knowledge, other firms m ay attempt to build on the knowledge

DEATH HURTS, BUT IT ISN’T FATAL:THE POSTEXIT DIFFUSION OF KNOWLEDGE CREATED BY

INNOVATIVE COMPANIES

GLENN HOETKERRAJSHREE AGARWAL

University of Illinois at Urbana-Champaign

There is little understanding of whether a firm’s innovative knowledge dies with it orif instead significant diffusion of knowledge occurs even after a firm exits an industry.Theoretical predictions about the differing effects of firm exit on private and publicknowledge and implications for interfirm knowledge transfer are forwarded. Weinvestigated main and moderating effects of a firm’s exit from the disk drive industryon knowledge diffusion to other firms, finding evidence that the ability to use a firm asa template plays a critical role in successfully replicating its knowledge. Absent thistemplate, knowledge “stickiness” reduces knowledge diffusion.

In 1999, despite millions of dollars of investmentand a portfolio of innovative technologies, flatpanel display manufacturer Optical InformationSystems (OIS) shut down operations, unable toachieve commercial success. Although OIS failed,its technology lived on. A letter by the firm’s formerdirector of advanced technologies to the editors ofthe magazine Information Display reported thateven after the firm’s exit, OIS technology continuedto make waves in the flat panel industry, withmany of its patents covering processes that becamemainstream technology. The letter then cited spe-cific innovations by other firms that had built onOIS breakthroughs. Although the firm had exitedthe industry in spite of its technological strength, itleft a lasting legacy for the industry’s technology.

Is the OIS story unusual, or does it highlight aregular occurrence? Since technological expertiseis an important determinant of firm success (Jo-vanovic & MacDonald, 1994; Teece, 1986), it couldbe argued that firms that exit an industry are typi-

cally lacking in this important area and thus havelittle impact on the technological progress in theindustry. This formulation would imply that firmslike OIS are outliers and that diffusion of theknowledge they create is generally low, both beforeand after they exit an industry. However, there isstrong evidence that many companies exit despitehaving developed innovative knowledge (Golder &Tellis, 1993; Katz & Shapiro, 1985) and that a lackof complementary assets (Teece, 1986) often resultsin firms’ untimely deaths. If this is the case, thenfirm exit will not be perfectly, negatively correlatedwith technological superiority. To the extent thatsome firms exit in spite of having created techno-logical knowledge, other firms may attempt tobuild on the knowledge created by departed firms.

Although the issue of whether other firms subse-quently capitalize on knowledge created by compa-nies that exit an industry remains underresearched,it is important to investigate for several reasons.First, 8 to 10 percent of all companies leave anindustry in an average year (Agarwal & Gort, 1996),but their exit may nonetheless create economicbenefits and impact social welfare (Dunne, Roberts,& Samuelson, 1988; Knott & Posen, 2005). Many ofthese companies may have been technologicallyinnovative and are thus underexploited sources oftechnological progress and increases in social wel-fare. Further, in some industries, substantial publicinvestment may have been made in these compa-nies, through either tax incentives or direct fund-ing. Does the value of that investment depend onthe commercial success of the firm receiving thefunding, or can other firms that remain commer-cially viable subsequently harness the resultinginnovation?

Both authors contributed equally. The research wassupported by grants received from the E. Marion Kauf-mann Foundation and the Campus Research Board, Uni-versity of Illinois. The manuscript has benefited fromcomments received from Editor Sara Rynes, three anon-ymous reviewers, Juan Alcacer, Joel Baum, Raj Echam-badi, Dan Levinthal, MB Sarkar, Michael Tushman, An-drew Van de Ven, Charles Williams, and seminarparticipants at the 2005 Academy of Management meet-ings, the University of Chicago, the 2004 Harvard Entre-preneurship and Innovation Conference, the Universityof Illinois, and the 2005 Wharton technology miniconfer-ence. We thank April Franco for access to data, andPersephone Doliner for her copy-editing services. Theusual disclaimer applies.

! Academy of Management Journal2007, Vol. 50, No. 2, 446–467.

446Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s expresswritten permission. Users may print, download or email articles for individual use only.

Theoretically, since the issue relates to interfirmknowledge transfer under the most challengingconditions, it offers an opportunity to examine theissues that may be relevant when firms seek tocapitalize on other firms’ technologies. The tradi-tional view of knowledge highlights the positiveexternalities inherent in knowledge creation andthe nonrival, nonexcludable nature of information,particularly when it is embodied in patents (Arrow,1962; Griliches, 1979; Jaffe, 1986). Patents repre-sent codified knowledge that has been publicly re-vealed through the publication of patent docu-ments, thus enabling the use of the knowledge byfirms other than the originators (Jaffe, 1986;Spence, 1984). In contrast, an alternative view em-phasizes that knowledge may have private as wellas public aspects (Nelson & Winter, 1982). Theseprivate aspects (Nelson & Romer, 1996) impartknowledge “stickiness” (von Hippel, 1994), a con-sequence either of the embeddedness of innova-tions in organizational routines and teams (Martin& Mitchell, 1998; Nelson & Winter, 1982) or ofcausal ambiguity (Lippman & Rumelt, 1982;Rumelt, 1984). Restriction of interfirm knowledgetransfer is the outcome. Researchers have foundthat significant tacit knowledge resides within thesocial structures of organizations, since innovationis the result of concerted and directed efforts byentire teams of employees.

Our paper builds on the complementarity of theprivate and public components of knowledge (Nel-son, 1990) to examine how the lack of accessibilityof private knowledge affects subsequent diffusionof the public knowledge embodied in a firm’s pat-ents. Exit means both loss of the private knowledgeembodied in a firm and loss of the possibility ofusing the firm’s activities as a template (Winter &Szulanski, 2001). Examining the effect of firm exiton knowledge diffusion can thus shed light on theimportance of private knowledge as a facilitator ofthe diffusion of public knowledge. Importantly, weaddress the competing explanation that firm exitmay represent a lack of relevance of the knowledgeand develop hypotheses for the interaction of firmexit with variables associated with the greater pres-ence of private knowledge. Thus, we contribute tothe literature on the extent of knowledge spilloversbetween firms by discussing how private knowl-edge may serve as a boundary condition for thepublic knowledge a firm creates. Our examinationof the postexit diffusion of knowledge also comple-ments studies of the importance of geographicallocation (Agrawal, Cockburn, & McHale, 2003; Au-dretsch & Feldman, 1996) and employee mobility(e.g., Rosenkopf & Almeida, 2003) for accessing

private knowledge and reducing the tacitness andstickiness of knowledge (von Hippel, 1994).

To the best of our knowledge, no study has sys-tematically examined the impact that the exit of afirm has on the diffusion of the knowledge it hascreated. The empirical setting of our study is thehard disk drive industry, because of its technolog-ical intensiveness and the availability of the datanecessary to examine our research questions(Christensen, 1993). We define firm exit, or“death,” as a firm’s having ceased operations in thedisk drive industry, excluding firms that were ac-quired. We ensure that for diversified firms, exitfrom the industry was concomitant with a cessationof their innovative activity related to the industry.We also investigate the possibility that firms thatexited were insignificant in the development ofhard drive technology and do not find this to be thecase. Using patent citations as a measure of knowl-edge diffusion, we examine the effects of firm exitnot only on the overall patent-citation life cycle,but also on the relationship between characteristicsof an innovation and its diffusion to other firms.

We find support for our hypothesis that exit im-pairs the ability of other firms to draw on theknowledge generated by a firm; firm exit results ina significant decline in citations received by a focalpatent. Further, we show that firm exit interactswith variables associated with more embeddednessof knowledge in a firm’s private routines (firm ageat time of patenting, degree to which an innovationbuilt on the innovating firm’s internal knowledgebase, and number of inventors) to have a negativeimpact on the patent’s citations. As a result, wefind broad support for our hypotheses that thehigher the private component of a firm’s knowl-edge, the more pronounced is the negative impactof the firm’s exit on subsequent citations. However,the firm’s exit—the death of its industry-relatedactivity—does not halt all further use of its tech-nology, and the effect of exit on subsequent cita-tions attenuates over time. Thus, firms that exit anindustry provide spillover benefits to others, inkeeping with the findings of Knott and Posen(2005).

THEORY

Private and Public Components of Knowledge

The idea that as firms pursue new knowledge,they create a public good dates back to Arrow(1962). Subsequent work in the area has discussedthe implications of the nonrival and nonexcludableproperties of knowledge for its subsequent diffu-sion, since both aspects increase the likelihood of

2007 447Hoetker and Agarwal

another firm benefiting from the knowledge createdby a focal firm. Because investments in knowledge-creating activities by a firm also increase the hu-man capital of its employees (Becker, 1964), em-ployee mobility has been identified as a keymechanism for knowledge diffusion (Almeida &Kogut, 1999), though knowledge diffusion can alsooccur through other mechanisms, including codifi-cation, reverse engineering and scientific reproduc-tion, and formal or informal interpersonal contacts(Arrow, 1996).

However, much attention in the last 20 years hasalso been paid to the tacit aspect of knowledge,particularly that which is team-based and sociallyembedded in firm routines (Nelson & Winter,1982). Highlighting the fact that not all the innova-tive knowledge firms create is public, Nelson(1990) argued that firms generate innovativeknowledge by combining generic, public knowl-edge with specific designs and practices that areprivate and known only to their creators. The suc-cess of other firms in replicating and building onthe knowledge created by a firm thus depends ontheir ability to understand the private knowledgewithin which the public knowledge is embedded(Rosenberg, 1982).

The private aspect of knowledge results inknowledge “stickiness” (von Hippel, 1994) due tocausal ambiguity and the embeddedness of innova-tions in individual human capital (Becker, 1964),and organizational or team-based rules and rou-tines (Lippman & Rumelt, 1982; Nelson & Winter,1982; von Hippel, 1994; Szulanski, 1996). For ex-ample, causal ambiguity, the “basic ambiguity con-cerning the nature of the causal connections be-tween actions and results,” impedes duplicatingand extending another firm’s innovative knowl-edge (Lippman & Rumelt, 1982: 420). It may beunclear which of the multiple research efforts thata firm engaged in ultimately led to its innovativesuccess. This lack of clarity may in part be becausethe knowledge resides at different levels within thefirm, including individual inventors, researchteams, and routines for combining complementaryresources. In addition to occurring at different lev-els, the private knowledge may vary in nature overorganizational levels. An individual inventor maypossess tacit knowledge about the underlying sci-entific basis of an innovation. The ability to managethe complexities of interactions within a team islikely to reside largely within team routines, whichno single individual may understand completely.The overall research and product line trajectory ismore likely to reside at the level of the firm. Also,the relative importance of private knowledge atdifferent levels is contingent on the nature of a

particular innovation. For example, for a leading-edge technology, the tacit knowledge of individualsmay be paramount, but for an innovation requiringa large team of inventors, the routines of the inno-vating team may be dominant.

Since most innovations embody private knowl-edge at multiple levels, ambiguity results regardingthe conditions under which their technologies canbe gainfully applied (Nelson & Winter, 1982). Itmay also be difficult to judge the potential value ofan innovation (Podolny & Stuart, 1995). Greaterambiguity on each of these dimensions limits thedegree to which a firm other than the source firmcan build on an innovation, even if the other firmhas access to the public component of the relevantknowledge.

Thus, the received literature suggests that privateand public knowledge are complementary requi-sites for the creation of new knowledge; in order tounderstand the private aspects of another firm’sinnovative knowledge, a firm must overcome theassociated embeddedness and causal ambiguity. Itcan attempt to do so by undertaking its own re-search efforts to build the required understandinginternally (Cohen & Levinthal, 1990). However, vi-carious learning—learning from the experience ofothers through observation (Cyert & March,1963)—is likely to be less costly than reinventingand learning experientially (Schulz, 2003). Sincetransferring knowledge often requires access totacit organizing principles that are not easily artic-ulated, the opportunity to consult a working exam-ple can be very valuable (Winter, 1987). As Winterand Szulanski wrote, “The recreation of a complex,imperfectly understood, productive routine is oftena protracted process that involves many referencesto an existing working model” (2001: 742). Thisstatement is consistent with Haunschild and Min-er’s (1997) finding that firms faced with uncertaintechnology rely on observing the organization thatis the source of the technology for clues on how toorganize and act. In essence, a source firm’s rou-tines and subsequent actions serve as a template forthose wanting to emulate its innovative activities.Interacting with or observing the source firm en-ables understanding which innovative trajectorieswere considered important to pursue, and whatassociated research efforts were subsequently em-phasized or dropped. Other firms also gain valu-able insights on how to manage roadblocks thatarise in advancing an innovation (Almeida &Kogut, 1999). Observing what innovations eventu-ally become commercial products provides a wayto evaluate the commercial potential of an innova-tion (Arrow, 1996). Thus, observing an innovatingfirm’s subsequent actions helps other firms in de-

448 AprilAcademy of Management Journal

termining the level(s) at which the private knowl-edge resides, deciding what innovative knowledgeis worth replicating and extending, assessing thehurdles, and assessing the directions to follow dur-ing replication and extension. The importance ofsuch direct or indirect interaction with an innova-tive firm has been well established in the vast lit-eratures on learning in alliances (Dyer & Singh,1998; Gulati, 1998), social networks (Burt, 1992;Granovetter, 1985), and knowledge spillovers viageographical proximity (Alcacer & Gittelman,2006).

Effect of Firm Exit on Knowledge Diffusion

The preceding discussion emphasizes the impor-tance of the continued existence of innovativefirms for the diffusion of their knowledge—thefirms themselves serve as templates, because theirroutines embody the interaction of the private andpublic components of their knowledge. Thus, justlike any artifact, whether a hammer or a computer,embodies knowledge that new producers of similarartifacts can use (Cowan, David, & Foray, 2000), afocal firm’s existence and activity represent embod-ied knowledge that subsequent developers can relyupon while building on their own knowledge. Wenow argue that the exit of a firm removes the pos-sibility of direct or indirect interaction with thefirm as a whole, thus limiting the extent to whichother firms can capitalize on its knowledge, even ifits employees and the codified knowledge areavailable.

In developing our hypotheses on the effect of itsexit on the diffusion of a firm’s knowledge, we notetwo issues. First, we deliberately focus on knowl-edge that is already codified and information avail-able to other firms via patents. Patent data providea stringent environment within which to test theimportance of private knowledge and firm exis-tence. If the private knowledge of a firm is not animportant complement to the explicit/codifiedknowledge available within patents, then firm exitshould have no appreciable impact on the rate atwhich other firms use and cite the patented knowl-edge. Second, we focus on source firm characteris-tics only, and not on recipient firms’ capabilitiesand strategies for harnessing the knowledge thatmay affect their absorptive capacity (Cohen &Levinthal, 1990). Thus, we are interested in “aver-age” postexit diffusion of knowledge and do notaddress differences among citing firms in their con-trol over complementary assets required to com-mercialize disk drive products, or in the relevanceor magnitude of their internal R&D efforts or hiringpractices.

The exit of a firm removes the opportunity toobserve and interact with the firm, which, as indi-cated above, is important for understanding theprivate aspects of the knowledge created by thefirm. Although access to the public good aspect ofthe knowledge remains (via reverse engineeringand reliance on codified knowledge), the firm’sactivities can no longer serve as a template for otherfirms seeking to build on its knowledge. After firmexit, the private knowledge that resides at levelsother than at the individual inventor level is likelyto undergo substantial disruption and loss. It maynot always be feasible to protect the private knowl-edge held at the team level, and the potential scat-tering of the firm’s innovative personnel to otherfirms may additionally complicate efforts to usesocial networks as a way of gathering informationon the firm. Even when all (or most) of a team areable to move en masse to another firm, they face thechallenge of functioning under a new managementand incentive system. At the firm level, it is notpossible, postexit, to observe how the innovatingfirm would have configured its complementary re-sources to build upon an innovation. Furthermore,since the firm’s commercialization efforts havestopped, other firms cannot use observations aboutwhat innovations eventually become commercialproducts to evaluate the commercial potential of aninnovation. The complementarity of the privateand public components of the knowledge a firm hascreated leads us to expect that its exit will reduceother firms’ ability to capitalize on its knowledge.

We note the possibility that, since technologicalcapabilities are positively related to firm survival,exiting firms represent lower levels of technologi-cal prowess. However, such a correlation wouldimpact the levels of citation received by a firm’spatents; there would be no reason to expect achange in the rate of citation before and after firmexit. There are two other reasons, though, that areconsistent with the observation of a postexit de-crease in firm citations. Patents often representstrategic behavior (Ziedonis, 2004), and it may beargued that firm exit reduces the threat of patentinfringement litigation. If the firm that created apatent is no longer around to defend the relevantintellectual property, the risk that litigation willoccur if subsequent patents omit its citation is re-duced. Given the market for intellectual property(Anton & Yao, 2002; Mann, 2005), other firms oftenacquire an exiting firm’s patent rights; thus it is notclear whether there is indeed a substantial declinein the risk of litigation. Finally, a firm’s knowledgemay lose relevance when it exits, either because ofexogenous shocks or the exit’s perceived signalvalue. Although we explicitly addressed the above


competing explanations both in our choice of em-pirical context and in our testing of our first hy-pothesis, we cannot discount the possibility thatthe decline in the citations associated with firmexit may be a result of a perceived reduction ineither the risk of litigation or the relevance of thepatented knowledge. Since all these reasons pointto a postexit decline, we hypothesize:

Hypothesis 1. Subsequent citation (use) of apatent by other firms in an industry is nega-tively impacted by the patenting firm’s exitfrom the industry.

We note, however, that if factors related to rele-vance or to risk of litigation, rather than to theaccessibility of private knowledge, are the truedrivers of our hypothesized decline in citationsafter firm exit, there should be no difference in therates of knowledge diffusion among characteristicsassociated with varying degrees of private and pub-lic components of knowledge. In the following sec-tion, we develop interaction hypotheses that enableus to isolate the role of accessibility of privateknowledge in determining postexit diffusion.

Interaction of Firm Exit with KnowledgeCharacteristics

The importance of the private knowledge held bya firm to the diffusion of its patented knowledgewill vary with the characteristics of an innovation.The greater the private component of knowledge,the greater will be the effect of firm exit on thesubsequent diffusion of knowledge. We examinethe interaction of exit with four variables that havebeen associated with the embeddedness of knowl-edge in a firm’s private routines: the age of theinnovating firm, the degree to which the innovationbuilt on the innovating firm’s internal knowledgebase, the number of inventors, and the diversity oftechnologies the innovation drew upon. Each vari-able influences the importance and/or accessibilityof the innovating firm’s private knowledge. Sincethe loss of the innovating firm as a template makesit more difficult to replicate the firm’s privateknowledge, we expect that exit will have a largernegative impact the more important or inaccessiblethe private knowledge was for that innovation. Wenow examine each variable in turn, exploring itsrelationship to the role of the private knowledgeassociated with innovations.

It is well established that the embeddedness ofinnovations in organizational routines increaseswith a firm’s age owing to greater formalization ofstructures and encoding of lessons in routines (Lev-itt & March, 1988; Nelson & Winter, 1982). A firm’s

core capabilities, particularly those related to tech-nology, are developed through learning and expe-rience, and this “path dependency” implies thatolder firms have higher stocks of private knowledge(Sorensen & Stuart, 2000). This is because olderfirms have gone through a longer process of learn-ing and have stored past learning in behavioralrules and routines (Dosi, Teece, & Winter, 1992;Nelson & Winter, 1982). Thus, it may be difficult tobuild on established firms’ capabilities as they aremore likely to be embedded in networks of in-trafirm relationships. Building on an older firm’sknowledge may require a recipient firm to observeor interact with the older firm more than would benecessary with a younger source firm, to learn bothits rules and routines and how its subsequent in-novations built on its earlier ones. Thus, the older afirm was at the time of a patent, the greater will bethe impact of the loss of the firm as a template.

A similar logic applies to innovations that resultfrom a firm building on its prior innovations (Jaffe& Trajtenberg, 2002). These innovations drawheavily on a firm’s internal knowledge base ratherthan on the knowledge of others and are said toreflect “localized search” (Anderson & Tushman,1990). They will therefore be closely bound withinthe routines and culture of the innovating firm(Nelson & Winter, 1982). Further, they are likely tobe couched in the idiosyncratic language of the firm(Arrow, 1974). As such, innovations that drawheavily upon a source firm’s internal knowledge basewill be highly tacit and difficult for others to imitateand extend, particularly after the exit of the sourcefirm. Again, we anticipate a larger postexit drop in thediffusion of an innovation if that innovation drewheavily on a firm’s internal knowledge base.

Since older firms and firms that draw on theirinternal knowledge bases will have more privateknowledge, we hypothesize:

Hypothesis 2. The older the patenting firm is atthe time of a patent application, the more neg-atively the firm’s exit impacts subsequent cita-tion (use) of that patent by other firms.

Hypothesis 3. The more related a patent is tothe patenting firm’s internal knowledge base,the more negatively the firm’s exit impacts sub-sequent citation (use) of that patent by otherfirms.

The larger the number of inventors associatedwith an innovation, the larger is the pool of mobileemployees upon the exit of the firm from the in-dustry. Indeed, a source firm’s employees may con-tinue to build on a technology once they join (orcreate) other firms, and this condition might lead


one to argue for a greater diffusion of knowledgeafter an exit. However, the greater the number ofinventors in a research team, the more numerousthe necessary interactions between individuals,and the more embedded the innovation in a com-plex web of relationships (Van de Ven, 1986).When a team of inventors is large, the range ofspecialized skills represented on it is also oftenlarge (Schilling, 2006; Valentin & Jensen, 2002).Such a large team represents not simply more in-teractions, but increasingly complex ones. Main-taining effective communication in a group whosemembers have diverse technical backgrounds is acomplex challenge (Pfeffer, 1981) requiring the de-velopment of routines and languages that spantechnical specializations.

Thus, the greater the number of inventors on ateam, the greater the degree of private knowledgethat is embedded in the team and its firm. Thisincrease in private knowledge increases the impor-tance of the continued existence of the knowledge-creating firm for other firms seeking to build on itsinnovations. Absent the routines of a departed firm,other firms and their individual inventors will, webelieve, have limited ability to replicate the exiter’sactivities. Further, the higher the number of inven-tors on a team, the more difficult it is for the entireteam to be hired or easily assimilated by anotherfirm. Thus, although individual employees may beable to leverage their knowledge at their new placeof employment, team- and firm-level privateknowledge may be more difficult to replicate. Over-all, we posit that a firm’s exit will have a strongerimpact on the diffusion of knowledge created by alarge team of inventors than it will have on thediffusion of knowledge created by a small team.Accordingly,

Hypothesis 4. The larger the team of inventorsa patent has, the more negatively the patentingfirm’s exit impacts subsequent citation (use) ofthat patent by other firms.

Similarly, innovations that draw upon a widerange of underlying technologies (e.g., organiclight-emitting diodes, which require expertise inelectronics, organic chemistry, and materials sci-ence) tend to be stickier than those that are exten-sions of a narrow field of knowledge, since theymay require exploration rather than exploitation(March, 1991). Knowledge that synthesizes diver-gent knowledge bases tends to be highly original(Trajtenberg, Henderson, & Jaffe, 1997), and combi-nations of multiple fields tend to occur at the tech-nological frontier. Knowledge surrounding suchbreakthrough research is likely to be highly tacit

and therefore hard for outsiders to imitate (Nelson& Winter, 1982). Further, just as tacit expertise isvital to the management of products with manyinteracting components (Chesbrough & Teece,1996), it is also important in the management ofresearch that draws on many interacting technolo-gies. Thus, direct interaction and vicarious learningshould be especially important for the diffusion oftechnologies that draw on a wide range of technol-ogies. This argument implies that firm exit willhave a greater detrimental impact on the subse-quent use of an innovation that embodies a widerange of technologies.

Hypothesis 5. The more diverse technologies apatent draws upon, the more negatively thepatenting firm’s exit impacts subsequent cita-tion (use) of that patent by other firms.

DATA

To address the research questions above, weneeded to examine knowledge diffusion acrossfirms for the census of corporations that entered(and exited) an industry. We tracked such knowl-edge diffusion as the subsequent use of a firm’stechnology by other firms via patent citations. Indoing so, we followed a large body of research thathas used the citations a patent receives as an indi-cation of the degree to which subsequent innova-tions have built upon it (Jaffe & Trajtenberg, 1996;Katila & Ahuja, 2002). The chief advantage of usingpatent data for our purposes was that these datareliably capture subsequent use of innovativeknowledge by other firms. An inventor who files apatent application is required by law to list all“prior art” of which she or he is aware. Unlikeacademic citations, these citations to earlier workhave the important legal function of limiting thescope of the property right granted to the patent.Further, the patent examiner in charge of the appli-cation, who is an expert in the technological area ofthe patent, can add citations that the inventor mayhave missed or concealed. This practice reducesthe probability that irrelevant patents will be citedor that relevant patents will be omitted. Not everycitation represents awareness of the cited patentwithin an organization filing the citing patent,since the patent examiner could have added thecitation (Alcacer & Gittelman, 2006; Cockburn, Kor-tum, & Stern, 2002); however, a variety of studieshave confirmed that patent citations are an accu-rate, though noisy, indicator of actual knowledgeflows (Jaffe, Trajtenberg, & Fogarty, 2002).


Context: The Disk Drive Industry

Given the data requirements of a study on knowl-edge diffusion before and after the exits of firms,the industry chosen for our empirical contextneeded to conform to certain boundaries. First, ithad to be relatively technologically intensive, be-cause technologically intensive industries havehigher rates of knowledge generation, and hencehigher rates of knowledge transfer. Second, weneeded longitudinal data on firms that were suc-cessful in the chosen industry and those that ulti-mately exited it. Third, although the industry hadto experience significant technological change, itneeded to have some stable underlying knowledgebase—that is, knowledge that continued to haverelevancy over time. We selected the hard diskdrive industry for our empirical context since itconformed to both the theoretical and empiricalrequirements of the study.

Disk drives are magnetic information storage de-vices used in computers. In 1973, IBM pioneeredthe 14-inch Winchester, the first completely sealedand removable disk drive, and the disk drive indus-try has since experienced rapid technological evo-lution (see Christensen [1993, 1997] for a detailedindustry history). The industry experienced signif-icant levels of both entry and exit in the relevantperiod, and it has followed the typical industry lifecycle of introduction, growth, shakeout, and matu-rity (Gort & Klepper, 1982). Since every productivefirm was included in our data, regardless of size,the data do not suffer from a survivor bias.1 Many ofthe entering firms represented employee entrepre-neurship and, thus, interfirm knowledge transfer(Agarwal, Echambadi, Franco, & Sarkar, 2004). Ad-ditionally, as McKendrick, Doner, and Haggard(2000) documented, both the employee mobilityand interfirm spillovers that shape new firms’ tech-nology and location choices are extensive in thedisk drive industry.

With regard to the pace of technological change,we knew that numerous architectural, modular,and incremental innovations occurred in this in-dustry after the radical innovation embodied in theWinchester drive. Importantly, although the archi-tectural innovations (the introduction of smallerdiameters) heralded access to new customers andsubmarkets, these innovations employed off-the-

shelf component technology, and “no new technol-ogy [was] involved in these disruptive products”(Christensen, 1993: 191). As a result, the underly-ing knowledge base for creating disk drives re-mained largely unchanged, even though marketdisruptions due to new customer bases caused sev-eral technologically superior firms to exit theindustry.

Finally, despite the validity of caveats regardingthe use of patents as a measure of both inventive-ness and knowledge diffusion (Jaffe, Trajtenberg, &Henderson, 1993), a strong and significant correla-tion (r ! .57, p " .001) exists between the patentingactivity of firms and their technological capabilitiesas measured by the areal density of their diskdrives, a measure commonly used for technologicalperformance in studies of this industry (Agarwal etal., 2004; Christensen, 1997). Thus, the disk driveindustry was a particularly appropriate setting forour study.

Data Sources

For firm-level information, we relied on the Disk/Trend Report, a market research publication thattracked annual productive activity by all firms, pub-lic and private, in the industry from 1977 to 1997, theperiod studied here. The detailed reports on eachfirm provided in Disk/Trend were used to track entryand exit dates. Numerous prior studies have used therich, reliable data provided by this source in empiri-cal testing (Christensen, 1993), and these studies at-test to the comprehensiveness of the data source,particularly its inclusion of small and private firms.Our own checks of these data against external sources(e.g., Lexis-Nexis, the Directory of Corporate Affilia-tions, and the Thomas Register of American Manu-facturers) confirmed the inclusiveness of the databaseand the accuracy of the entry and exit dates of thefirms and their indicated status as diversified or disk-only manufacturers (Agarwal et al., 2004; King &Tucci, 2002; Lerner, 1997). For information on pat-enting by firms operating in the disk drive industry,we relied on data drawn from the National Bureau ofEconomic Research (NBER) Patent Citations Data File(Hall, Jaffe, & Trajtenberg, 2002) and the databaseMicroPatent U.S. The choice of patent classes to in-clude in our sample involved a trade-off. Thompsonand Fox-Kean (2005) and Henderson, Jaffe, and Tra-jtenberg (2005) discuss issues that pertain to prob-lems in broadly or narrowly defining a technologythrough the choice of patent classes and subclasses.On the one hand, including a broader range of patentclasses implies that a sample will be more inclusiveof inventive activity and will represent more patents.On the other hand, the broader the range of patent

1 We note that our data do not cover the first threeyears of the industry. However, industry life cycle stud-ies (Agarwal & Gort, 1996; Gort & Klepper, 1982) indicatethat firm exit is very infrequent during such periods, sowe did not expect our results to be affected by the non-availability of data during these years.


classes, the more likely it is that the patents haveapplication outside one’s industry of interest.

We adopted a conservative strategy and re-stricted the pool of patents to the class most rele-vant to hard disks: U.S. patent classification code360, dynamic magnetic information storage or re-trieval. Since the NBER data only list the first, notall, classifications of a patent, we augmented thesedata with the MicroPatent database to ensure thatall patents that were listed under code 360 wereincluded in our data. The 360 patent class as awhole remained stable and relevant over the periodof our study, though there was considerable reor-ganization of the subclasses it contained, as is typ-ical for technologically intensive patent classes(Henderson et al., 2005). Since we used the three-digit classification, the reorganization of subclasseshad no effect on our analysis. An investigation ofthe patents held by “pure-play” hard disk drivemanufacturers—firms that do nothing but makehard disk drives (Agarwal et al., 2004; King &Tucci, 2002; Lerner, 1997)—confirmed that 57 per-cent of their patents were assigned to patent class360. The next two largest classes that the pure-playmanufacturers were assigned to were 348 (televi-sion) and 399 (electrophotography). Since the ma-jority of the patents assigned to these classes wouldhave been unrelated to hard disks, we did not in-clude these in our sample.

Data Description

The data from the above two sources were cross-checked against the information from the Disk/Trend Report. We checked the data manually torectify any inconsistencies in how firms were listedin the two patent databases. Further, for firms thathad subsidiaries, we used the NBER COMPUSTATdata file, which gives the parents of subsidiarycompanies, to ensure that patents assigned to sub-sidiaries were also included.2 We selected all disk-drive-related patents assigned to a firm that hadapplication dates between 1976 and 1997. Finally,we identified all patent citations for these patentsin each year until 1999, the final year in the NBERCitations database. This process generated a pool of5,179 patents in 57 firms that had at least onedisk-drive-related patent. The final data set con-sists of 43,161 patent-year observations—for in-

stance, patent 4,933,785 observed one year after theyear in which it was applied for, patent 4,933,785observed two years after its application year, and soforth—in an unbalanced panel that contains allyears between a patent’s application year and 1999.For every observation, the data contain detailedcharacteristics regarding both the patent and thefirm to which it was assigned.

In particular, of the 57 firms included in our data,40 exited the industry in the time period underanalysis. These firms were distributed relativelyevenly on status as diversified or pure-play diskdrive manufacturers; 28 firms were diversified, and29 were pure-play manufacturers. However, aswould be expected on the basis of size differences,the larger diversified firms patented significantlymore than the smaller pure-play firms. Among thefirms that survived (exited) the industry, 9 (19)were diversified firms and 8 (21) were pure-playfirms. Importantly, the exits of the diversified firmswere accompanied by a 93 percent decline in theirdisk-drive-related patenting activity. Indeed, 13 ofthe 19 firms had zero patents related to disk drivesafter exit. Consequently, even for the diversifiedfirms in the industry, exit from disk drives clearlymeant the “death” of their disk-drive-related activ-ity and, thus, loss of a template for other firms inthe industry.

Variables in the Study

We now turn to a description of the chief vari-ables in the study, which are summarized in Table1. Our dependent variable, citations received, wasthe number of citations received by a focal patentfrom firms other than the one holding the patent ineach year after its application year. We used appli-cation year to ensure consistency with other stud-ies of knowledge spillovers and diffusion that haveused application rather than grant year to bettertrack the vintage of a technology (e.g., Jaffe et al.,1993; Thompson & Fox-Kean, 2005). This variablemeasures interfirm knowledge flows in a mannersimilar to that used by Song, Almeida, and Wu(2003) and Rosenkopf and Almeida (2003). How-ever, we note that as Alcacer and Gittleman (2006)showed, early citations (particularly those to workthat has not yet received a patent) are more likelyadded by patent examiners and thus are not trulyreflective of knowledge flows. We omitted self-ci-tations—citations by a firm to its own earlier pat-ents—since we were primarily interested in interor-ganizational knowledge transfer. This omissionwas also conservative, since the mechanisms driv-ing self-citations may differ from those behind ci-tations by other firms (Caballero & Jaffe, 2002; Tra-

2 We note that it is very likely that not all subsidiarypatents are included in our data, given the limitations ofthe NBER database. Specifically, the NBER data capturesubsidiary structure in 1989, and only for those firms thatwere publicly listed on a U.S. exchange.


jtenberg et al., 1997). Further, since self-citationwas not possible after a firm had exited the indus-try, including self-citations could have falsely mag-nified the impact of firm exit on knowledgetransfer.

Firm exit was defined as the cessation of a firm’soperations in the disk drive industry. Since acquisi-tions represent a change in ownership and differ sub-stantially from exits, we did not include acquisitionsin our study. The indicator variable, exit, was set to 1for observations occurring after a patenting firm hadexited the industry and to 0 otherwise.

To capture the impact of firm exit on the effect ofour independent variables, we interacted exit witheach of them. The independent variables definedcharacteristics of a patent and patenting firm at thetime the firm applied for the patent. We calculatedfirm age at the time of a patent by subtracting theyear of firm entry into the disk drive industry fromthe application year of the patent. A patent’s inter-nal focus was the proportion of citations in it thatwere to the firm’s own prior patents and corre-sponded to the self-citation ratio calculated in theNBER database.3 The larger the value of this vari-able, the more an innovation drew upon the firm’sinternal knowledge base. Number of inventors wasthe number of inventors listed on a patent applica-tion, used here as an indication of the size of theteam involved in the innovative research being pat-ented. Range of technologies combined corre-sponded to the originality score calculated in theNBER data and first suggested by Trajtenberg andcolleagues (1997). By counting the number ofcitations a patent makes within each of the three-digit patent classes, this measure captures thedegree to which the patent draws upon a widerange of technological areas. The measure is definedfor patent i as:

1 ! !k!1

K "NCITEDik

NCITEDi# 2

, (1)

where Ncited represents the number of patents citedby a focal patent and k indexes three-digit patent

classes. Patents based on research that draws upon awider range of technological roots have a larger valueon this variable. Hall (2002) suggested a modificationof this measure to reflect the fact that patents with fewcitations are less likely to cite a broad range of classes.The modified measure multiplies the original mea-sure by n/(n " 1), where n is the number ofcitations made by a patent. We used the modifiedmeasure, having confirmed that our results wererobust to the choice of measure.

Among the control variables, we included firmdummies, to control for unobserved heterogeneitythat might affect citations to all of a firm’s patents,and application year dummies, to control for po-tential cohort effects. To control for the effect ofcitation lag—the difference in time between theapplication years of the citing and original pat-ents—we used a set of indicator variables, citationlag 1 to citation lag 24, setting the appropriatevariable to 1 for observations of the 1st through the24th year after a patent was applied for. Addition-ally, we included two control variables for the qual-ity of an innovation and an innovating firm: matu-rity of technology and recent technological activity.Maturity of technology was the number of citationsto prior patents made by a focal patent, divided bythe number of claims the patent made (a measure ofthe technological space a patent occupied). Morecitations to prior art per claim indicates a moredeveloped or mature technological field (Lanjouw& Schankerman, 2003). A mature technology maybe easier to understand (Sorensen & Stuart, 2000),yet it may also simply be of less interest to otherfirms. Further, because there is likely to be a largerstock of innovations for a more mature technology,any given innovation would be, ceteris paribus,less likely to be built upon. Recent technologicalactivity was computed as the mean number of disk-drive-related patents a firm had applied for in theprior three years. For patents applied for in thesecond or third year of a firm’s existence, it was themean of the number of patents applied for eachyear since firm entry. We included this measure ofa patenting firm’s technological activity at the timeof a patent because we expected that technologiesdeveloped by firms perceived as highly technolog-ically active might draw disproportionate attentionfrom other firms. Because their innovative effortswould be more broadly observed, they would bemore likely to be built upon by others (Podolny &Stuart, 1995).4

3 Several data challenges compelled Hall et al. to cal-culate lower and upper bounds for the estimate of self-citations. We used the lower bound, although the differ-ences are small and our results are invariant to the use ofeither measure. Alcacer and Gittleman (in press) notedthat a large number of self-citations are paradoxicallyadded by examiners, rather than inventors. Fortunatelyfor our purposes, high numbers of self-citations fromeither source indicate that a given patent is closely re-lated to a firm’s prior technological trajectory.

4 To avoid potential confusion, we note that our inde-pendent variables all related to information in a focalpatent—that is, the number of the firm’s own prior pat-


Table 2 provides the descriptive statistics andcorrelation matrix for the key variables in thestudy. An inspection of the correlations does notreveal any multicollinearity concerns, showing amean variance inflation factor (VIF) of 1.18 and amaximum VIF of 1.58.

METHODOLOGY

Our dependent variable was the number of cita-tions a patent received in each year after its appli-cation date, so we turned to the family of count datamodels for estimation (Greene, 2000). Our empiri-cal model was similar to that of Song et al. (2003)and Rosenkopf and Almeida (2003). Although

those papers measured the total citations a patentreceived from a given firm, we modeled the numberreceived from all firms in each year in order to beable to estimate the effect of firm exit over time.Specifically, the probability of a patent receiving agiven number of citations can be modeled as result-ing from a Poisson process:

Pr(Yit ! y) !e!"i"i

yi

yi!, (2)

where Yit represents the number of citations re-ceived by patent i in year t after the patent appli-cation. The mean value "i is parameterized in termsof xi, the vector of attributes, and coefficient vector#:

"i ! exp(x!it#). (3)

The Poisson process, however, restricts the meanand variance to be equal, which may not be a rea-sonable assumption. The negative binomial regres-sion model extends the Poisson regression modelby allowing the variance of the process to exceedthe mean (Cameron & Trivedi, 1998). The degree bywhich it does so, the overdispersion parameter,equals the variance of the process divided by itsmean. Because we had panel data, we used a ran-dom-effects negative binomial model (Hausman,Hall, & Griliches, 1984), which specified that allobservations for a given patent i shared a commonoverdispersion parameter $i, in which 1/(1 " $i) #beta(%, #), to avoid inflated standards errors. Be-cause many of our variables of interest were invari-ant within a patent, we were unable to use fixedeffects. The mean dispersion was greater than 1.3 inall models, indicating a variance at least 30 percentgreater than the mean (p[variance $ mean] % .05),indicating in turn that the negative binomial modelwas more appropriate than a Poisson model.

RESULTS

We first investigated the effect of firm exit onpatent citation counts to test if diffusion rates dif-fered significantly before and after a firm’s exit. Forease of exposition, we depict this effect graphicallyin Figure 1 and note that the results from the neg-ative binomial model presented later are consistentwith the graph. Figure 1 shows the average numberof citations received from other firms by patents ineach year after their application dates for threegroups of patents: (1) those belonging to firms thatdid not exit the industry through 1997, (2) thosebelonging to firms that would eventually exit buthad not yet done so for the relevant citation lagyear, and (3) those belonging to firms that had

ents that it cited. Our dependent variable related to thedecision by other companies to cite the focal patent sub-sequent to its granting.

TABLE 1Variable Definitions

Name Definition

Dependent variableCitations received The number of citations received in

a given year from other firms.Independent variablesFirm exit A 0/1 variable set to 1 if a firm had

exited at the time of anobservation.

Firm age at time ofpatent

The application year of a focalpatent minus the year of firmfounding.

Internal focus The percentage of citations in afocal patent to other patents ofthe same company (labeled“selfctlb” in the NBER PatentCitations Data File)

Number of inventors The number of inventors listed on afocal patent.

Range of technologiescombined

The heterogeneity of the patentclasses cited by a focal patent(labeled “original” in the NBERPatent Citations Data File).

Control variablesMaturity of

technologyThe number of citations made by a

focal patent divided by thenumber of claims it contains.

Recent technologicalactivity

The average number of patents by afirm over the three years prior topatent application.

Application yeardummies

0/1 variables for the year in whicha patent was applied for in the1977–97 period.

Citation lag 0/1 set for each of the 24 indicatorvariables (lag_1 to lag_24) forevery year after a patentapplication year.


already exited at the time of the observation. Forexample, if a firm A remained in the industry in theperiod under study, the citations its patents re-ceived for all citation lag years were placed in thefirst group, “firm surviving.” On the other hand,consider a firm B that filed a patent in 1980 andthen exited in 1985. The citations received by thisparticular patent were included in the secondgroup, “firm will exit, has not yet,” until citationlag 5 (1985 for this patent) and in the third group,“after firm exit,” for all subsequent citation lags.

Prior to the exits of their firms of origin, thepatents of firms that eventually exited the industryreceived approximately the same number of cita-tions as those of firms that remained in existence.Tests of homogeneity confirmed the visual impres-sion gained from examining the lines shown inFigure 1: the number of citations received is indis-tinguishable for the first two groups (p ! .10) forcitation lags of up to ten years. The exception is thelag of two years, in which patents of firms thatwould exit, but had not done so yet, received sig-

FIGURE 1Citations Received from Other Firms over Time

TABLE 2Summary Statistics and Correlationsa

Variable Mean s.d. Minimum Maximum 1 2 3 4 5 6 7

1. Citations by other firms 0.55 1.24 0.00 28.002. Firm has exited 0.22 0.41 0.00 1.00 ".043. Firm age at time of patent 14.01 10.39 0.00 43.00 .02 ".374. Internal focus 0.16 0.25 0.00 1.00 .01 .00 .255. Number of inventors 2.46 1.79 1.00 20.00 .05 ".12 .12 .036. Range of technologies combined 0.24 0.25 0.00 0.88 ".01 .04 .08 ".03 .027. Maturity of technology 0.87 1.37 0.02 26.33 ".03 .03 .01 ".04 ".02 .148. Recent technological activity of firm 35.72 31.78 0.00 128.00 .04 .01 .46 .27 .09 .08 .03

a n # 43,161. Because of the large n, every correlation is uninformatively significant at the 10 percent level or better.


nificantly more citations on average (0.74 versus0.59, p ! .005). Thus, patent citation counts do notdiffer, while a firm is still in existence, for firmsknown to have survived and firms known to havelater exited. Our finding of similar diffusion ratesprior to exit for exiting and surviving firms sup-ports earlier studies that indicate that firms mayexit this industry despite being technologically in-novative (Golder & Tellis, 1993; Katz & Shapiro,1985; Podolny & Stuart, 1995). It also confirms theanecdotal evidence provided by McKendrick andcolleagues (2000: 73) that led them to conclude thatthe industry landscape is “littered with the graves”of firms that were once considered technologicalleaders.

Once a firm exited, however, citations to its pat-ents dropped precipitously. As the third curve re-veals, the patents of firms that had exited the in-dustry by a citation lag year received fewercitations than either of the other two groups, andthis difference is particularly stark for the earliercitation lag years. For citation lags of 1 to 11 years,the difference between the numbers of citations tothe patents of firms that did not exit and to thepatents of those that already had exited ranges from18 to 40 percent (p[difference " 0] # 0.01). Cita-tions to the patents of firms that would exit but hadnot already done so and citations to the patents offirms that had already exited differ at the .05 levelor better for lags of 1–11 years, with the exceptionof the 5-year lag, with the difference ranging from

16 to 67 percent. Thus, we find evidence consistentwith Hypothesis 1. In spite of this drop, citations tothe patents of exited companies remained signifi-cantly above zero for most of the period (p # .05 forlags of 1–20 years). Firm exit, or death, has a det-rimental, but not a final—or “fatal”—effect on thediffusion of knowledge to other firms. For lateryears, the citation count is not significantly differ-ent from the citation counts received by the othertwo groups. This pattern is intuitively reasonableand consistent with findings that the advantages ofgeographic proximity for learning about the work ofothers “fade as the work is used and disseminated”over time (Jaffe et al., 1993: 591).

We now turn to our formal analysis of the effectof firm exit on the relationships between citationcounts and our key variables of interest. Tables 3and 4 present the results of a negative binomialestimate of citations received from other firms. Wenote that the coefficients represent semielastici-ties—that is, the proportionate change in the con-ditional mean caused by a one-unit change in theexplanatory variable (Cameron & Trivedi, 1998: 81–82). In Table 3, we present a simple exposition ofthe main effect of exit on patent citations, aggregat-ing over potential interaction effects with citationlags (given the nonlinear pattern of patents’ citationlags [e.g., Trajtenberg, 1990] observed in Figure 1)and with key explanatory variables. In Table 4, werelax this assumption and present the fully speci-fied model, allowing for a free-functional form by

TABLE 3Results of Negative Binomial Estimation of Citations Received before and after Exita

Variable

All Firms Diversified Firms

Model 1 Model 2 Model 3

Firm age at time of patent $0.01 (0.03) $0.01 (0.03) $0.01 (0.03)Internal focus $0.01 (0.06) $0.01 (0.06) $0.03 (0.06)Number of inventors 0.03** (0.01) 0.03** (0.01) 0.03** (0.01)Range of technologies $0.06 (0.04) $0.06 (0.04) $0.08* (0.05)Recent technological activity of firm 0.01** (0.00) 0.01** (0.00) 0.01** (0.00)Maturity of technology $0.03** (0.01) $0.03** (0.01) $0.03* (0.01)Firm has exited $0.06* (0.04) $0.09* (0.04)Constant $0.52 (0.57) $0.50 (0.57) $0.43 (0.58)

Firm dummy (joint significance) 0.00** 0.00** 0.00**Application year (joint significance) 0.00** 0.00** 0.00**Mean dispersion 1.48** 1.48** 1.44**Number of observations 42,927 42,927 36,426Log-likelihood $40,451.09 $40,449.59 $33,253.34

a Standard errors are in parentheses.* p # .05

** p # .01One-tailed tests.


TA

BLE

4N

egat

ive

Bin

omia

lEs

tim

atio

nof

Cit

atio

nsR

ecei

ved,

Inte

ract

ion

Mod

els

for

befo

rean

daf

ter

Exit

Var

iabl

es

All

Firm

s

Div

ersi

fied

Firm

s:M

odel

7M

odel

1M

odel

2M

odel

3M

odel

4M

odel

5M

odel

6

Firm

age

atti

me

ofpa

tent

!0.

02(0

.42)

!0.

02(0

.49)

!0.

02(0

.41)

!0.

02(0

.43)

!0.

02(0

.42)

!0.

02(0

.49)

!0.

02(0

.48)

Firm

age

atti

mes

ofpa

tent

"ex

it!

0.05

**(0

.00)

!0.

05**

(0.0

0)!

0.05

**(0

.00)

Inte

rnal

focu

s0.

02(0

.78)

0.02

(0.7

8)0.

08(0

.25)

0.02

(0.8

0)0.

02(0

.76)

0.11

(0.1

2)0.

08(0

.32)

Inte

rnal

focu

s"

exit

!0.

25*

(0.0

4)!

0.37

**(0

.00)

!0.

28*

(0.0

3)N

umbe

rof

inve

ntor

s0.

03**

(0.0

0)0.

03**

(0.0

0)0.

03**

(0.0

0)0.

03**

(0.0

0)0.

03**

(0.0

0)0.

03**

(0.0

0)0.

04**

(0.0

0)N

umbe

rof

inve

ntor

s"

exit

!0.

04*

(0.0

3)!

0.05

*(0

.02)

!0.

06*

(0.0

2)R

ange

ofte

chno

logi

esco

mbi

ned

!0.

05(0

.27)

!0.

05(0

.28)

!0.

05(0

.30)

!0.

05(0

.26)

!0.

03(0

.57)

!0.

03(0

.49)

!0.

08(0

.14)

Ran

geof

tech

nolo

gies

"ex

it!

0.10

(0.2

6)!

0.06

(0.5

0)!

0.02

(0.8

3)R

ecen

tte

chno

logi

cal

acti

vity

offi

rm0.

02**

(0.0

0)0.

02**

(0.0

0)0.

02**

(0.0

0)0.

02**

(0.0

0)0.

02**

(0.0

0)0.

02**

(0.0

0)0.

02**

(0.0

0)M

atur

ity

ofte

chno

logy

!0.

03**

(0.0

0)!

0.03

**(0

.00)

!0.

03**

(0.0

0)!

0.03

**(0

.00)

!0.

03**

(0.0

0)!

0.03

**(0

.00)

!0.

03*

(0.0

2)C

onst

ant

!1.

25*

(0.0

4)!

1.32

*(0

.03)

!1.

27*

(0.0

3)!

1.26

*(0

.03)

!1.

25*

(0.0

4)!

1.38

*(0

.02)

!1.

29*

(0.0

3)C

itat

ion

lag

(join

tsi

gnif

ican

ce)

0.00

**0.

00**

0.00

**0.

00**

0.00

**0.

00**

0.00

**Ex

it"

cita

tion

lag

(join

tsi

gnif

ican

ce)

0.00

**0.

00**

0.00

**0.

00**

0.00

**0.

00**

0.00

**Fi

rmdu

mm

y(jo

int

sign

ific

ance

)0.

00**

0.00

**0.

00**

0.00

**0.

00**

0.00

**0.

00**

App

lica

tion

year

(join

tsi

gnif

ican

ce)

0.00

**0.

00**

0.00

**0.

00**

0.00

**0.

00**

0.00

**M

ean

disp

ersi

on1.

36**

1.36

**1.

36**

1.36

**1.

36**

1.36

**1.

32**

Num

ber

ofob

serv

atio

ns43

,161

43,1

6143

,161

43,1

6143

,161

43,1

6136

,655

Log-

like

liho

od!

38,4

78.4

3!

38,4

53.0

1!

38,4

76.2

1!

38,4

76.1

3!

38,4

77.8

0!

38,4

45.5

3!

31,5

50.4

8Si

gnif

ican

ceof

incr

emen

tal

chi-

squa

reov

erm

odel

10.

00**

0.03

*0.

03*

0.26

0.00

**

*p

#.0

5**

p#

.01

including citation lag dummies. Model 1 in Table 3reports the results for the main effect of our explan-atory variables; it does not differentiate betweencitations received while an innovating firm wasactive and citations received after its exit from thefocal industry. In model 2, we include the maineffect of exit, which is negative and significant,thus providing support for Hypothesis 1.

The tests of our interaction hypotheses, Hypoth-eses 2–5, are reported in Table 4. Model 1 reportsthe coefficients of the main effects. The 24 citationlag variable dummies (not reported owing to spacelimitations) account for time-varying effects of thecitation lag years and, collectively, they are consis-tent with the patterns observed in Figure 1. Thenumber of citations increases quickly, peakingaround four years after a patent was applied for,and then slowly decreases.

To capture the effect of exit on the baseline, themodels in Table 4 include a full set of interactionterms between exit and citation lags. The coeffi-cients are jointly significant, though not reportedowing to space limitations. Instead, Figure 2 de-picts the trend in citation rates of patents beforeand after firm exit on the basis of the estimatedcoefficients in model 6 of Table 4 and computed atthe mean values of the variables. Consistently withFigure 1, the total effect of exit on the average

patent is negative and significant, indicating thatpatents received fewer citations once a patentingfirm exited the industry. We note that for the firstthree citation lags after application year, exit has apositive effect on citations. However, as Jaffe, Traj-tenberg, and Henderson (1993: 586, footnote 18)noted, citations to patents applied for but not yetgranted are most likely the result of inclusion by apatent examiner (since the citing firm could not beaware of an ungranted patent application). In ourdata, the average lag between application year andgrant year is 2.07 years; thus, we do not attributeany knowledge diffusion for these early years. Im-portantly, exit has a negative effect on citations forcitation lags 3 through 17. The predicted overallimpact of firm exit for a patent, at sample meanvalues for each covariate, is a loss of 0.05 citationsper year, a 9 percent decline. Accordingly, our fullyspecified interaction model, model 6 in Table 4,also provides support for Hypothesis 1.

Models 2–5 in Table 4 show the results of ourtests of Hypotheses 2 through 5 separately by intro-ducing each of the interactions with exit separately,and we report the full model in column 6. With theexception of the variable “range of technologiescombined,” the interaction terms significantly im-prove the fit of the model. Our test of the first twointeraction hypotheses (Hypotheses 2 and 3) re-

FIGURE 2Impact of Exit on Citations to an Average Patent


lated to firm age and internal focus yield similarresults. Neither variable is significant when a firmis active, and both have a significant, negative in-teraction with firm exit. Both variables are associ-ated with more embedded innovative routines: aninnovation that draws upon a firm’s own knowl-edge base is naturally stickier than other innova-tions, and older firms have more established inno-vative routines. Our results indicate that, when aninnovating firm is active, other firms are able toovercome these barriers through opportunities forlearning from the innovating firm’s activities. How-ever, absent this template, upon firm exit, otherfirms find it harder to build on an initial innova-tion. Thus, we find support for Hypotheses 2 and3.5

The coefficient of the variable measuring numberof inventors is positive and significant, indicatingthat patents involving more inventors receive morecitations. However, a significant, negative interac-tion with exit implies that after a firm’s exit, otherfirms find it very difficult to build on knowledgethat required larger inventor teams.6 In particular,since a high number of inventors implies a largepool of mobile employees, the large and significantcoefficient of the interaction term in model 2 un-derscores the need for the innovating firm to con-tinue to exist so that other firms may observe itsroutines and management of innovation. This find-ing supports our Hypothesis 4.

Finally, we find no effect for the variable mea-suring the range of technologies combined or itsinteraction with firm exit. Thus, Hypothesis 5 isnot supported. This absence of support suggeststhat the combination of technologies that occurs inthis industry poses little challenge for imitation.Once a firm has successfully combined multipletechnologies, other firms do not find it difficult to

build on the resulting combined technology, withor without access to the firm as a template. Wenote, though, that this result may reflect the rela-tively low value of this variable in our sample; thisvalue was 0.24 (in a range of 0–1). By comparison,the average over all sectors was approximately 0.30in 1975, and it rose to approximately 0.40 in the1990s (Hall et al., 2002: 430, Figure 15). The rangeof technologies combined in the hard drive indus-try may not have been diverse enough to pose abarrier to knowledge replication. Alternatively, itmay be due to the lack of variance resulting fromuse of an intraclass measure for the extent of diffu-sion, given our single-industry focus.

Among the control variables, we find that matu-rity of technology is negative and significant andthat recent technological activity is positive andsignificant. This pattern of findings accords withour expectations that patents building on maturetechnologies are less likely to be cited and thatpatents belonging to more technologically activefirms are cited more heavily.

To ensure robustness of our results, we con-ducted several additional tests. As noted in the datadescription, both diversified and nondiversified(pure-play) disk drive firms operated in the indus-try. This distinction is particularly important, inso-far as diversified firms continued to exist elsewhereonce they exited the disk drive industry. Accord-ingly, we tested our hypotheses for the subset ofdiversified firms. These results are reported inmodel 3 of Table 3 and model 7 of Table 4. All thehypotheses continue to be supported; the only sub-stantive change in the results is that the main effectof firm age for diversified firms is negative andsignificant.

We also ensured that our results were not sensi-tive to choice of model specification. Since most ofour interaction hypotheses relate to time-invariantvariables at the patent level, we were unable to runa panel-level fixed-effects specification. However,our results are robust to whether the firm-levelfixed effects are included or excluded in the spec-ification. We additionally conducted tests of ourhypotheses using the Poisson model. The resultsfor all the hypotheses remain unchanged. Further,Hypothesis 5, which relates to the range of technol-ogies combined, is supported in the Poissonspecification.

ALTERNATIVE EXPLANATIONS

Our hypotheses regarding the main and moder-ating effects of the exit of a firm from a technolog-ically intensive industry on the diffusion of itsknowledge centered around the importance of ac-

5 A high rate of self-citation might also indicate ahighly specialized firm that may be pursuing a line ofinquiry that others do not find promising. The signalgenerated by firm failure might provide additional nega-tive information regarding the approach and reduce cita-tions. However, we note that other firms would likely notfind such a line of inquiry promising, even before thefailure of the firm.

6 Given that both the main and interaction terms arestatistically significant, we find that each additional in-ventor increases the expected number of citations by 3.4percent when a firm is active (the coefficient on thenumber of inventors ! 0.03) and reduces the expectednumber of citations by 1.6 percent when the firm hasexited. (When exit ! 1, the coefficient equals –0.016[0.034 " {0.05} # 1], where –0.05 is the coefficient of theinteraction term of exit and number of inventors.).


cess to the private knowledge contained in thefirm’s routines. As indicated in the theory section,there may, however, be alternative explanations forthe observed effects. One possibility is that differ-ences in technological prowess caused firm exitand also manifested themselves in lower citationrates. Supporting Christensen (1993), Franco,Sarkar, Agarwal, and Echambadi (2005) found thatsuperior technological capabilities did not enhancesurvival rates in the absence of firm entry into thenew submarkets that developed in the disk driveindustry. We do not discount the possibility that alack of technological capabilities is positively re-lated to exit yet see several firms in the sample thatexited the industry in spite of their technologicalability. Indeed, the fountainhead of knowledge andcreator of the industry, IBM, ultimately ceased op-erations in disk drives in 2002, indicating the im-portance of factors beyond technological capabili-ties. Further, our data do not indicate that firmsthat exited the industry differed significantly intheir patenting behavior from firms that were stillin existence in 1997. Table 5 reports our tests ofhomogeneity for both diversified and pure-playsubsets of firms. Diversified firms that exited dur-ing the period of our study and those that did notshowed no statistically significant differences (p !.05) on the following: number of patents receivedper year, internal focus, range of technologiesdrawn upon, number of claims per patent, andnumber of inventors per patent. The same resultshold, with the exception of internal focus, for thesubset of pure-play firms. Moreover, as already in-dicated by Figure 1, no significant difference incitation rates was observed for the years in whichfirms existed between firms that remained in exis-tence over the study period and those that eventu-ally exited the industry. Most importantly, al-though differences in technological capabilitiesmay result in differences in the overall number of

patents a firm receives7 or in the overall level ofcitations those patents receive, our hypotheses cen-tered on the change in citations received by a givenpatent after firm exit. The quality of the knowledgeunderlying patents plays no role in the impact ofexit; rather, we ask to what degree other companiesbuild on a patent, contingent on the status of theinventing firm (surviving vs. defunct).

Another alternative explanation for the observedresults could be that other firms take the exit of acompany as a signal that its technology is no longerrelevant, an assumption that would lead them topay less attention to the innovations the firm gen-erated while active. As noted in our description ofthe disk drive industry, the relevance of the under-lying technology base in the industry did notchange significantly during the time period studied(Christensen, 1993). At the firm level, although wecontrolled for these issues to some degree by in-cluding variables measuring the maturity of tech-nology and recent technological activity, we couldnot rule out this explanation altogether for the maineffect of firm exit (Hypothesis 1), since it mighthave driven a portion of the overall drop in cita-tions a patent received after the exit of the firm thatheld it. However, the signaling of relevance drivercannot explain the effects of the interaction of firmexit with variables associated with varying degreesof private knowledge (Hypotheses 2–6). The sameis true for the potential explanation that firms wereless fearful of litigation on the part of source firms.In this context, we also highlight the results ob-tained above for the subset of diversified firms thatceased their disk-drive-related activity. Since thesefirms were still in existence, they were presumably

7 We note that in Table 5, patents per year seem to behigher for surviving than exiting firms, though the highstandard deviations imply that these are not statisticallysignificant.

TABLE 5Tests of Homogeneity of Firm and Patent Characteristics

Variables

Diversified Firms Pure-Play Firms

Surviving Firms Exiting Firms t Surviving Firms Exiting Firms t

Patents per year 15.70 (12.79) 5.88 (13.28) 1.87 8.06 (8.59) 1.61 (0.91) 2.12Internal focus 0.09 (0.08) 0.04 (0.07) 1.76 0.07 (0.06) 0.01 (0.02) 2.48*Range of technologies 0.31 (0.07) 0.26 (0.13) 1.49 0.39 (0.14) 0.26 (0.19) 1.95Number of claims 14.91 (5.46) 11.46 (4.81) 1.62 14.46 (3.70) 11.58 (6.27) 1.52Number of inventors 2.46 (0.97) 1.76 (0.71) 1.92 2.19 (0.74) 1.90 (0.85) 0.93Maturity of technology 0.77 (0.24) 0.88 (0.48) "0.83 1.33 (0.82) 1.13 (0.61) 0.65

* p ! .05


willing to protect their intellectual property. If alower likelihood of litigation by firms were themain driver of the drop in citations after their exit,we should not see any significant effect of exit onthe citations received by their patents. However, asnoted earlier, the patenting activity in disk drivesundertaken by the diversified firms droppedsharply after their exits from the industry. Thus,the evidence still points strongly to the loss ofprivate knowledge having an important effect.

In sum, a firm’s exit would taint all of its patentsas inferior, less likely to be defended, or irrelevantto the same degree, yet we find that the postexitreduction in citations received for a given firm’spatents is highly conditional on the characteristicsof each patent—characteristics that are closely tiedto the importance and inaccessibility of the privateknowledge associated with that patent. Indeed, thesupport that we find for the interaction hypotheses,coupled with the cross-patent variation in citationsof patents belonging to the same exited firm, helpsrule out a wide range of alternative explanations.

DISCUSSION AND CONCLUSION

The fate of innovative knowledge created byfirms that subsequently exit an industry is of prac-tical and theoretical importance. To the degree thatknowledge languishes after the exit of an innovat-ing firm, other industry participants and society atlarge lose a potential source of technologicalprogress. If, however, knowledge that a defunctfirm created is significantly diffused, these positiveexternalities result in some social benefit from theinvestments made when the firm was alive. Theo-retically, the fate of innovative knowledge afterfirm exit illuminates the impact of private knowl-edge on the diffusion of knowledge that is in theexplicit/codified domain and the extent to whichthe continued existence of a firm enhances spill-overs of its knowledge.

There is anecdotal evidence that a firm’s innova-tive knowledge can outlast its existence. The intro-duction to this study cited OIS as an example of afirm whose knowledge had considerable impact onan industry (the flat-panel display industry) afterits demise. Another such firm is Prairietek. Thiscompany was active in disk drives for only fiveyears, but one of its patents was cited 106 times inthe eight years between its exit and 1999. Our studysought to systematically study this issue and pro-vide empirical evidence on whether OIS and Prai-rietek are merely exceptions to the rule or examplesof a regular pattern of postexit diffusion.

Our study provides several insights regarding theeffect of firm exit, or death. First, death clearly

hurts knowledge diffusion. Examining the patentcitation trends before and after the exit of a firm, wefind a significant decline in the citation rate thatwas attributable to firm exit, even after controllingfor firm and patent characteristics. Our resultsshow that, in addition to the features identified inprior research as the characteristics of a firm that isthe source of knowledge, another important deter-minant of knowledge diffusion is the continuedexistence of the firm. Thus, our evidence suggeststhat studies of knowledge diffusion should addressnot only the impact of the quality of knowledge onits ultimate citation but also the fate of the firm thatoriginated the knowledge.

In this context, we note that the extent of em-ployee mobility in the disk drive industry is so highthat disk drive designers have said that “workersremain the same, they just shift periodically fromcompany to company” (McKendrick et al., 2000:44). Employee mobility and reverse engineering al-low for continued diffusion of a firm’s technologyafter its exit; however, these mechanisms are tem-pered by the inaccessibility, after firm death, ofprivate knowledge that was embedded in the firm’sorganizational structure. Indeed, the exit of a firmfrom an industry should release many employeeswho can act as conduits of knowledge transfer inthe organizations that subsequently hire them. AsIngram noted, “The experience of a failed organi-zation may be particularly likely to diffuse throughemployee mobility as participants in the failure goto new jobs” (2002: 657). As a result, to the extentthat an exited firm’s knowledge resided in the hu-man capital of individual employees (Becker,1964), mechanisms for its continued diffusion ex-ist. Although the possibility of knowledge transferto other firms through employee mobility is highestat the time of firm exit, we find no evidence ofincreased citation by other firms after firm exit.This absence of evidence may imply that knowl-edge transfer through employee mobility is moreeffective when a source firm remains active. Recip-ient firms, even when they hire employees from asource firm, may still need to either interact with,or at least observe, its rules and routines in order tofully benefit from the knowledge transfer.

However, death is not fatal to knowledge diffu-sion. OIS and Prairietek are not just anecdotal ex-ceptions. There is clear evidence of significant post-exit diffusion of knowledge. Indeed, our findingsare consistent with findings about social welfareenhancement reported by Knott and Posen, whofound that knowledge spillovers from exited firmswere associated with reduction in the costs of sur-viving firms. Thus, although citation rates did de-cline after firm exit, firms that exited still received


a significant proportion of the citations that theycould have expected to receive had they still ex-isted. Further, death may also not have a perma-nent impact on diffusion, since we found that dif-ferences in the diffusion of active and defunctfirms’ knowledge faded over time. Taken cumula-tively, the pre- and postexit citations received bythe patents of firms that ultimately exited the diskdrive industry imply that the firms provided signif-icant welfare benefits to society.

Our study is limited in many respects. As in allstudies that employ data from single industries, ourresults may not be generalizable to other industriesthat have very different conditions from those inour focal industry. In our use of patent data, ourstudy is also subject to the limitations recognizedin the literature, including the view that patentsmay not represent all inventive activity in an in-dustry. In particular, although several thousandpatents were granted to the firms in the disk driveindustry, we do not know the exact proportion ofthe total inventions in the industry that these pat-ents represent. Further, given our focus on whetheran inventing firm existed or had exited, and theempirical design of counting all citations receivedby each patent, we did not distinguish betweenrecipient firm characteristics and the mechanismsemployed for knowledge transfer. Also, given time-invariant explanatory variables, we could not testour hypotheses using a fixed-effects specificationand had to assume random effects when controllingfor patent-level unobserved heterogeneity. Further,although we identified and tried to address severalalternative explanations for the observed main ef-fect of exit and included firm-level fixed effects tocontrol for firm-level unobserved variation, wewere not able to disentangle the relative effects ofall the potential explanations and the extent towhich changes in the status and relevance of tech-nologies, along with type of firm (e.g., specialist vs.generalist) might have impacted the postexit diffu-sion of knowledge. Knowledge belonging to firmsthat exit because of a lack of complementary assetsmay be of greater postexit interest to others thanknowledge belonging to firms that fail because oftechnological weakness, but we could not isolatethe cause of each firm’s exit.

Both the results from our study and its limita-tions point to several intriguing questions for futureresearch. As we noted at the outset, given our in-terest in identifying how private knowledge com-plements public knowledge, our study highlightedthe impact of firm exit on knowledge that is explicitand codified in patents. We suspect that firm exithas an even more detrimental impact on tacit andnoncodified knowledge, resulting not just in a de-

cline in diffusion, but also in an actual loss ofknowledge. For instance, MacKenzie and Spinardi(1995) presented evidence on the “uninvention” oftacit knowledge in the nuclear weapons industrythat resulted from cessation of design and opera-tions. Similarly, Benkard (2000) found evidence oforganizational “forgetting” in the presence ofhighly tacit and human-capital-embodied knowl-edge. Future research assessing the impact of firmexit on other types of knowledge, particularly thosewith high tacitness, would help shed further lighton this issue.

Our results relate to the “average” effect of firmexit on knowledge diffusion and, in particular, ourstudy does not address the impact of heterogeneityin recipient firms’ capabilities and strategies on thepostexit diffusion of knowledge created by a sourcefirm. Future research needs to examine heterogene-ity in the use and further development of technol-ogies created by firms that exit an industry. In viewof our findings, we expect that absorptive capabil-ities (Cohen & Levinthal, 1990) will play an espe-cially important role. In the absence of an observ-able template, firms with a greater ability to dissectand absorb innovative knowledge on their own willhave an advantage in building on a departed firm’sknowledge. Relatedly, future research could distin-guish among recipient firms on the basis of locationrelative to source firm and compare the effects ofsource firm exit on knowledge diffusion to colo-cated and distant recipients. Doing so would alsoenable scholars to address whether potential geo-graphical effects are a result of language/culturalissues or a more systematic challenge associatedwith transferring technologies over geographicdistances.

Further studies are also needed to understand themechanisms underlying knowledge transfer andwhether firms should use different strategies whenseeking knowledge from existing versus defunctfirms. Such studies would have important manage-rial implications as well. Learning vicariously andforming a collaborative networks are not optionsfor learning from firms that have exited an indus-try; mechanisms for transferring technology fromdefunct firms may instead include hiring technicalemployees (Rosenkopf & Almeida, 2003), hiringmanagers (Boeker, 1997), and buying intellectualproperty (Arora, Fosfuri, & Gambardella, 2001). In-tellectual property, although useful, transfers onlyformal, public knowledge, omitting associated pri-vate, tacit know-how that may be critical. Techni-cal employees bring both explicit and tacit knowl-edge to a new firm, but they are removed from theroutines and culture of their previous employer.Managerial employees bring important organiza-


tional knowledge that can help in the re-creation ofteam dynamics and routines. Are these mecha-nisms more or less effective when a firm that is thesource of knowledge is still in existence or when ithas exited? Should the technology strategy for har-nessing knowledge created by a departed firm focuson one more than the others, or is it more effectiveto combine multiple mechanisms?

Our study has demonstrated that private knowl-edge matters, but isolating the impact of privateknowledge held at various levels in firms (individ-ual inventors, research teams, routines for achiev-ing complementarity in resources) was beyond itsscope. We hope that future research will theorizeand test the implications of these distinctions anddisentangle the effects that private knowledge ateach level may have on subsequent diffusion. Forinstance, is there greater diffusion of knowledgewhen the individual inventors scatter to differentfirms, each providing a seed for distinct trajectoriesthat build on an exited firm’s knowledge, or is theregreater diffusion when all inventors move as ateam, thereby preserving some of the knowledgeheld at the team level? If the latter is the case, whatare some impediments to collective movement ofteams from one firm to another, and how may thesebe addressed?

In conclusion, we find that stickiness owing toembeddedness is a major impediment to diffusionof knowledge, and our study highlights the impor-tance of using a firm’s activities as a template forsuccessfully replicating and extending its innova-tive knowledge. By establishing that the privateknowledge held by a firm is an important comple-ment to knowledge in the public domain and aidsin its diffusion, we have added another importantdimension to the study of knowledge diffusion.Further, we believe our results have significant im-plications for public policy and technology strategy,particularly in highly entrepreneurial industries.

Innovative and entrepreneurial firms providebenefits to society that outlast their existence; thus,public policy aimed at encouraging their activitieswill provide spillover benefits that go beyond thosemeasured through simple measures of their indi-vidual productivity and growth. Thus, evaluationof public investment in technology developmentshould include the longer-term impact of the tech-nology developed, independent of the commercialsuccess of a funded firm. However, given the loss ofprivate knowledge incurred when a funded firmfails, it would be useful to impose as a condition ofpublic funding a requirement that firms codifytheir innovative knowledge to preserve it againstthe possibility of firm failure. Combining a betterunderstanding of exited firms’ potential contribu-

tion to society with actions to encourage postfailurediffusion of knowledge might allow funding agen-cies to consider funding riskier, more entrepreneur-ial projects than would otherwise appear optimal.To the degree that these projects are embedded inyounger and smaller firms, our results suggest thatthe value of the knowledge they create will be morerobust to the possible failure of the firms.

The obvious implication of our findings for tech-nology strategy is that firms should actively incor-porate failed or failing companies in the sources ofinnovation from which they draw. Beyond that, ourfindings give guidance regarding what specific in-novations are most amenable to incorporation—those stemming for young companies, not overlyembedded in a failed firm’s idiosyncratic knowl-edge base, and having a smaller team of inventors.Future research will help us move beyond thesegeneral principles to understand what types of in-novations are most amenable to incorporation by agiven company with specific characteristics.

REFERENCESAgrawal, A., Cockburn, I., & McHale, J. 2003. Gone but

not forgotten: Labor flows, knowledge spillovers,and enduring social capital (Working paper no.9950, National Bureau of Economic Research).http://papers.nber.org/papers/w9950.

Agarwal, R., Echambadi, R., Franco, A., & Sarkar, M. B.2004. Knowledge transfer through inheritance: Spin-out generation, development and survival. Academyof Management Journal, 47: 501–522.

Agarwal, R., & Gort, M. 1996. The evolution of marketsand entry, exit and the survival of firms. Review ofEconomics and Statistics, 69: 567–574.

Alcacer, J., & Gittelman, M. 2006. How do I know whatyou know? Patent examiners and the generation ofpatent citations. Review of Economics and Statis-tics. 83: 774–779.

Almeida, P., & Kogut, B. 1999. Localization of knowledgeand the mobility of engineers in regional networks.Management Science, 45: 905–917.

Anderson, P., & Tushman, M. L. 1990. Technologicaldiscontinuities and dominant designs—A cyclicalmodel of technological change. Administrative Sci-ence Quarterly, 35: 604–633.

Anton, J. J., & Yao, D. A. 2002. The sale of ideas: Disclo-sure, property rights, and incomplete contracts. Re-view of Economic Studies, 69: 513–531.

Arora, A., Fosfuri, A., & Gambardella, A. 2001. Marketsfor technology: The economics of innovation andcorporate strategy. Cambridge, MA: MIT Press.

Arrow, K. J. 1962. Economic welfare and the allocation ofresources for invention. In R. R. Nelson (Ed.), Therate and direction of inventive activity: Economic


and social factors. Princeton, NJ: Princeton Univer-sity Press.

Arrow, K. J. 1974. The limits of organizations. NewYork: Norton.

Arrow, K. J. 1996. Technical information and industrialstructure. Industrial and Corporate Change, 5:645–652.

Audretsch, D. B., & Feldman, M. P. 1996. R&D spilloversand the geography of innovation and production.American Economic Review, 86: 630–640.

Becker, G. S. 1964. Human capital: A theoretical andempirical analysis with special reference to edu-cation. New York: Columbia University Press.

Benkard, C. L. 2000. Learning and forgetting: The dynam-ics of aircraft production. American Economic Re-view, 90: 1034–1054.

Boeker, W. 1997. Executive migration and strategicchange: The effect of top manager movement onproduct-market entry. Administrative ScienceQuarterly, 42: 213–236.

Burt, R. S. 1992. Structural holes. Cambridge, MA: Har-vard University Press.

Caballero, R. J., & Jaffe, A. B. 2002. How high are thegiants’ shoulders: An empirical assessment ofknowledge spillovers and creative destructure in amodel of economic growth. In A. B. Jaffe & M. Tra-jtenberg (Eds.), Patents, citations, and innovations:A window on the knowledge economy: 89–152.Cambridge, MA: MIT Press.

Cameron, A. C., & Trivedi, P. K. 1998. Regression anal-ysis of count data. Cambridge, U.K.: Cambridge Uni-versity Press.

Chesbrough, H. W., & Teece, D. J. 1996. When is virtualvirtuous? Organizing for innovation. Harvard Busi-ness Review, 74(1): 65–71!.

Christensen, C. 1993. The rigid disk drive industry: Ahistory of commercial and technological turbulence.Business History Review, 67: 531–588.

Christensen, C. 1997. The innovator’s dilemma: Whennew technologies cause great firms to fail. Boston:Harvard Business School Press.

Cockburn, I. M., Kortum, S., & Stern, S. 2002. Are allpatent examiners equal? The impact of examinercharacteristics, NBER working paper no. w8980,National Bureau of Economic Research, Cambridge,MA.

Cohen, W. M., & Levinthal, D. A. 1990. Absorptive ca-pacity: A new perspective on learning and innova-tion. Administrative Science Quarterly, 35: 128–152.

Cowan, R., David, P. A., & Foray, D. 2000. The expliciteconomics of knowledge codification and tacitness.Industrial and Corporate Change, 9: 211.

Cyert, R. M., & March, J. G. 1963. A behavioral theory ofthe firm. Englewood Cliffs, NJ: Prentice-Hall.

Dosi, G., Teece, D. J., & Winter, S. G. 1992. Toward atheory of corporate coherence. In G. Dosi, R. Gian-netti, & P. M. Toninelli (Eds.), Technology and en-terprise in a historical perspective: 185–211. Ox-ford, U.K.: Oxford University Press.

Dunne, T., Roberts, M. J., & Samuelson, L. 1988. Patternsof entry and exit in the U.S. manufacturing indus-tries. Rand Journal of Economics, 19: 495–515.

Dyer, J. H., & Singh, H. I. 1998. The relational view:Cooperative strategies and sources of interorganiza-tional competitive advantage. Academy of Manage-ment Review, 23: 660–679.

Franco, A., Sarkar, M., Agarwal, R., & Echambadi, R.2006. The moderating effect of technological capa-bilities on timing of entry. Academy of ManagementBest Paper Proceedings.

Golder, P. N., & Tellis, G. J. 1993. Pioneer advantage:Marketing logic or marketing legend? Journal ofMarketing Research, 30: 158–170.

Gort, M., & Klepper, S. 1982. Time paths in the diffusionof product innovations. Economic Journal, 92: 630–653.

Granovetter, M. S. 1985. Economic action and socialstructure: The problem of embeddedness. AmericanJournal of Sociology, 91: 481–510.

Greene, W. H. 2000. Econometric analysis (4th ed.).Upper Saddle River, NJ: Prentice-Hall.

Griliches, Z. 1979. Issues in addressing the contributionsof R&D to productivity growth. Bell Journal of Eco-nomics, 19: 92–116.

Gulati, R. 1998. Alliances and networks. Strategic Man-agement Journal, 19: 293–317.

Hall, B. H., Jaffe, A. B., & Trajtenberg, M. 2002. The NBERpatent citation data file: Lessons, insights and meth-odological tools. In A. B. Jaffe & M. Trajtenberg(Eds.), Patents, citations and innovation: A windowon the knowledge economy: 403–460. Cambridge,MA: MIT Press.

Haunschild, P. R., & Miner, A. S. 1997. Modes of interor-ganizational imitation: The effects of outcome sa-lience and uncertainty. Administrative ScienceQuarterly, 42: 472–500.

Hausman, J., Hall, B. H., & Griliches, Z. 1984. Economet-ric models for count data with an application to thepatents–R and D relationship. Econometrica, 52:909–938.

Henderson, R., Jaffe, A., & Trajtenberg, M. 2005. Patentcitations and the geography of knowledge spillovers:A reassessment—Comment. American EconomicReview, 95: 461–464.

Ingram, P. 2002. Interorganizational learning. In J. A. C.Baum (Ed.), Blackwell companion to organizations:652–663. Malden, MA: Blackwell.

Jaffe, A. 1986. Technological opportunity and spilloversof R & D: Evidence from firms’ patents, profits, and


market value. American Economic Review, 76:984–1001.

Jaffe, A. B., & Trajtenberg, M. 1996. Flows of knowledgefrom universities and federal laboratories. Proceed-ings of the National Academy of Sciences, 93:12671–12677.

Jaffe, A. B., & Trajtenberg, M. 2002. Patents, citations,and innovations: A window on the knowledgeeconomy. Cambridge, MA: MIT Press.

Jaffe, A. B., Trajtenberg, M., & Fogarty, M. S. 2002. Themeaning of patent citations: Report on the NBER/Case-Western Reserve survey of patentees. In A. Jaffe& M. Trajtenberg (Eds.), Patents, citations and inno-vations: 379–402. Cambridge, MA: MIT Press.

Jaffe, A. B., Trajtenberg, M., & Henderson, R. 1993. Geo-graphic localization of knowledge spillovers as evi-denced by patent citations. Quarterly Journal ofEconomics, 108: 577–598.

Jovanovic, B., & MacDonald, G. 1994. The life cycle of acompetitive industry. Journal of Political Economy,102: 322–347.

Katila, R., & Ahuja, G. 2002. Something old, somethingnew: A longitudinal study of search behavior andnew product introduction. Academy of Manage-ment Journal, 45: 1183–1195.

Katz, M. L., & Shapiro, C. 1985. Network externalities,competition, and compatibility. American Eco-nomic Review, 75: 424–440.

King, A. A., & Tucci, C. L. 2002. Incumbent entry intonew market niches: The role of experience and man-agerial choice in the creation of dynamic capabili-ties. Management Science, 48: 171–186.

Knott, A. M., & Posen, H. E. 2005. Is failure good? Stra-tegic Management Journal, 26: 617–641.

Lanjouw, J. O., & Schankerman, M. 2003. Enforcement ofpatent rights in the United States. In W. M. Cohen &S. A. Merill (Eds.), Patents in the knowledge-basedeconomy: 145–179. Washington, DC: National Acad-emies Press.

Lerner, J. 1997. An empirical exploration of a technologyrace. RAND Journal of Economics, 28: 228–247.

Levitt, B., & March, J. G. 1988. Organizational learning. InW. R. Scott (Ed.), Annual review of sociology, vol.14: 319–340. Palo Alto, CA: Annual Reviews.

Lippman, S. A., & Rumelt, R. P. 1982. Uncertain imita-bility: An analysis of interfirm differences in effi-ciency under competition. Bell Journal of Econom-ics, 13: 418–438.

MacKenzie, D., & Spinardi, G. 1995. Tacit knowledge,weapons design, and the uninvention of nuclearweapons. American Journal of Sociology, 101: 44.

Mann, R. J. 2005. An empirical investigation of liquida-tion choices of failed high-tech firms. WashingtonUniversity Law Quarterly, 82: 1375–1444.

March, J. 1991. Exploration and exploitation in organi-zational learning. Organization Science, 2: 71–87.

Martin, X., & Mitchell, W. 1998. The influence of localsearch and performance heuristics on new designintroduction in a new product market. ResearchPolicy, 26: 753–771.

McKendrick, D. G., Doner, R. F., & Haggard, S. 2000.From Silicon Valley to Singapore: Location andcompetitive advantage in the hard disk drive in-dustry. Stanford, CA: Stanford University Press.

Nelson, R. R. 1990. What is public and what is privateabout technology? CCC working paper 90-9, Univer-sity of California Center for Research in Manage-ments, Berkeley, CA.

Nelson, R. R., & Romer, P. M. 1996. Science, economicgrowth, and public policy. Challenge, 9(2): 9–21.

Nelson, R. R., & Winter, S. G. 1982. An evolutionarytheory of economic change. Cambridge, MA:Belknap.

Pfeffer, J. 1981. Management as symbolic action: Thecreation and maintenance of organizational para-digms. In L. L. Cummings & B. Staw (Eds.), Re-search in organizational behavior, vol. 3. Green-wich, CT: JAI Press.

Podolny, J. M., & Stuart, T. E. 1995. A role-based ecologyof technological change. American Journal of Soci-ology, 100: 1224–1260.

Rosenberg, N. 1982. Inside the black box: Technologyand economics. Cambridge, U.K.: Cambridge Uni-versity Press.

Rosenkopf, L., & Almeida, P. 2003. Overcoming localsearch through alliances and mobility. ManagementScience, 49: 751–766.

Rumelt, R. 1984. Towards a strategic theory of the firm.Englewood Cliffs, NJ: Prentice-Hall.

Schilling, M. A. 2006. Strategic management of techno-logical innovation (2nd ed.). New York: McGraw-Hill.

Schulz, M. 2003. Pathways of relevance: Exploring in-flows of knowledge into subunits of multinationalcorporations. Organization Science, 14: 440–459.

Song, J., Almeida, P., & Wu, G. 2003. Learning-by-hiring:When is mobility more likely to facilitate inter-firmknowledge transfer? Management Science, 49: 351–365.

Sorensen, J. B., & Stuart, T. E. 2000. Aging, obsolescence,and organizational innovation. Administrative Sci-ence Quarterly, 45: 81–112.

Spence, M. 1984. The learning curve and competition.Bell Journal of Economics, 12: 49–70.

Szulanski, G. 1996. Exploring internal stickiness: Imped-iments to the transfer of best practice within thefirm. Strategic Management Journal, 17: 27–43.

Teece, D. J. 1986. Profiting from technological innova-


tion: Implications for integration, collaboration, li-censing, and public policy. Research Policy, 15:285–305.

Thompson, P., & Fox-Kean, M. 2005. Patent citations andthe geography of knowledge spillovers: A reassess-ment. American Economic Review, 95: 450–460.

Trajtenberg, M. 1990. A penny for your quotes: Patentcitations and the value of innovations. Rand Journalof Economics, 21: 172–187.

Trajtenberg, M., Henderson, R. M., & Jaffe, A. B. 1997.University versus corporate patents: A window onthe basicness of invention. Economics of Innovationand New Technology, 5(1): 19–50.

Valentin, F., & Jensen, R. L. 2002. Reaping the fruits ofscience. Economic Systems Research, 14: 363–388.

Van de Ven, A. H. 1986. Central problems in the man-agement of innovation. Management Science, 32:590–608.

von Hippel, E. 1994. “Sticky information” and the locusof problem solving: Implications for innovation.Management Science, 40: 429–439.

Winter, S. G. 1987. Knowledge and competence as stra-tegic assets. In D. J. Teece (Ed.), The competitivechallenge: Strategies for industrial innovation andrenewal: 159–184. Cambridge, MA: Ballinger.

Winter, S. G., & Szulanski, G. 2001. Replication as strat-egy. Organization Science, 12: 730–743.

Ziedonis, R. H. 2004. Don’t fence me in: Fragmentedmarkets for technology and the patent acquisitionstrategies of firms. Management Science, 50: 804–820.

Glenn Hoetker ([email protected]) is an assistant pro-fessor of strategic management at the University of Illi-nois at Urbana-Champaign. He received a Ph.D. in inter-national business from the University of Michigan. Hisresearch interests focus on innovation and interfirmrelationships.

Rajshree Agarwal ([email protected]) is an associateprofessor of strategic management at the University ofIllinois at Urbana-Champaign. She received a Ph.D. ineconomics from the State University of New York atBuffalo. Her research interests focus on knowledge dif-fusion and the implications of entrepreneurship and in-novation for firm/industry evolution.


DEATH HURTS, BUT IT ISNÕT FATAL: THE POSTEXIT DIFFUSION …terpconnect.umd.edu/~rajshree/research/20 Hoetker... · logical knowledge, other firms m ay attempt to build on the knowledge

Documents