crossing the organizational species barrier: how...

CROSSING THE ORGANIZATIONAL SPECIES BARRIER:HOW VENTURE CAPITAL PRACTICES INFILTRATED THE

INFORMATION TECHNOLOGY SECTOR

VIBHA GABAINSEAD

ALAN D. MEYERUniversity of Oregon

We examine the contagion processes whereby practices originating in one organiza-tional population spread into and diffuse within a second. We theorize that “endemic”innovations native to one population spread to other populations through two distinctforms of contagion. We test this argument by observing information technology firms’adoption of corporate venture capital programs. Results suggest that geographic prox-imity triggers cross-population contagion, that within-population contagion arisesfrom different causal mechanisms, and that firms maintaining close cross-populationties pay less attention to the actions taken and outcomes experienced by other firmswithin their own industry.

Fueled by emerging technology and a robust econ-omy, corporate equity investing in start-up venturesexploded in the 1990s. The decade’s last five yearsalone saw a 3,800 percent increase, as corporate ven-ture capital investments grew from $172 million to$6.8 billion (Venture Economics, 2006). Corporationsscrambled to establish venture capital programs,identify best practices, and take equity in young en-trepreneurial ventures. Rene Savelsberg, head of Phil-ips’s corporate venture capital arm, described his pro-gram’s genesis as follows:

We first got interested in corporate venture capitalthrough our marketing relationship with WebTVback in 1997. At that time, this start-up company,literally working out of a garage, approached meabout signing a marketing relationship with them.What we found was that they were much furtheradvanced than our own internal developmentgroup. As a result, we decided to go ahead. The nextweek, they announced to the world that we hadsigned a marketing agreement with them. Fourmonths later, they were sold out to Microsoft for$400 million. Our board saw this deal and said,“Wait a minute, it was the Philips name and back-

ing that created that $400 million. Otherwise, theywould have had no credibility. We have to domore . . . to capture some of that value that hasbeen created. (Henderson & Leleux, 2002: 432)

The prospect of capturing financial value mayhave galvanized the Philips board into establishinga venturing arm, but many corporate programs citestrategic returns as their primary objective (Dush-nitsky & Lenox, 2006). Proponents hail corporateventure capital programs as vehicles for strategicrenewal that can import “disruptive technologies”and new business models into established corpora-tions. Recent research offers support for suchclaims by linking corporate venture capital (CVC)programs with increased knowledge creation(Wadhwa & Kotha, 2006) and higher rates of tech-nological innovation (Dushnitsky & Lenox, 2005).Some scholars have characterized corporate ven-ture capital programs as complements to orthodoxR&D (Dushnitsky & Lenox, 2005); others have gonefurther, suggesting that CVC can even replace in-house R&D—in effect, outsourcing it to start-ups(Rausch, 2004).

The venture capital model is widely acknowl-edged as a powerful, disciplined tool for makinginvestment decisions in the face of risk and uncer-tainty. However, transplantation of the private VCmodel into the public corporation is difficult andfraught with risks (Gompers & Lerner, 2001a). Theventure capital model was originally formulatedwithin a cottage industry where a small set of close-knit partnerships invented practices for vetting andsyndicating deals, mentoring and monitoring the

The authors thank David Strang, Elaine Romanelli,Patricia Thornton, Rick Mowday, Herminia Ibarra, Char-lie Galunic, Martin Gargiulo, John Weeks, Fabrizio Cas-tellucci, Phil Anderson, Theresa Lant, and Steve Meziasfor comments on previous versions of this article. Thisresearch was supported by National Science FoundationGrant SES-0217711, INSEAD research grant 2520-025R,and by the University of Oregon’s Lundquist Center forEntrepreneurship.

� Academy of Management Journal2008, Vol. 51, No. 5, 976–998.

976

Copyright 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.

companies they funded, and spurring them on ei-ther to fail early or to grow exponentially into busi-nesses that could be sold to stock market investors.Some of the knowledge that venture capitalists(VCs) possess is tacit, and certain VC practicesclash with the compensation systems and organi-zational cultures of the corporations seeking toadopt CVC programs. So corporate venture capitalinvesting programs are hard to design and operate(Gaba & Meyer, 2006). Aggregate levels of equityinvesting fluctuate with economic cycles. Manyfirms that set up CVC units during the 1990s closedthem after 2000 in the face of mounting financiallosses. One critic went so far as to characterize thecontagious spread of CVC programs in the 1990s as“the lemmings’ march toward financial immola-tion” (Edelson, 2001: 41).

When we set out to model the diffusion of cor-porate venturing programs in the information tech-nology (IT) sector, our exploratory analyses showedthat “CVC diffusion could not be understood with-out broadening the scope of observation to encom-pass the larger set of organizational forms and pop-ulations that make up the VC community” (Meyer,Gaba, & Colwell, 2005: 465). Accordingly, our focushere extends beyond the population of firms com-prising the IT industry to include the private part-nerships that pioneered venture capital practices inthe first place. In doing this, we follow the counselof Davis and Marquis (2005: 341) to pursue mech-anism-based theorizing, taking the organizationalfield (rather than the organizational population) asthe unit of analysis. These authors urged research-ers to focus explicitly on organizational fields andsocial mechanisms: “fields because substantial eco-nomic change does not stay contained within or-ganizational or industry boundaries, and mecha-nisms because the quality of explanation isenhanced by an explicit focus on the cogs andwheels behind the regression coefficients” (Davis &Marquis, 2005: 341; emphasis added).

The diffusion mechanisms we propose for CVCprograms are analogous to the transmission of anorganism native to one specific organizational popu-lation (private venture capital partnerships) acrossthe species barrier, into another organizational popu-lation (publicly held information technology corpo-rations). The spatial or geographic diffusion of ven-ture capital between regions and nations has beeninvestigated extensively in prior work (Bruton,Fried, & Manigart, 2005; Guler & Guillén, 2005;Sapienza, Manigart, & Vermier, 1996). However,the spread of venture capital and other innovativepractices among organizational forms or popula-tions has received scant attention. We focus uponinnovations endemic to an originating population

that mutate and jump to an adjacent organizationalpopulation. We develop theory specifying the so-cial mechanisms involved, and hypothesize theirdirect and interactive effects upon innovationadoption. Conceiving of diffusion as a social pro-cess directs one’s attention to the role that propin-quity plays in facilitating the transfer of knowledge.Accordingly, in spelling out the cogs and wheelsinvolved in CVC diffusion, we draw upon the lit-erature on knowledge spillover within geographi-cally proximate regional clusters (Jaffe, Trajtenberg,& Henderson, 1993; Saxenian, 1994).

We test our predictions by studying the bundle ofinnovative practices developed within clusteredenclaves of private VC partnerships that penetratedand spread within the population of publiclytraded IT corporations. First, we explicate andmodel the mechanisms through which VC practicescome to be modified for and adopted by IT corpo-rations. This argument focuses upon cross-popula-tion contagion arising from IT firms’ social ties andproximity to the VC population clusters, coupledwith evidence documenting the efficacy of VCpractices. We proceed to examine a second set ofcontagion processes, within-population contagion,that transmitted the modified VC model from earlyIT adopters to later IT adopters. Here, we focus onCVC programs’ popularity within the IT industry,the prominence of prior IT adopters, evidence bear-ing on benefits accruing to IT adopters, and geo-graphic proximity to prior IT adopters. Finally, weconsider how these cross-population and within-population contagion processes interact.

Our article makes three contributions. First, byinvestigating the spread of corporate venture capi-tal, we shed light on the swift but erratic diffusionof this novel form of strategic investing. Second,our analyses illustrate how innovations penetratethe boundaries separating organizational popula-tions—a relatively unexplored issue in the innova-tion diffusion literature. By disentangling cross-and within-population contagion mechanisms, weoffer a more nuanced understanding of the dynam-ics of innovation diffusion. Third, we focus atten-tion on the spread of “endemic” innovations, show-ing that geographic proximity to the originatingpopulation may be especially crucial in accelerat-ing the diffusion of such innovations.

INNOVATION, CONTAGION,AND POPULATION

Endemic Innovations

In biology and ecology, “endemic” means exclu-sively native to a place or biota, in contrast to var-

2008 977Gaba and Meyer

ious terms signifying “not native,” such as “exotic,”“alien,” “introduced,” or “naturalized.” A speciesthat is endemic is unique to that place or region,found naturally nowhere else. In this article, weapply the term “endemic innovation” to an inno-vation that, when first introduced, is exclusivelynative to a specific organizational population.

Many innovations, of course, spread readily amongpopulations. The boundaries separating organization-al populations have done little to slow the contagiousspread of equipment-embodied innovations like per-sonal computers, or innovative financial arrange-ments like golden parachutes and poison pills (Davis& Greve, 1997). Other innovations, however, areturned back at the population border. Innovations canremain endemic to the organizational populationconstituted by an industry, geographic region, nation-state, or cultural system. Potential barriers to extra-population contagion of endemic innovations in-clude institutional norms, industry standards,technology platforms, regulatory regimes, and cul-tural practices. However, some endemic innova-tions undergo changes that breach contagion barri-ers, allowing the innovations to take hold and diffusewithin new organizational populations. One exampleis the computer software industry’s “open-source”business model, whereby source code is shared freelyand all programmers are invited to contribute to itsfurther development (Wayner, 2000). After remainingendemic to software development for over a decade,the open-source model has “morphed” into variantsthat have been adopted by biotech research labs, in-dustrial design firms, and NASA’s mission to Mars(Goetz, 2003). Other examples are the “lean produc-tion” model that originally was endemic to Japanesemanufacturing (Womack, Jones, & Roos, 1990), andthe microfinancing model, endemic to nongovern-mental organizations but currently diffusing rapidlyin the commercial banking sector (Armendariz deAghion & Morduch, 2005). All of these innovationswere initially native to a defined organizational pop-ulation but underwent modification, enabling theirdiffusion into other populations.

We turn now to the set of innovative practicesdeveloped by private venture capital firms, argu-ing that they constitute an endemic innovationthat has undergone modification, enabling theinnovation’s adoption within information tech-nology corporations.

Venture Capital: The Genesis of anEndemic Innovation

The venture capital model incorporates a novelset of practices for finding, financing, growing, andcommercializing entrepreneurial start-ups (Gomp-

ers & Lerner, 2001b; Kenny & Florida, 2000). Ven-ture capitalists are professional investors who raisefunds from wealthy individuals, insurance compa-nies, pension funds, and other institutions wishingto take equity positions in entrepreneurial venturesbut lacking the ability to identify, manage, andharvest these investments themselves. VCs win-now, select, and take equity in young companieswith the potential to grow exponentially andachieve a dominant market position. The VCs scru-tinize an investment prospect’s business plan, tech-nology, intellectual property, and managementteam; invest their financial capital; and often be-come actively involved in advising, monitoring,and building the company. To mitigate the extremerisks attendant to investing in early-stage compa-nies, venture capitalists deploy their funds acrossportfolios of young companies and often syndicatetheir deals by coinvesting alongside other VCs.

Since there is no market for unregistered securi-ties, venture returns must be realized eitherthrough the acquisition of a VC-backed start-up byan established corporate buyer, or through the is-suance of shares in an initial public (stock) offering(IPO). The IPO is the outcome preferred by all par-ticipants; it achieves the highest valuation for theyoung company, provides liquidity to the inves-tors, and preserves the company’s independence(Gompers & Lerner, 1999).

Established financial institutions did not con-ceive the VC model. Instead, it emerged organicallywithin informal networks of geographically clus-tered private investors. The origins of venture cap-ital can be traced to American Research and Devel-opment in 1946 (Gompers & Lerner, 1999). The firstventure capital partnership was formed in 1958 byDraper, Gaither, & Anderson (DGA; Gompers & Le-rner, 1999). DGA’s enduring legacy was to establishthe limited partnership as an organizational form(Kenney & Florida, 2000), incorporating a distinc-tive set of practices for finding, funding, and com-mercializing novel technologies (Gompers & Le-rner, 2001b). The challenges faced in mobilizingcapital and the need to share information and ex-pertise led pioneering venture capitalists to joinforces, evolving into an interactive community ex-changing information, advice, gossip, and referralsof opportunities (Bruton, Fried, & Manigart, 2005;Kenney & Florida, 2000).

Forty years later, geographic clustering remains adistinctive feature of the venture capital population(Cook, 2001; Sorenson & Stuart, 2001). As sug-gested in Figure 1, almost two-thirds of all VCinvestments in the United States were concentratednear California’s Silicon Valley, New York City,

978 OctoberAcademy of Management Journal

and New England’s Route 128 during the time pe-riod of this study.

The Haphazard Diffusion of Venture Investingto Corporations

Over the years, corporations have tried repeat-edly to stimulate innovation and growth by sepa-rating their new business endeavors from their cur-rent business structures. Often, the results havebeen disappointing (Chesbrough, 2000). Many ofthe best corporate ideas languished or were commer-cialized only when defecting employees foundednew firms (Gompers & Lerner, 2001a). In the late1960s, a number of forward-looking corporationssought to emulate the venture capital model by mak-ing equity investments in external start-ups. By theearly 1970s, over 25 percent of the Fortune 500 hadinitiated venture capital programs (Rind, 1981). Butin 1973, the market for public stock offerings de-clined abruptly, the pool of private venture capitaldried up, and corporations began scaling back theirown venture capital programs. Corporate venturecapital efforts that began in the 1960s were typicallyshuttered after operating just four years (Gompers &Lerner, 2001a). Some private VCs labeled CVC pro-grams “fair-weather investors”; others scorned themas “dumb money,” because of their propensity tooverpay for deals pursued for strategic rather thanfinancial reasons (Alistair, 2000). Academic research-ers highlighted a number of factors thought to havecontributed to CVC program abandonment: poorlyarticulated venturing strategies (Seigel, Seigel, & Mac-Millan, 1988), insufficient top management commit-ment (Sykes, 1990), incompatible organizational cul-tures (Rind, 1981), and inadequate compensationschemes (Block & Ornait, 1987). Some observers con-

cluded that the venture capital model was, in effect,intrinsically endemic to the private VC population(Edelson, 2001).

Corporate Venturing in the 1990s

In the mid 1980s, disappointing returns from in-ternal R&D and shifts in federal antitrust policy ledmany firms to begin to externalize their R&D oper-ations (Mowery, 1999). Meanwhile, Silicon Valleyventure capitalists were gaining celebrity status byspotting promising business opportunities, acceler-ating the progress of new ventures through theirearly development, and helping these venturesachieve liquidity. By the late 1990s, the cachet ofventure-backed firms like eBay and Yahoo! hadreignited corporate interest in the VC model(Gompers & Lerner, 2001a).

This boom in venture capital was unprecedented(Gompers & Lerner, 2001b). Researchers calculatedthat a dollar invested in venture capital was, onaverage, three or four times as potent in stimulatinginnovation as a dollar invested in traditional R&D(Kortum & Lerner, 2000). Total VC investmentssurged from less than $4 billion in 1994 to $43billion in 1999. The corporate share of venturecapital investments increased even more dramati-cally, growing from 2 percent in 1994 to 17 percentin 2000. By mid 2000, some 350 corporate venturecapital funds were reported to exist worldwide, upfrom less than a dozen in 1990 (Venture Econom-ics, 2001).

During the 1990s, information technology corpo-rations were in the vanguard of CVC program adop-tion. Figure 2 shows that the IT sector was thedestination for over two-thirds of all venture capi-tal investments. Digital Equipment Corporation,

FIGURE 1Geographic Distribution of VC Investments in Millions of Dollars

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001New York

MassachusettsCalifornia

0

5

10

15

20

25

30

35

40

45

Percentage of VCInvestments

Year

State


Apple, Intel, Compaq, and Sun Microsystems areprominent examples of firms that received venturecapital early in their development. Given the cen-tral role of venture capital in funding IT, it waslogical that these firms pioneered efforts to assim-ilate the VC model into their own operations.

Modifying the VC Model

We define the adoption of a corporate venturecapital program as occurring when an establishedcorporation creates a structurally distinct entitydedicated to making external equity investments ina portfolio of high-potential young enterprises.CVC units can assume a variety of structural forms(Gaba & Meyer, 2006; Gompers, 2002), which in-clude autonomous subunits reporting to the corpo-ration’s top management team, limited partner-ships whose sole limited investor is the parentcorporation, and durable alliances between the cor-poration and a private VC firm intended to assem-ble a dedicated portfolio of investments tailored tofit corporate strategic objectives.

Many innovations must be modified to fit anadopting organization’s idiosyncratic context(Abrahamson, 2006; Rogers, 1995), and innova-tions endemic to an alien organizational form callfor even more extensive modification. Althoughthey are modeled upon the practices of private VCfirms, corporate programs have distinctive features.First, unlike private VCs, CVCs do not raise fundsfrom institutional investors; in most cases they tapthe corporate treasury to fund their CVC programs(Chesbrough, 2000). Second, unlike private VCs,whose central objective is to maximize financialreturns, most corporations pursue strategic objec-

tives as well: early exposure to disruptive technol-ogies, access to new markets and business models,and identification of prospective targets for acqui-sition (Chesbrough, 2000; Dushnitsky & Lennox,2006). Third, unlike private VCs, CVC staff seldombecome the lead investors in syndicated invest-ments or take seats on boards of start-up ventures,preferring to coinvest alongside private VCs wholead deals and take board seats (De Clercq, Fried,Lehtonen, & Sapienza, 2006). Fourth, financial re-turns to private VC partners come primarily fromcashing in equity stakes in portfolio companies, butcorporations cannot devise compensation schemesoffering equivalent payouts without creating intrac-table conflicts of interest and wreaking havoc oninternal corporate incentive systems (Block & Or-nait, 1987).

These differences notwithstanding, CVC pro-grams retain key operational characteristics of theprivate VC model, such as due diligence, deal mak-ing, and syndication. Corporate venture investorscan complement private investors, adding uniquevalue to start-up companies by tapping their deepknowledge of the IT industry and its core technol-ogies, serving as a “beta site,” providing access tomarketing and distribution channels, and confer-ring legitimacy and gravitas upon a new venture(Chesbrough, 2000; DeClercq et al., 2006).

In sum, innovative investing practices devised byprivate venture capital partnerships in the 1950sremained, by and large, endemic to the VC popu-lation until the 1990s, when these practices, repur-posed to pursue strategic in addition to financialobjectives and remodeled to fit corporate struc-tures, began diffusing to public corporations, withIT firms in the vanguard. In the next section, we

FIGURE 2Industry Distribution of VC Investments in Millions of Dollars

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001Other industries

Biotechnology and healthInformation technology

0

10

20

30

40

50

60

70

80

90

Percentageof VC

Investments

Year

Industry


propose theoretical mechanisms responsible fortheir cross-population diffusion and subsequentwithin-population diffusion.

THEORY AND HYPOTHESES

How and why do novel practices spread amongthe actors who make up a social system? Reviews ofthe diffusion literature show that few researchquestions have spanned so many social sciencedisciplines, inspired such an outpouring of re-search, or aroused such enduring interest (Rogers,1995; Strang & Soule, 1998). Most of this work“treats diffusion as a primarily, or even exclusively,relational phenomenon” (Strang & Meyer, 1993:487) whose fundamental rate varies with the levelof connectedness and interaction between priorand potential adopters of an innovation. Yet empir-ical findings are equivocal, and scholars’ cumula-tive knowledge of why and how innovations cometo be adopted “is considerably less than the sum ofits parts” (Meyer & Goes, 1988: 897).

Because purely relational models have yieldedunsatisfactory explanations, a number of scholarshave incorporated contextual factors thought to af-fect diffusion. For instance, the rate and form ofdiffusion have been said to be shaped by institu-tional and cultural conditions (Strang & Meyer,1993), network structures of regional geographicclusters (Saxenian, 1994), fieldwide macrocultures(Abrahamson & Fombrun, 1994), and the influenceof mass media (Strang & Macy, 2001).

A number of prior studies have addressed thespread of organizational innovations across geo-graphic space (e.g., Greve, 2005). In the case ofventure capital, after originating in the UnitedStates, the model spread first to England and West-ern Europe in the late 1970s, and then to Asia in themid 1980s (Bruton et al., 2005; Sapienza et al.,1996). Researchers have reported that the geo-graphic dispersion of VC seems to be linked todifferences in national innovation systems, financialmarkets, political institutions, and social proximity(Guler & Guillén, 2005). Venture capitalists’ depth ofinvolvement in portfolio companies and beliefs abouthow much value they add have been found to varyacross countries (Sapienza et al., 1996).

Saxenian (1994) argued that differences in re-gional network structure explain why Silicon Val-ley adapted more effectively to technologicalchange than Boston’s Route 128. Suchman and Ca-hill (1996) described how law firms boosted SiliconValley’s entrepreneurial vitality by helping forgethese network relationships in the first place andinstitutionalizing “term sheets”—the contractual

covenants that codify and accelerate new firmformation.

Although social scientists have studied the dif-fusion of innovation among geographic popula-tions, they have not considered how the character-istics of organizational populations influence thecontagious spread of innovations. No studies havescrutinized innovations that emerge as endemic toa specific organizational population of adopters,examined how such innovations breach populationboundaries to spread into adjacent organizationalpopulations, or specified the different contagionmechanisms underlying within- and between-pop-ulation diffusion.1

In this section we develop such a model. Ourobjective is to understand the transfer of innovativepractices originating elsewhere into the organiza-tions that comprise a separate and distinct organi-zational population. Our model distinguishes be-tween two forms of contagious diffusion: One formoperates across the boundary separating two organ-izational populations, and the other operateswithin a single population. Cross-population con-tagion transports an endemic innovation from anoriginating population, where it was originally con-ceived, into an adopting population, where it issubsequently adapted and implemented. Within-population contagion spreads a formerly endemicinnovation from prior adopter to prospectiveadopter within a new organizational population.Figure 3 illustrates our model, showing the prin-

1 Community ecologists and institutional theoristshave long been concerned with how two or more organ-izational populations interact in organizational fields,but neither group has focused upon innovation diffusionas a core mechanism linking two populations.

FIGURE 3Cross-Population Diffusion

of Endemic Innovations


cipal mechanisms we consider, and indicatinghow these map onto the empirical setting westudied.

Cross-Population Diffusion of Innovations

How do innovative practices endemic to one or-ganizational population spread into another popu-lation? We focus on two mechanisms: direct expo-sure to the innovation afforded by geographicproximity to the originating population, and evi-dence of beneficial consequences flowing from theinnovation.

Geographic proximity to the originating popu-lation. The role of geography in structuring inter-actions and shaping strategic decisions has longinterested sociologists and economists. Geo-graphic clustering creates external economiesarising from information spillovers and access tospecialized services and skilled labor (Audretsch& Feldman, 1996; Krugman, 1991). Researchershave argued that geographic clusters facilitate thetransfer of knowledge between firms operatingwithin a region (Jaffe et al., 1993; Saxenian, 1994;Sorenson & Audia, 2000). These knowledge spill-overs occur as firms collaborate, as their mem-bers interact at social and professional gather-ings, and as workers leave one firm to work foranother. Saxenian (1994) reported that in SiliconValley, geographic proximity has promoted re-peated interactions that build trust, foster collab-oration, and fuel a continual recombination oftechnologies and skills.

Geographic proximity to a VC population clusteraffords a prospective IT adopter direct exposure tothe innovation itself, in situ. It allows members toobtain first-hand knowledge about the VC model—the requisite skills, values, procedures, behaviors,and know-how involved (Audretsch & Feldman,1996). Proximity helps foster social relationshipsthat boost the prospective adopter’s confidence inthe accuracy and quality of innovation-related in-formation (Owen-Smith & Powell, 2004; Sorenson& Audia, 2000). We expect that IT firms headquar-tered near a venture capital cluster will find iteasier to acquire both tacit and codified knowledgeabout VC practices. Such firms are likely to be morecognizant of the “agency risks” implicit in fundingstart-ups, and better equipped to monitor theirportfolio companies to mitigate such risks. In addi-tion, executives of proximate firms are likely to beembedded in social networks that include privateventure capitalists who can feed them an ongoingflow of deals.

Accordingly, we hypothesize that variation in ITfirms’ proximity to a VC population cluster creates

differentials in the amount and quality of informa-tion they acquire, leading to variation in the likeli-hood of adoption of CVC programs.

Hypothesis 1. Geographic proximity to a VCpopulation cluster increases the probability ofCVC program adoption by an IT firm.

Observing success in the originating popula-tion. So far, our argument has emphasized directexposure to innovative practices and their pioneer-ing practitioners as the mechanism enabling an en-demic innovation to penetrate and spread within anew organizational population. However, mostfirms presumably adopt innovations with the ex-pectation of achieving substantive benefits. If so,vivid and reliable evidence that an endemic inno-vation is paying off for firms in another populationshould persuade performance-oriented decisionmakers to consider adopting it themselves. Surpris-ingly, mainstream diffusion theories have largelyignored how prospective adopters are influencedby the results achieved by prior adopters in otherpopulations. However, it was the discovery of newdrugs by small biotech start-ups that led large phar-maceutical firms to adopt recombinant DNA tech-nology (Barley, Freeman, & Hybels, 1992). Simi-larly, the microcredit investing model spread fromnongovernmental organizations (NGOs) into the fi-nancial services sector in the wake of a 1997 UnitedNations study reporting that default rates of impov-erished borrowers in developing countries werecomparable to or lower than the rates for traditionalbanks (United Nations, 1997).

The impacts of VC investment on new ventureformation, value creation, and regional economicdevelopment received widespread publicity in the1990s. Although venture capital firms invest inonly a few hundred of the nearly one million newU.S. businesses established each year, roughly one-third of those that went public in the past twodecades were backed by venture capitalists (Gomp-ers & Lerner, 1999). Despite their small numbers,VC-backed companies often prove to be well-springs of technological innovation and continue tooutperform non-venture-backed companies long af-ter they go public (Brav & Gompers, 1997). Kortumand Lerner (2000) found that although VC invest-ment in biotechnology has averaged less than 3percent of R&D investment, firms with VC backingproduced nearly 15 percent of biotech innovations.

The initial public offering, as noted earlier, pro-vides the most desired and closely monitored sig-nal of the success of a venture capital investment.The number of IPOs varies from year to year, drivenby movement through aggregate venture capital cy-cles and rates of innovation in technology and busi-


ness models. Thus, the incidence of recent VC-backed IPOs provides a salient indicator of theoverall success of the private venture capitalmodel, attracts the attention of potential corporateadopters, and increases their receptivity to estab-lishing a CVC program.

Hypothesis 2. The likelihood of CVC programadoption by an IT firm is positively related tothe efficacy of the investments made by the VCpopulation.

Within-Population Diffusion of Innovations

Once an endemic innovation has been revampedand implemented in a nonnative population, newcontagion mechanisms come into play. This neednot terminate the initial cross-population conta-gion processes, but it supplements them by intro-ducing new channels for the contagious spread ofthe modified innovation. As shown on the rightside of Figure 3, we expect these new within-pop-ulation contagion mechanisms to transmit the re-fashioned CVC model from early IT adopters tolater IT adopters, driven by CVC programs’ popu-larity within the IT industry, the prominence ofprior IT adopters, the strategic returns accruing toIT adopters, and geographic proximity to prior ITadopters.

Popularity among peers. An innovation’s popu-larity within a population, manifested by the num-ber of prior adopters in the population, has beenimplicated in the diffusion of civil service reforms(Tolbert & Zucker, 1983), university-initiated sportsprograms (Washington & Ventresca, 2004), and fad-dish managerial innovations (Abrahamson & Fair-child, 1999). Diffusion scholars have adduced var-ied microprocesses linking popularity to cascadesof further adoptions: neoinstitutionalists focus onmimicry in pursuit of legitimacy (DiMaggio & Pow-ell, 1983); learning theorists focus on vicariouslearning by individual decision makers or entireorganizational populations (Greve, 1995; Hauns-child & Miner, 1997); increasing-returns theoristsfocus on network externalities (Arthur, 1994); andrational-choice theorists treat popularity as an out-cropping of an innovation’s beneficial outcomes(Rogers, 1995).

Drawing on these persuasive arguments and evi-dence, we hypothesize that increases in an innova-tion’s popularity, as reflected in the extent to whichit has already diffused within an organizationalpopulation, are positively related to within-popu-lation contagion.

Hypothesis 3a. The probability of CVC pro-gram adoption by an IT firm is positively re-

lated to the number of prior adopters in the ITindustry.

Prominence of similar adopters. Prospectiveadopters are especially likely to be influenced bythe moves of more prestigious organizations withintheir population. Adoption of innovations involveshigh uncertainty, but innovations adopted by suc-cessful organizations are viewed as less uncertain,and hence are more likely to be imitated by others(DiMaggio & Powell, 1983). Prominent organiza-tions are also regarded as worthy of imitation be-cause of the legitimacy gains accruing from imitat-ing them (DiMaggio & Powell, 1983; Sherer & Lee,2002). Organization-level traits such as size andprofitability are often used to infer social promi-nence (Fombrun & Stanley, 1990; Haunschild &Miner, 1997; Haveman, 1993). Numerous studieshave reported that large and profitable firms areimitated more frequently (Greve, 1995; Haunschild& Miner, 1997; Mezias & Lant, 1994). Lant andBaum (1995) found that Manhattan hotel managerspaid closer attention to the strategic actions oflarger, more luxurious hotels. Other studies havereported similar results for savings and loan asso-ciations entering new markets (Haveman, 1993),choices of investment banks brokering acquisitions(Haunschild & Miner, 1997), and choices of branchlocation (Greve, 2000). Accordingly, we hypothe-size that IT firms seek to emulate their respectedpeers, rendering innovations adopted by presti-gious firms especially contagious.

Hypothesis 3b. The probability of CVC pro-gram adoption by an IT firm is positively re-lated to the extent of prior adoption amongprominent IT firms.

Observing peers’ success. Earlier, we arguedthat firms notice and react to the benefits realizedby adopters within the population where an inno-vation originated. But it would seem that firmsshould pay even closer attention to benefits accru-ing to prior adopters within their own population.Organizational learning theorists have argued thatorganizations learn vicariously, imitating or avoid-ing specific innovations on the basis of their per-ceived impact elsewhere (Levitt & March, 1988).Evidence for the salience of tangible benefits real-ized by similar others comes from Conell and Cohn(1995), who found that the success of wildcatstrikes by French coal miners increased thechances of future strikes. Haunschild and Miner(1997) reported that investment bankers brokeringacquisitions that generated higher returns weremore likely to be retained in subsequent acquisi-tions. Other indirect evidence comes from studies


suggesting that favorable results achieved by earlierentrants have encouraged investors to build ship-yards (Argote, Beckman, & Epple, 1990), hotels(Baum & Ingram, 1998), and nuclear power plants(Zimmerman, 1982). In sum, both logic and evi-dence suggest that beneficial results obtained byprior similar adopters will lead prospective adopt-ers to anticipate similar results:

Hypothesis 3c. The probability of CVC programadoption by an IT firm is positively related toefficacious outcomes experienced by IT firmsthat are prior adopters.

Geographic proximity to similar adopters. Anaxiomatic premise of diffusion theory is that proxim-ity facilitates interaction that spreads innovations(Strang & Soule, 1998). Personal observations andcommunication are stronger mediators of informationover short rather than long distances owing to thehigher density of contacts and information spillovers(Greve, 1998). Once the assimilation of an innovationdeveloped elsewhere has opened up the possibility ofwithin-population contagion, proximity to otheradopters within that population is likely to accel-erate the innovation’s transmission. Proximity fa-cilitates “sensemaking” within communities ofpractice (Brown & Duguid, 1991), fuels pressures formimetic adoption, and fosters vicarious learningthrough observation of innovative practices and theiroutcomes in similar settings. Stronger contagion ef-fects over smaller geographical distances have beenreported for such innovations as matrix management,golden parachutes, and unionization (Burns &Wholey, 1993; Davis & Greve, 1997; Hedstrom, 1994).Accordingly, we expect spatial proximity to previ-ously adopting population members to speed an in-novation’s diffusion.

Hypothesis 3d. The probability of CVC programadoption by an IT firm is positively related to thenumber of proximate prior IT adopters.

The Joint Effects of Within- andCross-Population Contagion

We have argued that cross-population contagionexports endemic innovations to a new organization-al population, and within-population contagionpropagates the modified innovation within thatnew population. So far, our arguments have iso-lated the two contagion processes, focusing on themain effects of each. We now consider how thesetwo contagion processes interact when they operatein tandem. After a formerly alien innovation mu-tates and becomes native, does cross-populationcontagion persist or does it shut down, giving way

to within-population contagion? We draw uponneoinstitutional theory to develop predictionsabout the joint effects of cross-population and with-in-population contagion, proposing that the level of“theorization” of the innovation governs the rela-tive potency of these two effects.

Theorization of innovations. Strang and Meyer(1993) argued that diffusion is directed and accel-erated by theorized accounts of the actors and prac-tices involved in an innovation. They defined the-orization as “the self-conscious development andspecification of abstract categories and the formu-lation of patterned relationships such as chains ofcause and effect” (Strang & Meyer, 1993: 492). Thetheorization of an innovative practice simplifiesand abstracts its properties, explains the outcomesit produces, facilitates communication and inter-pretation, and thus expands and accelerates its dif-fusion. The fundamental claim is that “practices donot flow . . . theorized models and careful framingsdo” (Strang & Soule, 1998: 277).

Diffusion of well-theorized innovations entails“translating concrete practices into abstractions forexport and then unpacking the abstractions into a(suitably modified) concrete practice upon arrival”(Strang & Soule, 1998: 276). Diffusion is expeditedby management consultants who “arrive in organi-zations like traveling salesmen with attaché-casesfull of quasi-objects to be translated into localizedideas” (Czarniawska & Joerges, 1990: 36). Encapsu-lating innovations in higher levels of abstractionaccelerates the speed and reach of diffusing prac-tices, because universal theories predict that thepractices can be adopted by many or all organiza-tions to produce similar outcomes. The spread ofhighly theorized innovations becomes less depen-dent upon proximity-induced social interaction,because theorization substitutes for direct experi-ence and encodes practices in a “language that doesnot presume directly shared experience” (Strang &Meyer, 1993: 499). Conversely, innovations that arepoorly or thinly theorized should diffuse moreslowly, rely more heavily upon relational transmis-sion, and encounter difficulty in breaching theboundaries of the population wherein they arose.

During the 1990s, venture capital practice wasnot highly theorized.2 Venture capital practicescombine VCs’ tacit knowledge, expertise, and ex-

2 Although the venture capital limited partnership hasbecome institutionalized as an organizational form (Bur-ton, Fried, & Manigart, 2005), the underlying practicesremain comparatively tacit, subsuming practitioners’ in-tuition, judgment, and skills acquired and honed on thejob.


perience (Wasserman, 2005). Many pioneering VCswere iconoclastic entrepreneur-investors, unable orunwilling to codify and formalize their modus ope-randi. Venture capitalists often dub the investmentsof those within their clan as “smart money”—moneythat comes infused with the entrepreneurial acumen,business contacts, executive talent, and patience of afinancier who knows how to help young companiessucceed (Doerflinger & Rivkin, 1987: 16). In con-trast, an informant in our research scorned corpo-rate venture investors as “dumb money—suckerswho overpay for deals because they are in it forstrategic returns.” This, according to our informant,made them useful “only for goosing up your fund toshow a humongous internal rate of return to snowyour limited partners, and make sure they pony upfor the next fund you’re raising.” Disdaining corpo-rate investors as amateurs whose investment strat-egy was to “spray and pray,” this VC deliberatelyavoided the theorization of his practices. A veterancorporate investor offered a complementary view:

Private VCs are stable and they’re cloaked in se-crecy. It’s a club that runs on trust, and that trust canbe established because the half-life of your networkis long. Corporate venturing is cyclical and unstable.The investors are inexperienced and naı̈ve. At [mycorporation] we’ve been doing this for a long time,and we’ve cracked the code. We’ve figured out howto do deals by making it simpler, and making itquicker.

Our claim that tacit knowledge underpins VCpractices was substantiated by an early-stage ven-ture capitalist:

These relationships are literally foreign to lots ofcorporations. When I’m working in Silicon Valley, Iknow who’s involved, and everybody knows what’scoming next. It’s like playing baseball—everybodyknows the rules. But in the corporate world, therules aren’t as well understood, the players keepchanging, and the team gets sold or moves unpre-dictably. I’ve waited three weeks to wrap a dealbecause the corporate guy who has to sign off is onsafari in Africa.

Because poorly theorized venture capital prac-tices are encoded in tacit knowledge rather thanauthoritative theorizations, we expect that whengeographic proximity enables frequent social inter-action between VCs and corporate investors, cross-population contagion will remain the prepotentmechanism driving IT corporations’ CVC adop-tions—more potent than the factors fueling within-population contagion. Geographic proximity fos-ters social relationships that have been found toincrease both the speed and fidelity of tacit knowl-edge transfers (Greve, 2005; Jaffe et al., 1993; So-

renson & Audia, 2000). Accordingly, we expect ITfirms situated near a VC cluster to tap cross-popu-lation channels in reaching CVC adoption deci-sions, while paying less attention to the adoptionbehavior, prominence, and outcomes experiencedby their IT industry peers. On the other hand, weexpect prospective IT adopters whose geographicisolation precludes direct and inductive examina-tion of VCs’ innovative practices to monitor theirindustry peers’ behaviors and actions more keenly.For far-away firms, we expect within-populationcontagion mechanisms to predominate, with adop-tion chances driven by CVC programs’ popularitywithin the IT sector, their adopters’ stature, andoutcomes that materialize. Stated more formally,we hypothesize that a prospective IT adopter’s geo-graphic proximity to a VC cluster will moderate theeffects of within-population contagion upon thelikelihood of adoption.

Hypothesis 4. IT firms that are geographicallyproximate to a VC population cluster are lessaffected by the adoption behavior of other ITfirms than those that are geographically dis-tant from a VC cluster.

METHODS

Sample and Data

To test our hypotheses, we gathered longitudinaldata from a sample of U.S. firms in the informationtechnology sector. Our data cover 264 IT firms from1992 to 2001. Ninety-four of these firms adoptedCVC programs during the ten-year time period ofthe study. We focused exclusively on IT firms tocontrol for unobserved heterogeneity at the indus-try level, and because many of the venture capitalinvestments during the time period of the studywere made in the IT sector, rendering IT firms morelikely adopters of CVC programs. The study’s sam-ple consists of IT firms drawn from the Forbes 500list, which ranks U.S. firms by sales, profits, assets,and market value. Firms that rank among the top500 on one or more of these criteria are included inthe Forbes list, which is compiled annually. Theresearch sample was constructed in two steps: (1)the names of all firms listed on the Forbes 500between the years 1997 and 2000 were compiled,and (2) firms in the information technology sectorwere selected from this list. We followed the Na-tional Science Foundation’s definition of the ITsector (NSF, 2000) to include firms in five industrysubsectors: (1) office, computing and accountingequipment (SIC code 357), (2) communicationsequipment (SIC code 366), (3) electronic compo-nents (SIC code 367), (4) communication services


(SIC codes 481–484, 489), and (5) computing anddata processing services (SIC code 737).

Model Estimation

We used a discrete-time event history methodol-ogy to model the adoption of CVC programs (Alli-son, 1982). We estimated Pi(t), the conditionalprobability that firm i adopts a CVC program at timet. Pi(t) is related to the covariates by a probit regres-sion equation,

Pi(t) � �� 1xi1(t) � . . . � �kxik(t)�,

where � is the cumulative density function of thenormal distribution and xi’s are covariates that maybe time-invariant. This methodology is preferredwhen information on the exact timing of an event isunavailable—that is, when interval “censoring” ex-ists (Allison, 1982). In our case, exact adoptiondates are not known, since we only have annualdata. A second advantage of this method is thatfirms that do not adopt contribute to the regressionmodel exactly what is known about them; right-censoring is moot (Allison, 1982). Finally, time-varying explanatory variables are easily includedbecause each period during which a firm is at riskis treated as a separate observation. Left-censoringposes no problem, since only one of the 264 firmsin our sample had established a CVC program in1991. We assumed that the cumulative densityfunction for the error term F(.) was normally dis-tributed, so we used a probit model to estimate theprobability of an adoption event in a given year, ina pooled sample consisting of each organizationobserved during each of the ten years.3

Measures

Dependent variable. Our dependent variable isthe probability that a firm will adopt a CVC pro-gram at time t. To operationalize it, we relied uponthe Corporate Venturing Directory and Yearbookfor 2000, 2001, and 2002. Among other data, this

directory reports the year in which firms adoptCVC programs. For confirmation and to obtainmissing data, we turned to the VentureXpert data-base, IT firms’ websites, and industry publications.In open-ended interviews, corporate venture capi-talists told us that when firms make multiple directinvestments in technology start-ups within a singlecalendar year, they nearly always house their ven-ture capital activity within a formal structure. Fol-lowing their recommendation, we treated any firmmaking five or more direct investments within asingle year as having adopted a CVC unit in thatyear.4 We confirmed that these companies had ac-tually adopted a CVC program by examining com-pany websites, through Lexis-Nexis searches ofbusiness press articles and venture capital newslet-ters, and finally, by direct communication with thecompanies’ business development executives.

Independent variables. To test Hypotheses 1and 2, we calculated a weighted average of geo-graphic distance to the three main VC clusters (Sil-icon Valley, New York, and Route 128), andcounted the number of VC-backed IPOs in the ITsector. We designated these two variables as cross-population contagion measures.

Silicon Valley houses the world’s dominant clus-ter of private venture capital firms. In 2000, approx-imately 40 percent of all U.S. venture capital orig-inated in Silicon Valley. The other two importantregions are New York and Route 128 in New En-gland, accounting for 12 and 11 percent of VC in-vestment, respectively. We measured geographicdistance to VC clusters as a weighted average of thenumber of miles from corporate headquarters toeach of these three clusters. The weights used werethe density of VC funds targeting IT start-ups ineach cluster, lagged by one year. Thus, the distance offirm i to VCs at time t (dit) was calculated as dit �¥j � SV, 128, NYdij�jt � 1, where dij is the distance betweenfirm i and cluster j (j � Silicon Valley, Route 128, orNew York) and �jt � 1 is the proportion of VC fundsin cluster j at (t – 1) with ¥j � SV, 128, NY�jt � 1 � 1. Thezip code for corporate headquarters was obtainedfrom the Forbes lists. We classified the countiesAlameda, Contra Costa, Marin, San Francisco, SanMateo, and Santa Clara as comprising Silicon Val-ley; Essex, Middlesex, Suffolk, and Norfolk as com-prising Route 128; and New York, Bronx, Kings,Queens, and Richmond as comprising New York.We measured the geographic distance to each clus-

3 Unobserved heterogeneity leading to potentially bi-ased estimates, a concern for all estimation techniques, isparticularly problematic in event history analysis. Fixed-effects analyses have been shown to introduce inconsis-tent estimates (Chamberlain, 1985). In our data, random-effects estimates provide results identical to pooledestimates, with the former exhibiting higher standarderrors. A test of poolability yields a p-value of 1, indicat-ing that pooled probit is appropriate and that unobservedheterogeneity through the random-effects componentmay be ignored.

4 We experimented by moving this threshold from fiveto ten investments in a single year (a more conservativeparameter). This changed the CVC adoption date for 7 outof the 94 adopters but left our results unaffected.


ter as the number of miles from headquarters to themost proximate of the counties in a particular clus-ter, using a spherical geometry formula to calculatethe distance between zip codes (Sorenson & Stuart,2001):

Distance in miles�3,963.0 �arccos[sin(zip1.lat)

� sin(zip2.lat) � cos(zip1.lat) � cos(zip2.lat)

� cos(zip2.lon � zip1.lon)],

where zip i.lat is latitude of zip i � 1, 2 and zip i.lonis the longitude of zip i � 1, 2.

We measured the observed benefits of VC prac-tices by counting the number of VC-backed IPOstendered in the IT sector during the previous cal-endar year (Gompers & Lerner, 1998). IPOs providean unequivocal measure of private venture capital-ists’ value creation. Venture capital firms’ pressreleases herald their investing acumen by high-lighting portfolio firms that have “gone public”(Gompers & Lerner, 2001b). Unlike VC firms, ITfirms do not focus exclusively on financial returns,but the incidence of VC-backed IPOs indexes thereturns generated by the venture capital model andsignals to prospective IT adopters the potential ef-fectiveness of VC practices in commercializing in-novative technologies (Gompers & Lerner, 2001a).This measure was obtained from VentureXpertdatabase.

Drawing upon the institutional and innovationdiffusion literatures, we constructed multiple mea-sures of within-population contagion influences.Specifically, we used the popularity of the innova-tion, prominence of prior adopters, outcomes expe-rienced by prior adopters, and proximity to prioradopters to capture the intensity of within-popula-tion contagion (Greve, 1995; Haunschild & Miner,1997). Prior adopters were defined as IT firms thathad adopted a CVC program during or before theprevious year.

Following the diffusion literature (Haunschild &Miner, 1997; Kraatz & Zajac, 1996; Rao, Greve, &Davis, 2001), we measured the popularity of aninnovation at time t as the cumulated total adoptersup to (t – 1) in a single two-digit industry sector.5

We limited this measure to a two-digit SIC code inview of evidence that potential adopters pay greaterattention to more comparable organizations (Have-man, 1993).

As noted earlier, potential adopters also pay

more attention to prior adopters that have higherstatus. We used the average sales of prior adoptersin the same two-digit industry sector as a measureof their prominence (Greve, 2000; Haunschild &Miner, 1997; Haveman, 1993). We called this vari-able prominence of prior adopters.

Although VCs realize returns on their invest-ments by taking portfolio companies public, corpo-rations seek to achieve strategic returns in additionto financial returns. CVC programs enable their ITcorporate parents to screen entrepreneurial start-ups at the forefront of emerging technologies thathave the potential to change the competitive dy-namics of the industry (Chesbrough, 2003). If astart-up proves strategically relevant, the parentfirm is well positioned to acquire it. Accordingly,the rate at which firms select ventures from theirCVC portfolios as acquisitions is an indicator of thevitality of CVC programs (Gompers & Lerner, 1999).We used the cumulated number of CVC-backedacquisitions as a measure of the benefits realized byadopters of CVC programs. Although CVC-backedacquisitions are not the sole strategic objective of ITfirms, acquisitions are a visible and tangible indi-cator of the benefits accruing from their CVC pro-grams.6 Such acquisitions are likely to be noticedby industry peers and signal the potential value ofCVC programs to prospective IT adopters. We col-lected these data from two sources. First, we usedthe Securities Data Corporation’s (SDC) Mergersand Acquisitions database to obtain a list of allprivate acquisitions by all the CVC adopters in oursample for each of the years from 1992 to 2001. Wethen matched the acquisitions by each adopter withthe VentureXpert database to include only thoseacquisition targets in which CVC adopters had in-vested during the time period of the study. Weidentified a total of 94 CVC-backed acquisitionsmade by the IT firms in our sample. Finally, wecumulated these acquisitions since the year ofadoption to obtain a measure of beneficial out-comes experienced by prior adopters.

We used proximate prior adopters to measurethe availability of innovation-related informationthrough observation of and interaction with indus-try peers. For each firm at time t, we counted the

5 Using a three-digit classification left our results un-affected. A four-digit industry classification schemeproved too restrictive because very few firms were in-cluded in each four-digit classification category.

6 Alternative measures might include the number ofCVC-enabled new product introductions, returns real-ized from such products, and technologies licensed fromportfolio companies. However, such data are not avail-able to us, and IT firms also obtain intangible benefitsfrom CVC programs that this measure fails to capture,such as exposure to new business models. Thus, CVC-backed acquisitions constitutes a conservative measureof the strategic benefits realized from CVC programs.


total number of adopters of CVC programs up totime (t – 1) in the same geographic region as thefocal firm (Burns & Wholey, 1993; Davis & Greve,1997). This procedure allowed us to distinguishproximity to CVC adopters from proximity to theVC cluster, and evaluate their relative importancein driving the adoption process.

Control variables. To isolate theorized variables’impacts upon CVC adoption, we controlled for sev-eral firm-level variables in estimating the model.As they age, organizations acquire both experiencein assimilating innovations (Sorenson & Stuart,2000) and inertia that may impede strategic change(Hannan & Freeman, 1989). To control for age, wecounted years since a firm’s founding, using Stan-dard & Poor’s Million Dollar Directory as a datasource. Because stockpiled slack resources facili-tate experimentation with innovative practices(Levinthal & March, 1981), we controlled for slackby including a firm’s current ratio (Bourgeois,1981), calculated from Compustat data. Past suc-cess in innovating through in-house R&D couldeither increase or inhibit a firm’s likelihood ofadopting a CVC program. Some studies have re-ported that internal innovation capabilities sparkand complement external efforts (Cassiman &Veugelers, 2006; Dushnitsky & Lenox, 2005); othershave reported that internal innovation capabilitiessubstitute for and curb external efforts. We con-trolled for innovation propensity by counting pat-ents awarded to a firm divided by sales (Cohen,Levin, & Mowery, 1987). We used patent applica-tion filing dates to assign a granted patent to afirm in a given year, using data compiled by Hall,Jaffe, and Trajtenberg (2001) that record all utilitypatents granted between January 1, 1963, andDecember 30, 1999.7 Since larger firms haveamassed resources that help them undertake stra-tegic change, we controlled for firm size by mea-suring total corporate sales in the appropriateyear, using Compustat data.

Next, IT firms that themselves were foundedwith venture capital investors may have VC prac-tices imprinted into their organizational routinesand cultures (Stinchcombe, 1965). Such firms seem

likely to have a congenital affinity for VC practicesand to retain social ties to the VC community.Moreover, since VCs prefer to invest in firms lo-cated nearby (Sorenson & Stuart, 2001), these firmsare likely to be headquartered near Silicon Valley,Route 128, or New York, thus confounding esti-mates for our primary independent variable, geo-graphic distance to VC clusters. We controlled forIT firms’ venture-backed origins, using an indicatorvariable, VC-backed at founding, assigned thevalue 1 if a firm itself was backed by private VCfunding prior to its first initial public offering, and0 otherwise, using information from VentureXpert.Finally, since a firm’s decision to adopt a CVCprogram may be influenced by the availability ofopportunities to actually make equity investmentsin start-ups (Ahuja, 2000; Rosenberger, Katila, &Eisenhardt, 2007), we controlled for variation inthe number of such opportunities over time andacross geographic settings. Using the VentureXpertdatabase, we constructed a measure of local invest-ment opportunities for each firm i headquartered in

state k at time t asnt � 1k

¥jnt � 1j

, where j is an index for

states and nt�1j is the number of entrepreneurial

startups that received VC investments in state j attime t – 1. Table 1 provides summary statistics andcorrelations between the predictor variables.

RESULTS

Tables 2, 3, and 4 show the maximum-likelihoodestimates of 13 models predicting CVC programadoption using a discrete-time event history model.Model 1 includes only the control variables—age,slack, innovative propensity, size, venture backing,and density of start-up deals funded in the focalfirm’s state. In line with earlier research on inno-vation (Drazin & Schoonhoven, 1996), we foundthat these firm-level characteristics influence inno-vation adoption. The negative coefficient on ageindicates that younger firms are more likely toadopt CVC programs. The positive coefficients onslack and size indicate that firms with greater ac-cess to slack resources and firms with larger salesrevenues were more likely to adopt CVC programs.8

For innovation propensity, the results show thatfirms that generate more innovations internally(i.e., those having higher patent-sales ratios) areless likely to adopt CVC programs. Pisano (1990)reported a similar finding; firms with in-house R&D

7 Our study period ended in 2001, so we extended theHalle et al. data set by collecting primary data on patentsgranted to sample firms and their subsidiaries during2000–2001 directly from the U.S. patent office website(http://www.uspto.gov). We used three databases (SDC’sMergers & Acquisition Database, Capital IQ, and the Di-rectory of Corporate Affiliations) to identify the subsid-iaries for all IT firms in a given year and assigned utilitypatents granted to these subsidiaries to the IT corporateparent.

8 We used assets as a measure of size as well. However,with a correlation between sales and assets of .95, it madevery little difference which measure of size we used.


TA

BL

E1

Sam

ple

Mea

ns

and

Cor

rela

tion

sa

Var

iabl

eM

ean

s.d

.1

23

45

67

89

1011

1213

1415

16

1.D

ista

nce

toV

Ccl

ust

ers i

t�

1b

5.07

2.41

2.N

um

ber

ofV

C-b

acke

dIP

Os t

b4.

520.

54.0

13.

Pop

ula

rity

ofin

nov

atio

nit

�1

7.51

8.53

�.0

6�

.04

4.P

rom

inen

tp

rior

adop

ters

it�

1b

8.31

2.92

.13

.01

.13

5.O

utc

ome

exp

erie

nce

dby

pri

orad

opte

rsit

�1b

�1.

736.

69�

.04

.12

.48

.24

6.P

roxi

mat

ep

rior

adop

ters

it�

15.

489.

83�

.34

�.0

5.5

6.0

5.3

17.

Pop

ula

rity

ofin

nov

atio

nit

�1

�ge

ogra

ph

icd

ista

nce

toV

Cit

�1b

36.7

448

.23

.35

�.0

2.8

0.1

3.4

1.2

0

8.P

rom

inen

tp

rior

adop

ters

it�

1�

geog

rap

hic

dis

tan

ceto

VC

it�

1b

43.0

124

.74

.89

.01

.00

.55

.08

�0.

27.3

5

9.O

utc

ome

exp

erie

nce

dby

pri

orad

opte

rs�

geog

rap

hic

dis

tan

ceto

VC

s it

�1b

33.8

940

.56

.37

.02

.62

.15

.52

.26

.82

.38

10.P

roxi

mat

ep

rior

adop

ters

it�

1�

geog

rap

hic

dis

tan

ceto

VC

s it

�1b

8.87

19.6

2.0

5.2

8.1

9.0

4.2

0.4

1.2

0.0

6.3

0

11.

Age

itb

2.46

1.10

�.3

70

.09

�.0

2.0

7.2

7�

.07

�.3

3�

.04

.15

12.

Inn

ovat

ive

pro

pen

sity

it2.

0814

.27

.25

�.0

1�

.10

�.0

2�

.14

�.1

0.0

2.2

0.0

3�

.02

�.3

613

.S

lack

itb

0.80

0.78

�.0

9.0

1.0

2.0

0.0

2.0

6�

.04

�.0

8�

.01

.00

.00

�.0

114

.S

ize i

tb6.

152.

33�

.26

.06

.12

�.0

8.0

9.2

4.0

1�

.26

.08

.14

.31

�.2

4.1

115

.V

C-b

acke

dat

fou

nd

ing

0.52

0.50

.28

.04

�.0

2.0

4�

.08

�.0

7.1

0.2

5.1

1.0

1�

.25

.64

�.1

0�

.34

16.L

ocal

inve

stm

ent

opp

ortu

nit

ies i

t�

1b

1.05

4.17

�.4

7�

.01

.04

�.0

9.0

1.3

2�

.15

�.4

4�

.15

.24

�.1

1.0

3.2

0�

.18

.20

�.4

7

an

�1,

726.

bL

ogar

ith

m.

experience were less likely to move toward exter-nalization of their innovation functions. Whetheran IT firm was itself backed by VCs at time offounding does not seem to exercise a significantinfluence on the adoption decision.9 Finally, avail-ability of local investment opportunities in the fo-cal firm’s geographic vicinity increased the likeli-hood of CVC program adoption. The overall modeltest at the bottom of Table 2 is a Wald chi-squaretest of the joint significance of all the explanatoryvariables; this test shows that model 1, as a whole,is highly significant.

Cross-Population Contagion

We added the cross-population contagion vari-ables to model 1 to get models 2 and 3. Specifically,we examined the adoption-enhancing effects uponprospective IT adopters of geographic proximity toa VC population cluster, and of observing benefitsgenerated by VC practices. Hypothesis 1 predictedthat firms exposed directly to a VC cluster due togeographic proximity would be more likely to

adopt CVC programs. In model 2 we obtained asignificant and negative coefficient on geographicdistance from VC clusters, showing that firms head-quartered farther from a VC cluster are less likely toadopt CVC programs, while nearby firms are morelikely to adopt such programs. To test Hypothesis2, we added our measure of observed benefits of VCpractices to model 3. We obtained a positive andsignificant coefficient, indicating that higher inci-dence of VC-backed IPOs escalates IT firms’ adop-tion of CVC programs. Consequently, model 2 is asignificant improvement on model 1 (Wald �2 �163.98), and model 3 is a significant improvementon model 2 (Wald �2 � 20.11).

Within-Population Contagion

Next we investigated the effects of within-popu-lation contagion influences. Specifically, we exam-ined how observing CVC adoptions and outcomesfor other firms in the IT sector affected the likeli-hood of CVC program adoption by a focal IT firm.

We captured the effects on the likelihood of CVCprogram adoption of popularity of CVC programs,prominence of prior adopters, outcomes experi-enced by prior adopters, and distance to prioradopters in models 4–8 by adding variables to theirbaseline specification. Hypothesis 3a states that anIT firm is more likely to adopt when CVC programsgain popularity within the industry. Model 4 showsa positive and significant coefficient on the cumu-lative number of prior adopters in the IT industry,

9 In an exploratory analysis, we found that approxi-mately 45 percent of the IT firms in our sample had beenventure backed. Thus, restriction of range on this vari-able may be one reason for its insignificance. To checkwhether adoption by IT firms headquartered far from aVC cluster was affected by VC backing at the time theywere founded, we reran the analysis for a subsample ofdistant IT firms, and found no effect.

TABLE 2Cross-Population Contagiona

Variable Model 1 Model 2 Model 3

Distance to VC clustersit � 1 �0.02** (0.00) �0.02** (0.00)Number of VC-backed IPOst 0.57** (0.13)

ControlsAgeit �0.11** (0.02) �0.10** (0.02) �0.08** (0.02)Innovation propensityit �0.02** (0.00) �0.02** (0.00) �0.02** (0.00)Slackit 0.25** (0.01) 0.25** (0.01) 0.21** (0.00)Sizeit 0.20** (0.02) 0.21** (0.02) 0.21** (0.03)VC-backed at foundingi 0.04 (0.08) 0.02 (0.08) 0.04 (0.06)Local investment opportunitiesit � 1 0.03** (0.01) 0.02** (0.01) 0.02** (0.01)Constant �2.85** (0.00) �2.78** (0.00) �5.44** (0.66)Observations 1,726 1,726 1,726Number of firms 264 264 264Number of adoptions 94 94 94

Overall model test 62.09** 64.96** 88.05**Joint significance test (cross population) 20.11**df 2

a Standard errors are in parentheses.** p � .01One-tailed test for hypothesized variables, two-tailed for controls.


TA

BL

E3

Wit

hin

-Pop

ula

tion

Con

tagi

ona

Var

iabl

esM

odel

4M

odel

5M

odel

6M

odel

7M

odel

8M

odel

9

Dis

tan

ceto

VC

clu

ster

s it

�1

�0.

02†

(0.0

1)N

um

ber

ofV

C-b

acke

dIP

Os t

0.39

**(0

.15)

Pop

ula

rity

ofin

nov

atio

nit

�1

0.02

**(0

.00)

0.01

**(0

.00)

0.01

**(0

.00)

Pro

min

ent

pri

orad

opte

rsit

�1

0.05

**(0

.00)

0.02

**(0

.00)

0.02

**(0

.00)

Ou

tcom

esex

per

ien

ced

byp

rior

adop

ters

it�

10.

04**

(0.0

0)0.

03**

(0.0

0)0.

02**

(0.0

1)P

roxi

mat

ep

rior

adop

ters

it�

10.

06**

(0.0

0)0.

04**

(0.0

0)0.

03**

(0.0

1)

Con

trol

sA

geit

�0.

09**

(0.0

2)�

0.10

**(0

.02)

�0.

09**

(0.0

2)�

0.08

(0.0

2)�

0.07

**(0

.02)

�0.

06**

(0.0

2)In

nov

atio

np

rop

ensi

tyit

�0.

02**

(0.0

0)�

0.02

**(0

.00)

�0.

02**

(0.0

0)�

0.02

**(0

.00)

�0.

02**

(0.0

0)�

0.02

**(0

.00)

Sla

ckit

0.23

**(0

.01)

0.25

**(0

.01)

0.21

**(0

.01)

0.20

(0.0

1)0.

18**

(0.0

1)0.

19**

(0.0

1)S

ize i

t0.

21**

(0.0

2)0.

20**

(0.0

2)0.

21**

(0.0

2)0.

20(0

.02)

0.21

**(0

.02)

0.21

**(0

.03)

VC

-bac

ked

atfo

un

din

g0.

02(0

.01)

0.03

(0.0

9)0.

04(0

.08)

�0.

08(0

.05)

�0.

06(0

.05)

�0.

02(0

.03)

Loc

alin

vest

men

top

por

tun

itie

s it

�1

0.03

**(0

.01)

0.03

**(0

.01)

0.03

**(0

.01)

0.01

(0.0

1)0.

01†

(0.0

1)0.

02**

(0.0

1)

Con

stan

t�

3.11

**(0

.02)

�3.

23**

(0.0

1)�

2.89

**(0

.03)

�2.

93**

(0.0

1)�

3.18

**(0

.03)

�4.

98**

(0.7

2)

Nu

mbe

rof

obse

rvat

ion

s1,

726

1,72

61,

726

1,72

61,

726

1,72

6N

um

ber

offi

rms

264

264

264

264

264

264

Nu

mbe

rof

adop

tion

s94

9494

9494

94O

vera

llm

odel

test

84.0

**61

.92*

*83

.62*

*90

.65*

*98

.26*

*10

5.85

**Jo

int

sign

ific

ance

test

(wit

hin

pop

ula

tion

)d

f�

445

.74*

*26

.76*

*Jo

int

sign

ific

ance

test

(cro

ssp

opu

lati

on)

df

�2

15.5

4**

aS

tan

dar

der

rors

are

inp

aren

thes

es.

†p

�.1

0**

p�

.01

On

e-ta

iled

test

for

hyp

oth

esiz

edva

riab

les,

two-

tail

edfo

rco

ntr

ols.

supporting Hypothesis 3a. Hypothesis 3b arguesthat potential adopters tend to pay more attentionto prominent prior adopters. Model 5 supports Hy-pothesis 3b by showing that prior adopters’ prom-inence has a positive and significant influence onthe adoption of CVC programs. Model 6 confirmsHypothesis 3c’s prediction that success is conta-gious—when IT corporations acquire start-upsfunded by their own CVC programs, other IT firmsare encouraged to adopt. Model 7 shows that adop-tion by peers in close geographic proximity to afocal firm increases its probability of adoption, of-fering support for Hypothesis 3d. Finally, model 8includes all the within-population contagion vari-ables and confirms the results of the earlier models.We found that prospective adopters are influencedby the popularity of CVC programs in their indus-try, by prior adopters’ prominence, by the out-comes experienced by prior adopters, and by prox-imate prior adopters. A Wald chi-square (45.74)comparing models 1 and 8 indicates a significantimprovement in model fit. Our results are consis-tent with those reported in prior studies examining

the within-population contagion influences on theadoption of new organizational practices (e.g.,Greve, 1995; Haunschild & Miner, 1997).

Finally, in model 9 we included both cross- andwithin-population influences on the probability ofinnovation adoption. A Wald chi-square (15.54)comparing models 8 and 9 indicates a substantialimprovement in model fit. To compare the effectsof cross- and within-population contagion, we cal-culated the magnitudes of the influences of thevariables in model 9. We evaluated the marginaleffects at the means of the independent variables.The two variables with the most powerful effect onCVC program adoption were geographic distancefrom VC clusters and the observed benefits of VCpractices—the cross-population contagion vari-ables. A 1 percent increase in distance cut the prob-ability of CVC program adoption by 0.04 percent,and a 1 percent increase in observed benefits of VCpractices boosted the probability of adoption by 0.9percent. In contrast, our within-population conta-gion variables had a substantially smaller influ-ence. A 1 percent increase in the within-industry

TABLE 4Cross Versus Within-Population Contagiona

Variable Model 10 Model 11 Model 12 Model 13

Distance to VC clustersit � 1 �0.04** (0.00) �0.12** (0.03) �0.04** (0.01) �0.04** (0.00)Number of VC-backed IPOst 0.53** (0.12) 0.56** (0.12) 0.48** (0.12) 0.48** (0.14)Popularity of innovation � distance to VC

clustersit � 1

0.00** (0.00)

Popularity of innovationit � 1 0.02** (0.00)Prominent prior adoptersit � 1 � distance to VC

clustersit � 1

0.01* (0.01)

Prominent prior adoptersit � 1 0.00 (0.02)Outcomes experienced by prior adopterit � 1 �

geographic distance to VCsit � 10.00* (0.00)

Outcome experienced by prior adopterit � 1 0.03** (0.00)Proximate prior adoptersit � 1 � distance to VC

clustersit � 10.01** (0.00)

Proximate prior adoptersit � 1 0.00** (0.00)

ControlsAgeit �0.07** (0.02) �0.07** (0.02) �0.07** (0.02) �0.08** (0.02)Innovation propensityit �0.02** (0.00) �0.02** (0.00) �0.02** (0.00) �0.02** (0.00)Stackit 0.20** (0.00) 0.23** (0.01) 0.17** (0.01) 0.17** (0.01)Sizeit 0.22** (0.04) 0.21** (0.03) 0.21** (0.03) 0.19** (0.03)Venture backed at foundingi 0.02 (0.06) 0.05 (0.07) 0.03 (0.07) �0.01 (0.03)Local investment opportunitiesit � 1 0.02** (0.01) �0.04* (0.02) 0.03** (0.01) 0.01 (0.01)Constant �5.49** (0.70) �5.33** (0.74) �5.03** (0.68) �4.89** (0.75)

Observations 1,726 1,726 1,726 1,726Number of firms 264 264 264 264Number of adoptions 94 94 94 94Overall model test 96.68** 89.58** 100.88** 96.63**

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

** p � .01One-tailed test for hypothesized variables, two-tailed for controls.


popularity, prominence, outcome, and proximityvariables raised the probability of CVC programadoption by 0.03 percent, 0.04 percent, 0.04 per-cent, and 0.06 percent, respectively. Clearly, ourestimation models indicate that cross-populationinfluences are significant from both a statistical andsubstantive point of view, and that ignoring themwould offer at best a partial view of diffusion dy-namics. In sum, our results suggest that in additionto the more familiar within-population contagionmechanisms, cross-population contagion triggersother powerful mechanisms galvanizing IT firms toadopt CVC programs.

When Cross- and Within-PopulationContagion Interact

Next, we examined the role of geographic dis-tance to the VC population in moderating the po-tency of within-population contagion influences.Hypothesis 4 predicts that firms geographicallyproximate to a VC population cluster are less sub-ject to influences arising from the actions and out-comes of prior adopters within their own industry.Accordingly, we tested Hypothesis 4 by interactingwithin-population contagion measures with geo-graphic distance to VC clusters.

In model 10, we obtained a significant negativecoefficient for distance to the VC clusters and asignificant, positive coefficient for the interactionterm. This finding suggests that IT firms that aregeographically distant from a VC cluster are heavilyinfluenced by the popularity of CVC programs intheir industry, whereas proximate IT firms are in-clined to ignore their peers. In other words, theimpact of the within-population popularity of CVCprograms upon the probability of adoption is con-tingent on proximity to a VC cluster, with proxi-mate firms discounting popularity-based contagioninfluences in making adoption decisions.

In model 11, we interacted prior adoption byprominent IT firms with geographic distance to theVC clusters. The coefficient on distance is negativeand significant, and the interaction term is positiveand significant, suggesting that geographically dis-tant firms are more susceptible to the actions ofprominent prior adopters within their industry.This finding supports our prediction that IT firmsproximate to a VC cluster discount the actions ofprominent firms within their industry.

In model 12, we interacted outcomes experi-enced by prior corporate adopters with distance toVC clusters. As in models 10 and 11, we found thatthe coefficient for the interaction term is positiveand significant. This suggests that firms that aredistant from the main VC clusters pay close atten-

tion to the outcomes experienced by prior adopterswithin their industry, and firms located near VCclusters are less affected. So here too, geographicproximity moderates the effect of within-popula-tion contagion upon the likelihood of CVC programadoption.

In model 13, we interacted proximate prioradopters with distance from VC clusters. The re-sults show that, as hypothesized, IT firms geo-graphically distant from a VC cluster were lesslikely to adopt CVC programs, and that they weremore susceptible to contagion from an increase inthe number of IT adopters in their close vicinity. Insum, our results show that at least within the con-text we studied, firms were differentially suscepti-ble to the actions and experiences of their industrypeers. Firms located near the originating VC popu-lation were less susceptible to within-populationcontagion influences.

DISCUSSION AND CONCLUSION

In this article, we have proposed and tested amidrange theory explaining the incursion of ven-ture capital practices into the information technol-ogy sector of the U.S. economy, and their subse-quent diffusion in the form of corporate venturecapital programs. In the 1950s, enclaves of privateventure capital investors devised a set of innova-tive approaches to identifying promising entrepre-neurial start-ups, accelerating them through theirearly developmental stages, and helping themachieve liquidity. In the 1960s, alarmed by VC-backed start-ups’ forays into their product-markets,public corporations tried to emulate the VC model,expecting strategic renewal and financial returns tofollow (Gompers, 2002). But 1960s-style corporateventuring proved ineffectual, so most programswere terminated and the venture capital model beata retreat to Sand Hill Road and Route 128. In the1990s, corporate interest in the VC model was re-ignited by stellar performances of VC-backed firmsin the technology sector and by corporations’ grow-ing disenchantment with traditional R&D pro-grams. VC practices once again infiltrated publiclytraded corporations, this time undergoing more ex-tensive modification. Corporate venture capitalprograms diffused rapidly in the 1990s and werefound to yield better results, on average, than pri-vate venture capital investments (Gompers, 2002).

We conceptualized the slow and faltering spreadof VC practices as a specific instance of the diffu-sion of an endemic innovation—one that, at theoutset, is exclusively native to its organizationalpopulation of origin. We argued that endemic in-novations spread into adjacent organizational popu-


lations through two different forms of contagion. Thefirst, cross-population contagion, flows through closeinterpersonal relations that afford direct exposure tothe innovation, observation of its tangible payoffs,and inductive learning of its innovative practices.We reasoned that cross-population contagion re-quires protracted interpersonal contact, becausethe initial transfer of an endemic innovation entailsthe appropriation of tacit knowledge and its recom-bination and application in an alien organizationalsetting.

We tested hypotheses based on this argument bycollecting longitudinal data from 264 informationtechnology corporations from 1992 through 2001.Event history analyses provided support for ourpredictions. Results suggest that IT firms’ managerslearned and appropriated venture capital practicesthrough direct contact with VC partners and inter-preted the rate of venture-backed IPOs as trackingthe value created by VC practices. Our cross-popu-lation findings suggest that personal relationshipsand physical proximity facilitate the transmissionof an endemic innovation into a new population.

We further theorized that a second set of diffu-sion mechanisms starts to operate once an endemicinnovation takes root in the new setting. Contagionwithin the new population begins, supplementingwithout necessarily supplanting the initial cross-population contagion process. We reasoned thatdiffusion among a single industry’s more culturallyhomogeneous firms would rely less on direct rela-tionships fostered by close physical proximity toVCs, and would rely more on mimicry and socialcomparison processes. Our results support theseconjectures. They suggest that the popularity ofCVC programs within the IT industry, the proxim-ity and prominence of IT firms adopting prior pro-grams, and signals that the programs were payingoff for their corporate parents by generating attrac-tive acquisition targets fueled the diffusion of cor-porate venturing programs in the 1990s. Thesefindings are consistent with those of previous re-search, and they suggest that prior diffusion theo-ries and results are especially germane to thespread of innovations within sets of culturally sim-ilar organizations.

Finally, we considered how cross-population dif-fusion and within-population diffusion mecha-nisms interact. We reasoned that cross-populationcontagion remains prepotent whenever a prospec-tive adopter’s physical proximity to members of aninnovation’s originating population allows directcontact with the innovation’s practices and practi-tioners. In the absence of physical proximity, wepredicted that within-population contagion wouldhave a greater impact on adoption behavior. Our

findings support these expectations: the more dis-tant a prospective IT adopter’s headquarters from aVC cluster, the more closely the firm’s adoptiondecisions were aligned with its IT industry peers.These results suggest that where propinquity al-lows cross-population contagion mechanisms tooperate, they are more potent than within-popula-tion mechanisms. Among IT corporations, in-creases in the observed rates of VC-backed IPOshad the strongest impact upon IT firms’ adoption ofcorporate venturing, followed by proximity to a VCcluster. These two diffusion mechanisms appear tohave overwhelmed the effects of the within-popu-lation mechanisms that have been the focus ofmuch previous work.

Implications for Theory and Research

Our investigation of corporate venture programshas implications for the study of organizationalfields and for theories of innovation diffusion.First, the study’s results underscore the value ofmechanism-based theorizing, taking the organiza-tional field as the fundamental unit of analysis(Davis & Marquis, 2005). In particular, they showthe value of moving beyond single-population re-search samples to consider the social mechanismsconnecting the different organizational populationsthat inhabit a field (Meyer et al., 2005). Second, ouranalyses illustrate how innovations penetrate organ-izational population boundaries—an issue seldomexplored in the innovation diffusion literature. Byexplicitly delineating cross- and within-populationcontagion mechanisms, we offer a more completeand nuanced analysis. The study highlights theendemic character of certain innovations and therole that weak theorization plays in sequestering anendemic innovation within the population where i

crossing the organizational species barrier: how...

Documents