Hoetker, Glenn and Rajshree Agarwal, Death Hurts but it is Not Fatal

23 Pages·2007·0.57 MB·English

!AcademyofManagementJournal
2007,Vol.50,No.2,446–467.
DEATH HURTS, BUT IT ISN’T FATAL:
THE POSTEXIT DIFFUSION OF KNOWLEDGE CREATED BY
INNOVATIVE COMPANIES
GLENN HOETKER
RAJSHREE AGARWAL
University of Illinois at Urbana-Champaign
Thereislittleunderstandingofwhetherafirm’sinnovativeknowledgedieswithitor
ifinsteadsignificantdiffusionofknowledgeoccursevenafterafirmexitsanindustry.
Theoretical predictions about the differing effects of firm exit on private and public
knowledge and implications for interfirm knowledge transfer are forwarded. We
investigatedmainandmoderatingeffectsofafirm’sexitfromthediskdriveindustry
onknowledgediffusiontootherfirms,findingevidencethattheabilitytouseafirmas
a template plays a critical role in successfully replicating its knowledge. Absent this
template,knowledge“stickiness”reducesknowledgediffusion.
In1999,despitemillionsofdollarsofinvestment cally lacking in this important area and thus have
and a portfolio of innovative technologies, flat little impact on the technological progress in the
panel display manufacturer Optical Information industry. This formulation would imply that firms
Systems (OIS) shut down operations, unable to like OIS are outliers and that diffusion of the
achieve commercial success. Although OIS failed, knowledgetheycreateisgenerallylow,bothbefore
itstechnologylivedon.Aletterbythefirm’sformer and after they exit an industry. However, there is
director of advanced technologies to the editors of strong evidence that many companies exit despite
the magazine Information Display reported that having developed innovative knowledge (Golder &
evenafterthefirm’sexit,OIStechnologycontinued Tellis, 1993; Katz & Shapiro, 1985) and that a lack
to make waves in the flat panel industry, with ofcomplementaryassets(Teece,1986)oftenresults
manyofitspatentscoveringprocessesthatbecame in firms’ untimely deaths. If this is the case, then
mainstream technology. The letter then cited spe- firmexitwillnotbeperfectly,negativelycorrelated
cific innovations by other firms that had built on with technological superiority. To the extent that
OIS breakthroughs. Although the firm had exited some firms exit in spite of having created techno-
theindustryinspiteofitstechnologicalstrength,it logical knowledge, other firms may attempt to
left a lasting legacy for the industry’s technology. build on the knowledge created by departed firms.
Is the OIS story unusual, or does it highlight a Althoughtheissueofwhetherotherfirmssubse-
regular occurrence? Since technological expertise quentlycapitalizeonknowledgecreatedbycompa-
is an important determinant of firm success (Jo-
niesthatexitanindustryremainsunderresearched,
vanovic&MacDonald,1994;Teece,1986),itcould
it is important to investigate for several reasons.
be argued that firms that exit an industry are typi-
First, 8 to 10 percent of all companies leave an
industryinanaverageyear(Agarwal&Gort,1996),
but their exit may nonetheless create economic
Both authors contributed equally. The research was
benefitsandimpactsocialwelfare(Dunne,Roberts,
supported by grants received from the E. Marion Kauf-
mannFoundationandtheCampusResearchBoard,Uni- &Samuelson,1988;Knott&Posen,2005).Manyof
versity of Illinois. The manuscript has benefited from these companies may have been technologically
comments received from Editor Sara Rynes, three anon- innovative and are thus underexploited sources of
ymous reviewers, Juan Alca´cer, Joel Baum, Raj Echam- technological progress and increases in social wel-
badi, Dan Levinthal, MB Sarkar, Michael Tushman, An- fare.Further,insomeindustries,substantialpublic
drew Van de Ven, Charles Williams, and seminar
investment may have been made in these compa-
participants at the 2005 Academy of Management meet-
nies, through either tax incentives or direct fund-
ings,theUniversityofChicago,the2004HarvardEntre-
ing. Does the value of that investment depend on
preneurship and Innovation Conference, the University
the commercial success of the firm receiving the
ofIllinois,andthe2005Whartontechnologyminiconfer-
funding, or can other firms that remain commer-
ence. We thank April Franco for access to data, and
Persephone Doliner for her copy-editing services. The cially viable subsequently harness the resulting
usualdisclaimerapplies. innovation?
446
CopyrightoftheAcademyofManagement,allrightsreserved.Contentsmaynotbecopied,emailed,postedtoalistserv,orotherwisetransmittedwithoutthecopyrightholder’sexpress
writtenpermission.Usersmayprint,downloadoremailarticlesforindividualuseonly.
2007 HoetkerandAgarwal 447
Theoretically, since the issue relates to interfirm private knowledge and reducing the tacitness and
knowledge transfer under the most challenging stickiness of knowledge (von Hippel, 1994).
conditions, it offers an opportunity to examine the To the best of our knowledge, no study has sys-
issues that may be relevant when firms seek to tematically examined the impact that the exit of a
capitalize on other firms’ technologies. The tradi- firm has on the diffusion of the knowledge it has
tional view of knowledge highlights the positive created. The empirical setting of our study is the
externalities inherent in knowledge creation and hard disk drive industry, because of its technolog-
thenonrival,nonexcludablenatureofinformation, ical intensiveness and the availability of the data
particularlywhenitisembodiedinpatents(Arrow, necessary to examine our research questions
1962; Griliches, 1979; Jaffe, 1986). Patents repre- (Christensen, 1993). We define firm exit, or
sent codified knowledge that has been publicly re- “death,”asafirm’shavingceasedoperationsinthe
vealed through the publication of patent docu- disk drive industry, excluding firms that were ac-
ments, thus enabling the use of the knowledge by quired. We ensure that for diversified firms, exit
firms other than the originators (Jaffe, 1986; fromtheindustrywasconcomitantwithacessation
of their innovative activity related to the industry.
Spence, 1984). In contrast, an alternative view em-
We also investigate the possibility that firms that
phasizes that knowledge may have private as well
exited were insignificant in the development of
as public aspects (Nelson & Winter, 1982). These
harddrivetechnologyanddonotfindthistobethe
private aspects (Nelson & Romer, 1996) impart
case.Usingpatentcitationsasameasureofknowl-
knowledge “stickiness” (von Hippel, 1994), a con-
edge diffusion, we examine the effects of firm exit
sequence either of the embeddedness of innova-
not only on the overall patent-citation life cycle,
tions in organizational routines and teams (Martin
butalsoontherelationshipbetweencharacteristics
& Mitchell, 1998; Nelson & Winter, 1982) or of
of an innovation and its diffusion to other firms.
causal ambiguity (Lippman & Rumelt, 1982;
We find support for our hypothesis that exit im-
Rumelt, 1984). Restriction of interfirm knowledge
pairs the ability of other firms to draw on the
transfer is the outcome. Researchers have found
knowledge generated by a firm; firm exit results in
that significant tacit knowledge resides within the
asignificantdeclineincitationsreceivedbyafocal
socialstructuresoforganizations,sinceinnovation
patent. Further, we show that firm exit interacts
is the result of concerted and directed efforts by
withvariablesassociatedwithmoreembeddedness
entire teams of employees.
of knowledge in a firm’s private routines (firm age
Our paper builds on the complementarity of the
attimeofpatenting,degreetowhichaninnovation
privateandpubliccomponentsofknowledge(Nel-
built on the innovating firm’s internal knowledge
son,1990)toexaminehowthelackofaccessibility
base, and number of inventors) to have a negative
of private knowledge affects subsequent diffusion
impact on the patent’s citations. As a result, we
of the public knowledge embodied in a firm’s pat-
find broad support for our hypotheses that the
ents.Exitmeansbothlossoftheprivateknowledge
higher the private component of a firm’s knowl-
embodied in a firm and loss of the possibility of
edge, the more pronounced is the negative impact
using the firm’s activities as a template (Winter &
ofthefirm’sexitonsubsequentcitations.However,
Szulanski, 2001). Examining the effect of firm exit
the firm’s exit—the death of its industry-related
on knowledge diffusion can thus shed light on the
activity—does not halt all further use of its tech-
importance of private knowledge as a facilitator of
nology, and the effect of exit on subsequent cita-
thediffusionofpublicknowledge.Importantly,we
tions attenuates over time. Thus, firms that exit an
address the competing explanation that firm exit industry provide spillover benefits to others, in
mayrepresentalackofrelevanceoftheknowledge keeping with the findings of Knott and Posen
and develop hypotheses for the interaction of firm (2005).
exitwithvariablesassociatedwiththegreaterpres-
ence of private knowledge. Thus, we contribute to
theliteratureontheextentofknowledgespillovers THEORY
between firms by discussing how private knowl-
Private and Public Components of Knowledge
edge may serve as a boundary condition for the
public knowledge a firm creates. Our examination The idea that as firms pursue new knowledge,
ofthepostexitdiffusionofknowledgealsocomple- they create a public good dates back to Arrow
ments studies of the importance of geographical (1962). Subsequent work in the area has discussed
location(Agrawal,Cockburn,&McHale,2003;Au- theimplicationsofthenonrivalandnonexcludable
dretsch & Feldman, 1996) and employee mobility properties of knowledge for its subsequent diffu-
(e.g., Rosenkopf & Almeida, 2003) for accessing sion, since both aspects increase the likelihood of
448 AcademyofManagementJournal April
anotherfirmbenefitingfromtheknowledgecreated particular innovation. For example, for a leading-
byafocalfirm.Becauseinvestmentsinknowledge- edgetechnology,thetacitknowledgeofindividuals
creating activities by a firm also increase the hu- maybeparamount,butforaninnovationrequiring
man capital of its employees (Becker, 1964), em- a large team of inventors, the routines of the inno-
ployee mobility has been identified as a key vating team may be dominant.
mechanism for knowledge diffusion (Almeida & Since most innovations embody private knowl-
Kogut,1999),thoughknowledgediffusioncanalso edgeatmultiplelevels,ambiguityresultsregarding
occurthroughothermechanisms,includingcodifi- the conditions under which their technologies can
cation,reverseengineeringandscientificreproduc- be gainfully applied (Nelson & Winter, 1982). It
tion,andformalorinformalinterpersonalcontacts mayalsobedifficulttojudgethepotentialvalueof
(Arrow, 1996). an innovation (Podolny & Stuart, 1995). Greater
However,muchattentioninthelast20yearshas ambiguity on each of these dimensions limits the
also been paid to the tacit aspect of knowledge, degree to which a firm other than the source firm
particularly that which is team-based and socially can build on an innovation, even if the other firm
embedded in firm routines (Nelson & Winter, has access to the public component of the relevant
1982).Highlightingthefactthatnotalltheinnova- knowledge.
tive knowledge firms create is public, Nelson Thus,thereceivedliteraturesuggeststhatprivate
(1990) argued that firms generate innovative and public knowledge are complementary requi-
knowledge by combining generic, public knowl- sitesforthecreationofnewknowledge;inorderto
edge with specific designs and practices that are understand the private aspects of another firm’s
private and known only to their creators. The suc- innovative knowledge, a firm must overcome the
cess of other firms in replicating and building on associated embeddedness and causal ambiguity. It
the knowledge created by a firm thus depends on can attempt to do so by undertaking its own re-
their ability to understand the private knowledge search efforts to build the required understanding
within which the public knowledge is embedded internally (Cohen & Levinthal, 1990). However, vi-
(Rosenberg, 1982). carious learning—learning from the experience of
The private aspect of knowledge results in others through observation (Cyert & March,
knowledge “stickiness” (von Hippel, 1994) due to 1963)—is likely to be less costly than reinventing
causalambiguityandtheembeddednessofinnova- and learning experientially (Schulz, 2003). Since
tions in individual human capital (Becker, 1964), transferring knowledge often requires access to
and organizational or team-based rules and rou- tacit organizing principles that are not easily artic-
tines (Lippman & Rumelt, 1982; Nelson & Winter, ulated,theopportunitytoconsultaworkingexam-
1982; von Hippel, 1994; Szulanski, 1996). For ex- ple can be very valuable (Winter, 1987). As Winter
ample,causalambiguity,the“basicambiguitycon- andSzulanskiwrote,“Therecreationofacomplex,
cerning the nature of the causal connections be- imperfectlyunderstood,productiveroutineisoften
tween actions and results,” impedes duplicating a protracted process that involves many references
and extending another firm’s innovative knowl- to an existing working model” (2001: 742). This
edge (Lippman & Rumelt, 1982: 420). It may be statement is consistent with Haunschild and Min-
unclear which of the multiple research efforts that er’s (1997) finding that firms faced with uncertain
a firm engaged in ultimately led to its innovative technology rely on observing the organization that
success.Thislackofclaritymayinpartbebecause is the source of the technology for clues on how to
theknowledgeresidesatdifferentlevelswithinthe organize and act. In essence, a source firm’s rou-
firm, including individual inventors, research tinesandsubsequentactionsserveasatemplatefor
teams, and routines for combining complementary those wanting to emulate its innovative activities.
resources. In addition to occurring at different lev- Interacting with or observing the source firm en-
els,theprivateknowledgemayvaryinnatureover ables understanding which innovative trajectories
organizational levels. An individual inventor may were considered important to pursue, and what
possess tacit knowledge about the underlying sci- associated research efforts were subsequently em-
entificbasisofaninnovation.Theabilitytomanage phasized or dropped. Other firms also gain valu-
the complexities of interactions within a team is able insights on how to manage roadblocks that
likelytoresidelargelywithinteamroutines,which arise in advancing an innovation (Almeida &
no single individual may understand completely. Kogut, 1999). Observing what innovations eventu-
The overall research and product line trajectory is ally become commercial products provides a way
more likely to reside at the level of the firm. Also, to evaluate the commercial potential of an innova-
the relative importance of private knowledge at tion (Arrow, 1996). Thus, observing an innovating
different levels is contingent on the nature of a firm’s subsequent actions helps other firms in de-
2007 HoetkerandAgarwal 449
termining the level(s) at which the private knowl- The exit of a firm removes the opportunity to
edge resides, deciding what innovative knowledge observe and interact with the firm, which, as indi-
is worth replicating and extending, assessing the cated above, is important for understanding the
hurdles,andassessingthedirectionstofollowdur- private aspects of the knowledge created by the
ing replication and extension. The importance of firm. Although access to the public good aspect of
such direct or indirect interaction with an innova- the knowledge remains (via reverse engineering
tive firm has been well established in the vast lit- and reliance on codified knowledge), the firm’s
eratures on learning in alliances (Dyer & Singh, activitiescannolongerserveasatemplateforother
1998; Gulati, 1998), social networks (Burt, 1992; firms seeking to build on its knowledge. After firm
Granovetter, 1985), and knowledge spillovers via exit, the private knowledge that resides at levels
geographical proximity (Alca´cer & Gittelman, otherthanattheindividualinventorlevelislikely
2006). to undergo substantial disruption and loss. It may
notalwaysbefeasibletoprotecttheprivateknowl-
edge held at the team level, and the potential scat-
Effect of Firm Exit on Knowledge Diffusion
tering of the firm’s innovative personnel to other
Theprecedingdiscussionemphasizestheimpor- firms may additionally complicate efforts to use
tance of the continued existence of innovative social networks as a way of gathering information
firms for the diffusion of their knowledge—the on the firm. Even when all (or most) of a team are
firms themselves serve as templates, because their abletomoveenmassetoanotherfirm,theyfacethe
routines embody the interaction of the private and challengeoffunctioningunderanewmanagement
public components of their knowledge. Thus, just and incentive system. At the firm level, it is not
likeanyartifact,whetherahammeroracomputer, possible, postexit, to observe how the innovating
embodiesknowledgethatnewproducersofsimilar firm would have configured its complementary re-
artifacts can use (Cowan, David, & Foray, 2000), a sourcestobuilduponaninnovation.Furthermore,
focalfirm’sexistenceandactivityrepresentembod- since the firm’s commercialization efforts have
iedknowledgethatsubsequentdeveloperscanrely stopped,otherfirmscannotuseobservationsabout
upon while building on their own knowledge. We what innovations eventually become commercial
now argue that the exit of a firm removes the pos- productstoevaluatethecommercialpotentialofan
sibility of direct or indirect interaction with the innovation. The complementarity of the private
firm as a whole, thus limiting the extent to which andpubliccomponentsoftheknowledgeafirmhas
otherfirmscancapitalizeonitsknowledge,evenif created leads us to expect that its exit will reduce
its employees and the codified knowledge are other firms’ ability to capitalize on its knowledge.
available. We note the possibility that, since technological
In developing our hypotheses on the effect of its capabilities are positively related to firm survival,
exitonthediffusionofafirm’sknowledge,wenote exiting firms represent lower levels of technologi-
two issues. First, we deliberately focus on knowl- cal prowess. However, such a correlation would
edgethatisalreadycodifiedandinformationavail- impact the levels of citation received by a firm’s
able to other firms via patents. Patent data provide patents; there would be no reason to expect a
a stringent environment within which to test the change in the rate of citation before and after firm
importance of private knowledge and firm exis- exit. There are two other reasons, though, that are
tence. If the private knowledge of a firm is not an consistent with the observation of a postexit de-
important complement to the explicit/codified crease in firm citations. Patents often represent
knowledge available within patents, then firm exit strategic behavior (Ziedonis, 2004), and it may be
should have no appreciable impact on the rate at argued that firm exit reduces the threat of patent
whichotherfirmsuseandcitethepatentedknowl- infringement litigation. If the firm that created a
edge. Second, we focus on source firm characteris- patent is no longer around to defend the relevant
tics only, and not on recipient firms’ capabilities intellectual property, the risk that litigation will
and strategies for harnessing the knowledge that occur if subsequent patents omit its citation is re-
may affect their absorptive capacity (Cohen & duced. Given the market for intellectual property
Levinthal, 1990). Thus, we are interested in “aver- (Anton&Yao,2002;Mann,2005),otherfirmsoften
age” postexit diffusion of knowledge and do not acquireanexitingfirm’spatentrights;thusitisnot
addressdifferencesamongcitingfirmsintheircon- clear whether there is indeed a substantial decline
trol over complementary assets required to com- intheriskoflitigation.Finally,afirm’sknowledge
mercialize disk drive products, or in the relevance may lose relevance when it exits, either because of
ormagnitudeoftheirinternalR&Deffortsorhiring exogenous shocks or the exit’s perceived signal
practices. value. Although we explicitly addressed the above
450 AcademyofManagementJournal April
competing explanations both in our choice of em- corecapabilities,particularlythoserelatedtotech-
pirical context and in our testing of our first hy- nology, are developed through learning and expe-
pothesis, we cannot discount the possibility that rience, and this “path dependency” implies that
the decline in the citations associated with firm olderfirmshavehigherstocksofprivateknowledge
exit may be a result of a perceived reduction in (Sorensen & Stuart, 2000). This is because older
either the risk of litigation or the relevance of the firms have gone through a longer process of learn-
patented knowledge. Since all these reasons point ing and have stored past learning in behavioral
to a postexit decline, we hypothesize: rules and routines (Dosi, Teece, & Winter, 1992;
Nelson&Winter,1982).Thus,itmaybedifficultto
Hypothesis 1. Subsequent citation (use) of a
build on established firms’ capabilities as they are
patent by other firms in an industry is nega-
more likely to be embedded in networks of in-
tively impacted by the patenting firm’s exit
trafirm relationships. Building on an older firm’s
from the industry.
knowledge may require a recipient firm to observe
We note, however, that if factors related to rele- orinteractwiththeolderfirmmorethanwouldbe
vance or to risk of litigation, rather than to the necessarywithayoungersourcefirm,tolearnboth
accessibility of private knowledge, are the true its rules and routines and how its subsequent in-
drivers of our hypothesized decline in citations novationsbuiltonitsearlierones.Thus,theoldera
afterfirmexit,thereshouldbenodifferenceinthe firm was at the time of a patent, the greater will be
rates of knowledge diffusion among characteristics the impact of the loss of the firm as a template.
associatedwithvaryingdegreesofprivateandpub- A similar logic applies to innovations that result
liccomponentsofknowledge.Inthefollowingsec- from a firm building on its prior innovations (Jaffe
tion,wedevelopinteractionhypothesesthatenable & Trajtenberg, 2002). These innovations draw
us to isolate the role of accessibility of private heavily on a firm’s internal knowledge base rather
knowledge in determining postexit diffusion. than on the knowledge of others and are said to
reflect “localized search” (Anderson & Tushman,
1990).Theywillthereforebecloselyboundwithin
Interaction of Firm Exit with Knowledge
the routines and culture of the innovating firm
Characteristics
(Nelson&Winter,1982).Further,theyarelikelyto
Theimportanceoftheprivateknowledgeheldby becouchedintheidiosyncraticlanguageofthefirm
a firm to the diffusion of its patented knowledge (Arrow, 1974). As such, innovations that draw
will vary with the characteristics of an innovation. heavilyuponasourcefirm’sinternalknowledgebase
The greater the private component of knowledge, willbehighlytacitanddifficultforotherstoimitate
the greater will be the effect of firm exit on the and extend, particularly after the exit of the source
subsequent diffusion of knowledge. We examine firm.Again,weanticipatealargerpostexitdropinthe
theinteractionofexitwithfourvariablesthathave diffusion of an innovation if that innovation drew
been associated with the embeddedness of knowl- heavily on a firm’s internal knowledge base.
edge in a firm’s private routines: the age of the Since older firms and firms that draw on their
innovatingfirm,thedegreetowhichtheinnovation internal knowledge bases will have more private
built on the innovating firm’s internal knowledge knowledge, we hypothesize:
base, the number of inventors, and the diversity of
technologies the innovation drew upon. Each vari- Hypothesis2.Theolderthepatentingfirmisat
ableinfluencestheimportanceand/oraccessibility thetimeofapatentapplication,themoreneg-
of the innovating firm’s private knowledge. Since ativelythefirm’sexitimpactssubsequentcita-
thelossoftheinnovatingfirmasatemplatemakes tion (use) of that patent by other firms.
it more difficult to replicate the firm’s private
Hypothesis 3. The more related a patent is to
knowledge, we expect that exit will have a larger
the patenting firm’s internal knowledge base,
negativeimpactthemoreimportantorinaccessible
themorenegativelythefirm’sexitimpactssub-
the private knowledge was for that innovation. We
sequent citation (use) of that patent by other
now examine each variable in turn, exploring its
firms.
relationship to the role of the private knowledge
associated with innovations. The larger the number of inventors associated
It is well established that the embeddedness of withaninnovation,thelargeristhepoolofmobile
innovations in organizational routines increases employees upon the exit of the firm from the in-
with a firm’s age owing to greater formalization of dustry.Indeed,asourcefirm’semployeesmaycon-
structuresandencodingoflessonsinroutines(Lev- tinue to build on a technology once they join (or
itt&March,1988;Nelson&Winter,1982).Afirm’s create) other firms, and this condition might lead
2007 HoetkerandAgarwal 451
one to argue for a greater diffusion of knowledge and therefore hard for outsiders to imitate (Nelson
after an exit. However, the greater the number of & Winter, 1982). Further, just as tacit expertise is
inventors in a research team, the more numerous vital to the management of products with many
the necessary interactions between individuals, interacting components (Chesbrough & Teece,
and the more embedded the innovation in a com- 1996), it is also important in the management of
plex web of relationships (Van de Ven, 1986). research that draws on many interacting technolo-
When a team of inventors is large, the range of gies.Thus,directinteractionandvicariouslearning
specialized skills represented on it is also often should be especially important for the diffusion of
large (Schilling, 2006; Valentin & Jensen, 2002). technologies that draw on a wide range of technol-
Such a large team represents not simply more in-
ogies. This argument implies that firm exit will
teractions, but increasingly complex ones. Main-
have a greater detrimental impact on the subse-
taining effective communication in a group whose
quent use of an innovation that embodies a wide
members have diverse technical backgrounds is a
range of technologies.
complexchallenge(Pfeffer,1981)requiringthede-
velopment of routines and languages that span
Hypothesis5.Themorediversetechnologiesa
technical specializations.
patent draws upon, the more negatively the
Thus, the greater the number of inventors on a
patenting firm’s exit impacts subsequent cita-
team, the greater the degree of private knowledge
tion (use) of that patent by other firms.
that is embedded in the team and its firm. This
increaseinprivateknowledgeincreasestheimpor-
tanceofthecontinuedexistenceoftheknowledge-
DATA
creatingfirmforotherfirmsseekingtobuildonits
innovations.Absenttheroutinesofadepartedfirm, To address the research questions above, we
other firms and their individual inventors will, we needed to examine knowledge diffusion across
believe,havelimitedabilitytoreplicatetheexiter’s firms for the census of corporations that entered
activities.Further,thehigherthenumberofinven- (and exited) an industry. We tracked such knowl-
torsonateam,themoredifficultitisfortheentire edge diffusion as the subsequent use of a firm’s
team to be hired or easily assimilated by another technology by other firms via patent citations. In
firm.Thus,althoughindividualemployeesmaybe doingso,wefollowedalargebodyofresearchthat
abletoleveragetheirknowledgeattheirnewplace
has used the citations a patent receives as an indi-
of employment, team- and firm-level private
cation of the degree to which subsequent innova-
knowledgemaybemoredifficulttoreplicate.Over-
tions have built upon it (Jaffe & Trajtenberg, 1996;
all, we posit that a firm’s exit will have a stronger
Katila&Ahuja,2002).Thechiefadvantageofusing
impact on the diffusion of knowledge created by a
patent data for our purposes was that these data
large team of inventors than it will have on the
reliably capture subsequent use of innovative
diffusion of knowledge created by a small team.
knowledge by other firms. An inventor who files a
Accordingly,
patent application is required by law to list all
“prior art” of which she or he is aware. Unlike
Hypothesis 4. The larger the team of inventors
academic citations, these citations to earlier work
apatenthas,themorenegativelythepatenting
have the important legal function of limiting the
firm’sexitimpactssubsequentcitation(use)of
scope of the property right granted to the patent.
that patent by other firms.
Further,thepatentexaminerinchargeoftheappli-
cation,whoisanexpertinthetechnologicalareaof
Similarly, innovations that draw upon a wide
the patent, can add citations that the inventor may
range of underlying technologies (e.g., organic
have missed or concealed. This practice reduces
light-emitting diodes, which require expertise in
the probability that irrelevant patents will be cited
electronics, organic chemistry, and materials sci-
ence) tend to be stickier than those that are exten- or that relevant patents will be omitted. Not every
sions of a narrow field of knowledge, since they citation represents awareness of the cited patent
may require exploration rather than exploitation within an organization filing the citing patent,
(March, 1991). Knowledge that synthesizes diver- since the patent examiner could have added the
gent knowledge bases tends to be highly original citation(Alca´cer&Gittelman,2006;Cockburn,Kor-
(Trajtenberg,Henderson,&Jaffe,1997),andcombi- tum, & Stern, 2002); however, a variety of studies
nationsofmultiplefieldstendtooccuratthetech- have confirmed that patent citations are an accu-
nological frontier. Knowledge surrounding such rate, though noisy, indicator of actual knowledge
breakthrough research is likely to be highly tacit flows (Jaffe, Trajtenberg, & Fogarty, 2002).
452 AcademyofManagementJournal April
Context: The Disk Drive Industry shelfcomponenttechnology,and“nonewtechnol-
ogy [was] involved in these disruptive products”
Giventhedatarequirementsofastudyonknowl-
(Christensen, 1993: 191). As a result, the underly-
edge diffusion before and after the exits of firms,
ing knowledge base for creating disk drives re-
the industry chosen for our empirical context
mained largely unchanged, even though market
needed to conform to certain boundaries. First, it
disruptionsduetonewcustomerbasescausedsev-
had to be relatively technologically intensive, be-
eral technologically superior firms to exit the
cause technologically intensive industries have
industry.
higher rates of knowledge generation, and hence
Finally, despite the validity of caveats regarding
higher rates of knowledge transfer. Second, we
the use of patents as a measure of both inventive-
needed longitudinal data on firms that were suc-
nessandknowledgediffusion(Jaffe,Trajtenberg,&
cessful in the chosen industry and those that ulti-
Henderson, 1993), a strong and significant correla-
mately exited it. Third, although the industry had
tion(r!.57,p".001)existsbetweenthepatenting
to experience significant technological change, it
activityoffirmsandtheirtechnologicalcapabilities
needed to have some stable underlying knowledge
as measured by the areal density of their disk
base—that is, knowledge that continued to have
drives,ameasurecommonlyusedfortechnological
relevancy over time. We selected the hard disk
performanceinstudiesofthisindustry(Agarwalet
drive industry for our empirical context since it
al., 2004; Christensen, 1997). Thus, the disk drive
conformed to both the theoretical and empirical
industry was a particularly appropriate setting for
requirements of the study.
our study.
Disk drives are magnetic information storage de-
vices used in computers. In 1973, IBM pioneered
the14-inchWinchester,thefirstcompletelysealed Data Sources
andremovablediskdrive,andthediskdriveindus-
For firm-level information, we relied on the Disk/
try has since experienced rapid technological evo-
Trend Report, a market research publication that
lution (see Christensen [1993, 1997] for a detailed
trackedannualproductiveactivitybyallfirms,pub-
industry history). The industry experienced signif-
licandprivate,intheindustryfrom1977to1997,the
icant levels of both entry and exit in the relevant
period studied here. The detailed reports on each
period,andithasfollowedthetypicalindustrylife
firmprovidedinDisk/Trendwereusedtotrackentry
cycleofintroduction,growth,shakeout,andmatu-
andexitdates.Numerouspriorstudieshaveusedthe
rity(Gort&Klepper,1982).Sinceeveryproductive
rich,reliabledataprovidedbythissourceinempiri-
firm was included in our data, regardless of size,
cal testing (Christensen, 1993), and these studies at-
thedatadonotsufferfromasurvivorbias.1Manyof
test to the comprehensiveness of the data source,
the entering firms represented employee entrepre-
particularly its inclusion of small and private firms.
neurship and, thus, interfirm knowledge transfer
Ourownchecksofthesedataagainstexternalsources
(Agarwal,Echambadi,Franco,&Sarkar,2004).Ad-
(e.g., Lexis-Nexis, the Directory of Corporate Affilia-
ditionally, as McKendrick, Doner, and Haggard
tions, and the Thomas Register of American Manu-
(2000) documented, both the employee mobility
facturers)confirmedtheinclusivenessofthedatabase
andinterfirmspilloversthatshapenewfirms’tech-
and the accuracy of the entry and exit dates of the
nology and location choices are extensive in the
firmsandtheirindicatedstatusasdiversifiedordisk-
disk drive industry.
only manufacturers (Agarwal et al., 2004; King &
With regard to the pace of technological change,
Tucci, 2002; Lerner, 1997). For information on pat-
we knew that numerous architectural, modular,
enting by firms operating in the disk drive industry,
and incremental innovations occurred in this in-
wereliedondatadrawnfromtheNationalBureauof
dustryaftertheradicalinnovationembodiedinthe
EconomicResearch(NBER)PatentCitationsDataFile
Winchester drive. Importantly, although the archi-
(Hall, Jaffe, & Trajtenberg, 2002) and the database
tectural innovations (the introduction of smaller
MicroPatent U.S. The choice of patent classes to in-
diameters) heralded access to new customers and
cludeinoursampleinvolvedatrade-off.Thompson
submarkets, these innovations employed off-the-
and Fox-Kean (2005) and Henderson, Jaffe, and Tra-
jtenberg (2005) discuss issues that pertain to prob-
lems in broadly or narrowly defining a technology
1We note that our data do not cover the first three
through the choice of patent classes and subclasses.
yearsoftheindustry.However,industrylifecyclestud-
Ontheonehand,includingabroaderrangeofpatent
ies(Agarwal&Gort,1996;Gort&Klepper,1982)indicate
that firm exit is very infrequent during such periods, so classes implies that a sample will be more inclusive
we did not expect our results to be affected by the non- ofinventiveactivityandwillrepresentmorepatents.
availabilityofdataduringtheseyears. Ontheotherhand,thebroadertherangeofpatent
2007 HoetkerandAgarwal 453
classes, the more likely it is that the patents have stance,patent4,933,785observedoneyearafterthe
application outside one’s industry of interest. year in which it was applied for, patent 4,933,785
We adopted a conservative strategy and re- observedtwoyearsafteritsapplicationyear,andso
stricted the pool of patents to the class most rele- forth—in an unbalanced panel that contains all
vant to hard disks: U.S. patent classification code yearsbetweenapatent’sapplicationyearand1999.
360, dynamic magnetic information storage or re- For every observation, the data contain detailed
trieval. Since the NBER data only list the first, not characteristics regarding both the patent and the
all, classifications of a patent, we augmented these firm to which it was assigned.
data with the MicroPatent database to ensure that Inparticular,ofthe57firmsincludedinourdata,
all patents that were listed under code 360 were 40 exited the industry in the time period under
included in our data. The 360 patent class as a analysis. These firms were distributed relatively
wholeremainedstableandrelevantovertheperiod evenly on status as diversified or pure-play disk
of our study, though there was considerable reor- drivemanufacturers;28firmswerediversified,and
ganization of the subclasses it contained, as is typ- 29 were pure-play manufacturers. However, as
ical for technologically intensive patent classes would be expected on the basis of size differences,
(Henderson et al., 2005). Since we used the three- the larger diversified firms patented significantly
digitclassification,thereorganizationofsubclasses more than the smaller pure-play firms. Among the
had no effect on our analysis. An investigation of firms that survived (exited) the industry, 9 (19)
the patents held by “pure-play” hard disk drive were diversified firms and 8 (21) were pure-play
manufacturers—firms that do nothing but make firms.Importantly,theexitsofthediversifiedfirms
hard disk drives (Agarwal et al., 2004; King & were accompanied by a 93 percent decline in their
Tucci,2002;Lerner,1997)—confirmedthat57per- disk-drive-related patenting activity. Indeed, 13 of
cent of their patents were assigned to patent class the19firmshadzeropatentsrelatedtodiskdrives
360.Thenexttwolargestclassesthatthepure-play after exit. Consequently, even for the diversified
manufacturers were assigned to were 348 (televi- firms in the industry, exit from disk drives clearly
sion) and 399 (electrophotography). Since the ma- meantthe“death”oftheirdisk-drive-relatedactiv-
jorityofthepatentsassignedtotheseclasseswould ity and, thus, loss of a template for other firms in
have been unrelated to hard disks, we did not in- the industry.
clude these in our sample.
Variables in the Study
Data Description
We now turn to a description of the chief vari-
Thedatafromtheabovetwosourceswerecross- ables in the study, which are summarized in Table
checked against the information from the Disk/ 1. Our dependent variable, citations received, was
Trend Report. We checked the data manually to the number of citations received by a focal patent
rectifyanyinconsistenciesinhowfirmswerelisted fromfirmsotherthantheoneholdingthepatentin
in the two patent databases. Further, for firms that eachyearafteritsapplicationyear.Weusedappli-
hadsubsidiaries,weusedtheNBERCOMPUSTAT cation year to ensure consistency with other stud-
data file, which gives the parents of subsidiary iesofknowledgespilloversanddiffusionthathave
companies, to ensure that patents assigned to sub- used application rather than grant year to better
sidiarieswerealsoincluded.2Weselectedalldisk- track the vintage of a technology (e.g., Jaffe et al.,
drive-related patents assigned to a firm that had 1993; Thompson & Fox-Kean, 2005). This variable
application dates between 1976 and 1997. Finally, measures interfirm knowledge flows in a manner
we identified all patent citations for these patents similar to that used by Song, Almeida, and Wu
in each year until 1999, the final year in the NBER (2003) and Rosenkopf and Almeida (2003). How-
Citationsdatabase.Thisprocessgeneratedapoolof ever, we note that as Alca´cer and Gittleman (2006)
5,179 patents in 57 firms that had at least one showed, early citations (particularly those to work
disk-drive-related patent. The final data set con- that has not yet received a patent) are more likely
sists of 43,161 patent-year observations—for in- added by patent examiners and thus are not truly
reflective of knowledge flows. We omitted self-ci-
tations—citations by a firm to its own earlier pat-
ents—sincewewereprimarilyinterestedininteror-
2We note that it is very likely that not all subsidiary
ganizational knowledge transfer. This omission
patentsareincludedinourdata,giventhelimitationsof
the NBER database. Specifically, the NBER data capture was also conservative, since the mechanisms driv-
subsidiarystructurein1989,andonlyforthosefirmsthat ing self-citations may differ from those behind ci-
werepubliclylistedonaU.S.exchange. tations by other firms (Caballero & Jaffe, 2002; Tra-
454 AcademyofManagementJournal April
jtenberg et al., 1997). Further, since self-citation classes.Patentsbasedonresearchthatdrawsupona
was not possible after a firm had exited the indus- widerrangeoftechnologicalrootshavealargervalue
try,includingself-citationscouldhavefalselymag- onthisvariable.Hall(2002)suggestedamodification
nified the impact of firm exit on knowledge ofthismeasuretoreflectthefactthatpatentswithfew
transfer. citationsarelesslikelytociteabroadrangeofclasses.
Firm exit was defined as the cessation of a firm’s Themodifiedmeasuremultipliestheoriginalmea-
operations in the disk drive industry. Since acquisi- sure by n/(n " 1), where n is the number of
tionsrepresentachangeinownershipanddiffersub- citationsmadebyapatent.Weusedthemodified
stantiallyfromexits,wedidnotincludeacquisitions measure, having confirmed that our results were
inourstudy.Theindicatorvariable,exit,wassetto1 robust to the choice of measure.
for observations occurring after a patenting firm had Among the control variables, we included firm
exitedtheindustryandto0otherwise. dummies, to control for unobserved heterogeneity
Tocapturetheimpactoffirmexitontheeffectof that might affect citations to all of a firm’s patents,
our independent variables, we interacted exit with and application year dummies, to control for po-
each of them. The independent variables defined tential cohort effects. To control for the effect of
characteristicsofapatentandpatentingfirmatthe citation lag—the difference in time between the
timethefirmappliedforthepatent.Wecalculated application years of the citing and original pat-
firm age at the time of a patent by subtracting the ents—we used a set of indicator variables, citation
yearoffirmentryintothediskdriveindustryfrom lag 1 to citation lag 24, setting the appropriate
the application year of the patent. A patent’s inter- variableto1forobservationsofthe1stthroughthe
nal focus was the proportion of citations in it that 24th year after a patent was applied for. Addition-
were to the firm’s own prior patents and corre- ally,weincludedtwocontrolvariablesforthequal-
sponded to the self-citation ratio calculated in the ity of an innovation and an innovating firm: matu-
NBER database.3 The larger the value of this vari- rityoftechnologyandrecenttechnologicalactivity.
able, the more an innovation drew upon the firm’s Maturityoftechnologywasthenumberofcitations
internal knowledge base. Number of inventors was to prior patents made by a focal patent, divided by
thenumberofinventorslistedonapatentapplica- thenumberofclaimsthepatentmade(ameasureof
tion, used here as an indication of the size of the the technological space a patent occupied). More
teaminvolvedintheinnovativeresearchbeingpat- citations to prior art per claim indicates a more
ented. Range of technologies combined corre- developed or mature technological field (Lanjouw
sponded to the originality score calculated in the & Schankerman, 2003). A mature technology may
NBER data and first suggested by Trajtenberg and be easier to understand (Sorensen & Stuart, 2000),
colleagues (1997). By counting the number of yet it may also simply be of less interest to other
citationsapatentmakeswithineachofthethree- firms.Further,becausethereislikelytobealarger
digit patent classes, this measure captures the stockofinnovationsforamorematuretechnology,
degree to which the patent draws upon a wide any given innovation would be, ceteris paribus,
rangeoftechnologicalareas.Themeasureisdefined less likely to be built upon. Recent technological
forpatentias: activitywascomputedasthemeannumberofdisk-
drive-related patents a firm had applied for in the
!K "NCITED #2 prior three years. For patents applied for in the
ik
1! , (1)
NCITED secondorthirdyearofafirm’sexistence,itwasthe
i
k!1 mean of the number of patents applied for each
whereNcitedrepresentsthenumberofpatentscited year since firm entry. We included this measure of
by a focal patent and k indexes three-digit patent apatentingfirm’stechnologicalactivityatthetime
of a patent because we expected that technologies
developed by firms perceived as highly technolog-
icallyactivemightdrawdisproportionateattention
3Several data challenges compelled Hall et al. to cal- from other firms. Because their innovative efforts
culate lower and upper bounds for the estimate of self-
would be more broadly observed, they would be
citations.Weusedthelowerbound,althoughthediffer-
more likely to be built upon by others (Podolny &
encesaresmallandourresultsareinvarianttotheuseof
Stuart, 1995).4
either measure. Alca´cer and Gittleman (in press) noted
that a large number of self-citations are paradoxically
added by examiners, rather than inventors. Fortunately
for our purposes, high numbers of self-citations from 4Toavoidpotentialconfusion,wenotethatourinde-
either source indicate that a given patent is closely re- pendent variables all related to information in a focal
latedtoafirm’spriortechnologicaltrajectory. patent—that is, the number of the firm’s own prior pat-
2007 HoetkerandAgarwal 455
TABLE 1 those papers measured the total citations a patent
Variable Definitions receivedfromagivenfirm,wemodeledthenumber
received from all firms in each year in order to be
Name Definition able to estimate the effect of firm exit over time.
Specifically, the probability of a patent receiving a
Dependentvariable
givennumberofcitationscanbemodeledasresult-
Citationsreceived Thenumberofcitationsreceivedin
agivenyearfromotherfirms. ing from a Poisson process:
Independentvariables
Firmexit A0/1variablesetto1ifafirmhad e!"i"iyi
Pr(Y !y)! , (2)
exitedatthetimeofan it y!
i
observation.
Firmageattimeof Theapplicationyearofafocal where Y represents the number of citations re-
it
patent patentminustheyearoffirm
ceived by patent i in year t after the patent appli-
founding.
cation.Themeanvalue" isparameterizedinterms
Internalfocus Thepercentageofcitationsina i
focalpatenttootherpatentsof ofxi,thevectorofattributes,andcoefficientvector
thesamecompany(labeled #:
“selfctlb”intheNBERPatent
CitationsDataFile) " !exp(x!#). (3)
i it
Numberofinventors Thenumberofinventorslistedona
focalpatent. The Poisson process, however, restricts the mean
Rangeoftechnologies Theheterogeneityofthepatent
and variance to be equal, which may not be a rea-
combined classescitedbyafocalpatent
sonable assumption. The negative binomial regres-
(labeled“original”intheNBER
PatentCitationsDataFile). sion model extends the Poisson regression model
Controlvariables by allowing the variance of the process to exceed
Maturityof Thenumberofcitationsmadebya themean(Cameron&Trivedi,1998).Thedegreeby
technology focalpatentdividedbythe
which it does so, the overdispersion parameter,
numberofclaimsitcontains.
equals the variance of the process divided by its
Recenttechnological Theaveragenumberofpatentsbya
activity firmoverthethreeyearspriorto mean. Because we had panel data, we used a ran-
patentapplication. dom-effects negative binomial model (Hausman,
Applicationyear 0/1variablesfortheyearinwhich Hall, & Griliches, 1984), which specified that all
dummies apatentwasappliedforinthe
observations for a given patent i shared a common
1977–97period.
overdispersion parameter$ in which 1/(1 "$) #
Citationlag 0/1setforeachofthe24indicator i, i
variables(lag_1tolag_24)for beta(%, #), to avoid inflated standards errors. Be-
everyyearafterapatent causemanyofourvariablesofinterestwereinvari-
applicationyear. ant within a patent, we were unable to use fixed
effects.Themeandispersionwasgreaterthan1.3in
allmodels,indicatingavarianceatleast30percent
greater than the mean (p[variance $ mean] % .05),
Table 2 provides the descriptive statistics and
indicatinginturnthatthenegativebinomialmodel
correlation matrix for the key variables in the
was more appropriate than a Poisson model.
study. An inspection of the correlations does not
reveal any multicollinearity concerns, showing a
mean variance inflation factor (VIF) of 1.18 and a RESULTS
maximum VIF of 1.58.
We first investigated the effect of firm exit on
patent citation counts to test if diffusion rates dif-
METHODOLOGY feredsignificantlybeforeandafterafirm’sexit.For
easeofexposition,wedepictthiseffectgraphically
Our dependent variable was the number of cita-
in Figure 1 and note that the results from the neg-
tions a patent received in each year after its appli-
ativebinomialmodelpresentedlaterareconsistent
cationdate,soweturnedtothefamilyofcountdata
withthegraph.Figure1showstheaveragenumber
models for estimation (Greene, 2000). Our empiri-
ofcitationsreceivedfromotherfirmsbypatentsin
cal model was similar to that of Song et al. (2003)
each year after their application dates for three
and Rosenkopf and Almeida (2003). Although
groups of patents: (1) those belonging to firms that
did not exit the industry through 1997, (2) those
ents that it cited. Our dependent variable related to the belonging to firms that would eventually exit but
decisionbyothercompaniestocitethefocalpatentsub- had not yet done so for the relevant citation lag
sequenttoitsgranting. year, and (3) those belonging to firms that had