Learning While Innovating: The Abandonment of Corporate Venture Capital Programs

Vibha Gaba INSEAD 1 Ayer Rajah Avenue Singapore 138676 Tel: +65 6799 5268 Fax: +65 6799 5399 Email: [email protected]

Working Paper March, 2008

The author thanks Ziv Carmon, Pushan Dutt, Kathy Eisenhardt, Martin Gargiulo, Henrich Greve, Alan Meyer, and Ilian Mihov for helpful comments. The research reported in this paper was supported by INSEAD research grant 2520-025R. 1

Learning While Innovating: The Abandonment of Corporate Venture Capital Programs

ABSTRACT

This paper draws upon prior research on organizational learning and innovation diffusion to empirically investigate the abandonment of an administrative innovation- corporate venture capital (CVC) programs - by information technology (IT) firms. CVC programs are modeled upon the venture capital (VC) practices of finding, funding, and guiding entrepreneurial startups. We examine the effect of firms’ direct experiences with the innovation on its ability to retain or abandon the innovation. We also investigate how this experience interacts with the contagion influences that bias firms towards abandonment. Results suggest that direct experience with the CVC program decreases the likelihood of its abandonment by a focal firm. We also find that firms with a richer understanding of their CVC program are less susceptible to contagion influences.

1

Learning While Innovating: The Abandonment of Corporate Venture Capital Programs

INTRODUCTION Why do firms abandon innovations they once adopted? A review of innovation diffusion literature shows an extensive collection of research papers on innovation adoption but only a handful of studies on innovation abandonment. Despite the fact that many organizational innovations become widely popular only to decline or even disappear, we know much less about what causes firms to abandon innovations they once adopted (Abrahamson and Fairchild, 1999; Strang and Macy, 2001.) The innovation diffusion literature suggests contagion, social learning, or mimicry as a general process explaining the diffusion of innovations (Strang and Soule, 1998). Because firms have no direct experience with the innovation prior to adoption, they use social comparison to evaluate innovations, i.e., the evaluation of an innovation is influenced by the observation that others are adopting it. This explanation has been extensively tested and validated in the context of innovation adoption. However, to extrapolate the findings in the context of innovation abandonment may be misleading because it assumes that innovation adoption and abandonment are fundamentally similar processes. Unlike the adoption decision, the abandonment decision is also likely to be influenced by firms’ direct experience with the innovation (Burns and Wholey, 1993). This suggests that in the context of innovation abandonment firms draw inferences not only from the behavior of other firms but also from their own direct experience. Prior research on organizational learning argues that firms acquire information and knowledge about administrative practices, innovations, or technologies in a variety of ways

2

(Levitt and March, 1988). Learning from own experience is one of the fundamental sources of learning. As firms spent time implementing an innovation they accumulate knowledge, enhance capabilities, and reduce uncertainty associated with the innovation (Levitt and March, 1988; Argote, Beckman, and Epple, 1990). However, firms also learn vicariously by observing and imitating the actions and behaviors of other firms. In the case of insufficient or ambiguous information from their own experience, firms often supplement their own experience with those of others (Baum, Li and Usher, 2000). While prior research acknowledges that firms generally use a combination of different learning modes in making strategic decisions, there is little work done to evaluate the relative importance of these different modes (Lieberman and Asaba, 2006). Innovation abandonment seems an appropriate context to explore not only the direct but also the relative influences of alternative modes of learning. Like innovation adoption, innovation abandonment maybe a consequence of contagion influences where firms feel the pressure to abandon the innovation when they see others doing so. However, does this cause firms to ignore their direct experience with the innovation and abandon it even if they find the innovation useful and appropriate? Furthermore, are all firms equally susceptible to these contagion influences? Or is there heterogeneity in firms’ susceptibility to contagion influences? In this paper we empirically investigate the abandonment of an administrative innovation - corporate venture capital (CVC) programs - by information technology (IT) firms from 19922003. These programs are modeled upon the practices of private venture capital (VC) firms for finding, funding and guiding external startups. The time period of our study coincides with the explosive popularity and subsequent decline of CVC programs. We draw upon prior research on organizational learning and innovation diffusion to explain the heterogeneity in firms’ ability to retain or abandon an innovation they once adopted. We examine the effect of firms’ direct

3

experience with the innovation on the likelihood of innovation abandonment. We further investigate how this experience interacts with the contagion influences that bias firms towards abandonment. We hope to make three contributions with this research. First, we elaborate and examine the important role of innovation related experience in the abandonment of innovations. We argue that firms enhance their innovation related experience – acquisition of the skills, values, procedures, behaviors, and know-how - in multiple ways. This is directly consequential for their ability to retain the innovation. Second, we explore the moderating role of firms’ direct experience with the innovation. Firms act based on the inferences that they drawn from their own and others experiences. We contend that both these learning processes factor into the innovation abandonment decision of firms. We show that the firm-level heterogeneity with the innovation related experience manifests itself as heterogeneity in firms’ susceptibility to contagion influences. Firms with rich experience are relatively immune to the contagion influences emanating from the abandonment decisions of their peers. Third, we empirically investigate the determinants of the abandonment of an administrative innovation - an important but relatively under studied issue in the innovation literature. Understanding innovation abandonment can yield theoretical insights into the temporal instability of organizational innovation, and why and how reasonable innovations are discarded or discredited. It is important because in reality only a small minority of innovations actually become institutionalized while most end up looking more like fads or fashions (Zucker, 1988; Strang and Macy, 2001). Our paper is organized as follows. We begin by describing our research setting, the innovation we study, and the abandonment of that innovation by information technology firms. We then draw upon prior research on organizational learning and innovation diffusion to develop

4

our hypotheses. Next, we describe our measures and analytical methods, and go on to report findings. The paper’s final section discusses the implications of our results.

RESEARCH SETTING We study the abandonment of corporate venture capital (CVC) programs. CVC programs are modeled upon the practice of private venture capitalist (VC) firms. The venture capital model is an organizational form that has become increasingly popular as a vehicle for finding, financing, and guiding the development of entrepreneurial startups (Kenny and Florida, 2000). Over the last three decades, annual venture capital investments have swelled from less than $1 billion to over $103 billion, far outstripping growth in any other investment class (Gompers and Lerner, 1999). In the decade of 1990s, private venture capital firms, located in distinct geographic clusters (see Figure 1 for a geographic breakdown of VC investments) began outperforming public technology firms in identifying promising business opportunities, accelerating the progress of new ventures through their early development, and helping these ventures achieve liquidity (Gompers and Lerner, 1999). Incumbent firms alarmed at the erosion of their markets by new VC-backed entrants, concluded that the VC model might be a powerful tool for harnessing technological change. Subsequently, to complement, or even to substitute for in-house R&D units, corporations began setting up their own CVC programs to make external equity investments in startups. The corporate share of overall venture capital investing rose rapidly from 2 percent in 1994 to 15 percent in 2000 and nearly $16 billion was invested by over 300 corporations (Venture Economics, 2001). Then, economic recession and a collapse of equity and IPO markets in 2000 ended the boom in the venture capital industry. During the first quarter of 2001, corporate venture investing fell 81 percent and many corporations shut down or idled their CVC programs (Venture Economics, 2003). Figure 1 shows the dollar amount of

5

investments as well as the number of firms making CVC investments each year during the time period of our study. -----------------------------------------------------------Insert Figure 1 about here ------------------------------------------------------------

CVC programs are structured and managed differently than traditional in-house corporate R&D or business development functions and they typically report into the firm’s top executive team at the senior vice president level, usually through the corporate strategy or business development function. Corporate venture capital investments are typically riskier and less subject to rigid management of internal costs than conventional in-house R&D (Hamel, 1999). Ideally, these units are designed to facilitate a coordinated and proactive approach to new business development by focusing on external sources of new technologies. Firms establish CVC programs for a variety of strategic and financial reasons. While some firms are undoubtedly enticed by the financial returns of venturing, most adopters report that their foremost objectives are strategic ones – gaining exposure to new and disruptive technologies, access to new markets and business models, and identification of prospective acquisition targets. Firms can easily assess the financial return on their investment in start-ups (through taking the entrepreneurial venture public) by looking to IPO markets. However, strategic returns from CVC programs are not so easily assessed – they are long term, potentially risky, and not easily quantifiable. Firms, however, find it difficult to replicate VC practices and outcomes – many of these practices are encoded in tacit knowledge, and some clash with corporate systems and cultures (Chesbrough, 2000). The novel practices that make up the VC business model fall into two categories: pre-investment activities and post-investment activities (Sorenson and Stuart, 2001). VCs build social capital in the form of networks of contacts that are the source of their “deal

6

flow”. A large VC firm screens over 1,000 investment proposals per year, and typically invests in between one and two percent of them. The selection of high-caliber opportunities occurs through a process termed “due diligence” in which VCs scrutinize a target’s business plan, technology, intellectual property, management team, and business alliances. After investing their financial capital in a startup, experienced VCs shift to advising, monitoring, and building the company. They supply contacts to law firms, headhunters, real estate brokers, and prospective customers. VCs take seats on startups’ boards, providing infusions of financial advice and management expertise as needed. Prominent VC firms tap their stocks of reputational and social capital, lending their prestige and connections to startups in forging new alliances and attracting new resources. Finally, returns are realized through either the acquisition of the VC-backed startup by an established corporate buyer, or through the issuance of shares in an initial public stock offering (IPO). The IPO is the preferred outcome for all participants – it generally results in the highest valuation of the company, provides liquidity to the investors, and preserves the young company’s independence (Gompers and Lerner, 1999). The repertoire of skills possessed by the VCs that are required in the selection, advisement and IPO phases are not easily available to or accessible by all firms. The actual know-how or modus operandi call on venture capitalists’ intuition, judgment, and skills acquired and honed on the job over the years. In fact, a savvy venture capitalist is often characterized as “smart money” – money that is imbued with the entrepreneurial savvy, business contacts, executive talent and patience of financiers with long experience in helping entrepreneurial companies succeed (Doerflinger & Rivkin, 1987:16). In addition, internal R&D personnel often resist CVC programs as they prefer funding be allocated to internal R&D programs (Gompers and Lerner, 1999).

7

We define abandonment of CVC programs as cessation of all the new investment related activity in entrepreneurial startups. It is generally accompanied by the dissolution of a distinct structural entity that was initially created for the very purpose of the investing in external entrepreneurial startups. THEORY AND HYPOTHESES Why do organizations’ abandon innovations they once adopted? The literature on fads and fashions highlights the transient nature of many organizational innovations. It focuses on the role of managerial discourse or collective beliefs in explaining the origins and the trajectories of a variety of organizational innovations. For example, Abrahamson & Fairchild (1999) examined the lifecycles of four innovations - job enrichment, quality circle, total quality management, and business process reengineering. They found that “the upswing in the amount of discourse about the quality circles paralleled the diffusion of decision to adopt them, whereas the downswing in the amount of discourse coincided with the diffusion to reject them” (p.731). These patterns of collective beliefs at least in part give rise to contagion pressures as more and more firms observe their peers adopting an innovation (Abrahamson, 1991). This suggests a reverse diffusion process where the innovation is evaluated less positively the more other organizations abandon it, leading to contagion of innovation abandonment. Greve (1995) studied the abandonment of a product-market strategy (radio format) by a sample of radio stations from 1984-93. He found strong evidence for contagion influences as a source for strategy abandonment, i.e., radio stations discontinued the use of easy listening format when they observed others abandoning it. Burns & Wholey (1993) examined the adoption and abandonment of an administrative innovation, matrix structures, in a sample of hospitals. They too found that regional proportion of hospitals

8

abandoning matrix structures positively influenced the likelihood of matrix structure for a focal hospital. However, innovation abandonment can also be viewed as a function of firm’s innovation related experience. Unlike in the adoption process, where firms have no direct experience with the innovation and therefore seek external cues to guide their decision, in the post adoption phase, firms are also likely to base their decision on their own experience with the innovation. Lanier, Carson & Guidry (2000) compared the lifespan of sixteen management innovations and found that more difficult and complex innovations tend to have shorter life spans. In a more recent study, Rao, Greve, & Davis (2001) examined the initiation and abandonment of analysts’ coverage of NASDAQ firms. They found that analysts’ relied more on their own experience in making abandonment decisions, i.e., those who significantly overestimated the earnings per share for a particular firm were more likely to abandon coverage. However, it is relatively easier for security analysts to observe and evaluate errors in their earnings forecasts than for managers to judge the true value of a complex administrative innovation. The costs and benefits of complex administrative innovations may take time to become evident, may be hard to observe, or may have ambiguous interpretations (Repenning & Sterman, 2002). Burns & Wholey (1993) found that, in addition to contagion effects, less diverse hospitals were more likely to abandon the matrix structures. They speculate that firms’ the direct experience with the matrix structures seem to influence their decision to abandon these structures. However, this conclusion is inferred from tenuous evidence, such as observing changes in the coefficient signs of organization-level predictor variables. Although diffusion scholars have studied the origins and the trajectories of various organizational innovations, the focus has been mainly on innovations that have been adopted or

9

rejected widely – a form of selection bias (Strang & Soule, 1998). Importantly, in explaining the upswings and downswings of organizational innovations, researchers do not explicit distinguish between the adoption and abandonment of innovations. No studies offer an in-depth examination of role of direct experience in the abandonment phase and how it interacts and operates in tandem with the contagion influences that impel firms to abandon innovations. In this section, we develop such a model. Our objective is to understand how does innovation related experience contribute to the eventual retention or abandonment of administrative innovations. First, we elaborate and examine the direct effect of firms’ innovation related experience on its ability to retain or abandon an innovation. Second, we propose that firm’s experience with the innovation also moderates its susceptibility towards contagion influences. Firms that develop a deep understanding of the innovation and experience positive results with the innovation are less likely to imitate the abandonment decision of their peers.

Innovation Related Experience Much of the innovation diffusion research focuses on the initial adoption of innovations with the assumption that firms learn about the innovation prior to its adoption, and subsequently, become more effective at implementing the innovation. However, the inherent uncertainty in implementing the innovation precludes exhaustive knowledge of the precise characteristics and consequences of an innovation prior to adoption. Rarely, if ever, are novel innovations universally applicable across time, circumstances, and organizational contexts. Learning about an innovation entails acquisition of a causal model that links novel practices to valued outcomes for the adopters. It also entails developing routines on how to staff and transact within a given administrative innovation. However, acquisition of such knowledge rarely occurs without

10

difficulty. Gaps in information acquired and errors in interpretation imply that adopters actively seek to build on their existing base of knowledge and experience even in the post adoption phase (Sorenson, Rivkin, & Fleming, 2006). This implies that firms continue to engage in an information-seeking and information-processing exercise in the post adoption phase to reduce the inherent uncertainty surrounding an innovation (Repenning & Sterman, 2002; Schwab, 2007). Firms can gather experience with an innovation and learn about its objectives, practices, and outcomes in multiple ways. Prior research on experiential learning has mainly focused on performance improvements as the main indicator for organizational learning (Argote, 1999). However, when the performance feedback is slow or ambiguous firms may enhance the richness of their experience by attending to multiple events and actions such as the consequences associate with the making of decisions and its implementation (March, Sproull & Tamuz, 1991). This may be especially true in the context of administrative innovations where the adopters may often find it difficult to accurately assess an innovation’s underlying costs and benefits. Long delays between introducing new processes and reaping its benefits may further conflate their evaluate of the innovation. Therefore, we conceptualize innovation related experience broadly: it includes the depth of learning at the time of adoption, knowledge acquired while implementing the innovation, and the results experienced from implementing the innovation. Each of these aspects of the innovation related experience enhance and build adopters understanding of the innovation and enable them to translate the abstract properties of the innovation into more concrete set of practices. Next, we describe how each of these experiences specifically shape their decision to abandon or retain innovations.

11

----------------------------------------------------Insert Figure 2 about here. ------------------------------------------------------Information at the time of adoption. One of the consistent themes in the innovation diffusion literature is the distinction between early and late adopters. Scholars argue that early adopters search for innovations more actively and diligently, appraise likely costs and benefits, and adopt because their analyses project positive returns (Tolbert and Zucker, 1983; Fligstein, 1985; Barron, Dobbins, and Jennings, 1986). They often modify innovations to suit local conditions and that fit to their unique problems and opportunities (Rogers, 1995; Westphal, Gulati & Shortell, 1997). Late adopters, on the other hand, tend to economize on search costs by relying on cognitive heuristics of social proof, yield to bandwagon pressures, make ceremonial decisions, and adopt innovations as symbolic displays of their propriety and modernity (Tolbert & Zucker, 1983; Abrahamson and Rosenkopf, 1997; Rao, Davis & Greve, 2001). Consequently, late adopters are said to implement innovations in a more perfunctory and standard fashion, avoiding the cost and effort customization would entail (Westphal, Gulati, & Shortell, 1997). While prior empirical research has well documented the behavioral determinants of early versus late adoption, it has not directly examined the consequences of these differences on adopters’ propensity to abandon innovations. Although, Burns & Wholey (1993) distinguished between the early and late adopters, they did not directly test the consequences of time of adoption on the abandonment of matrix programs. Rao et al (2001) examined the initiation and abandonment of coverage of NASDAQ firms by security analysts. They find that analysts that initiated the coverage of a firm based on heuristics of social proof tend to overestimate firm’s future profitability and are subsequently more likely to abandon coverage of the firm. However, as mentioned earlier, the negative feedback on earnings estimates is much more salient in a

12

relatively short period of time as compared to any feedback on adopting complex administrative innovations. Even though Westphal et al. (1997) found that conformity to normative patterns of TQM implementation was negatively associated with perceived efficiency of these programs, they did not examine if this perceived inefficiency resulted in subsequent abandonment of TQM initiatives within the hospitals. Since early and late adopters differ in their motivation, level of customization, and expectations of benefits from the CVC programs at the time of adoption, they are also likely to qualitatively differ in their initial stock of know-how and understanding of VC practices. Early adopters are likely to spend more effort assessing and evaluating VC model and its applicability to their unique context. They are likely to be more familiar and cognizant of the challenges inherent in implementing these programs. In contrast, late adopters adopt the programs mainly in response to contagion influences, spend less time acquiring unique information about the VC practices, and possess little knowledge of the implementation risks associated with the VC model. This initial search for innovation related information will be consequential for the subsequent abandonment decision of the firms. Accordingly, we expect late adopters are unlikely to have a deep understanding of CVC program to sustain them in the long term. Therefore, Hypothesis 1: Late adopters of CVC programs are more likely to abandon their CVC programs.

Information acquired while implementing the innovation. Once an organization has adopted the innovation, what happens next? Does an organization’s ability to implement and retain the innovation depend only on its initial stock of knowledge? Or does it also depend on its ability to utilize, modify, and build on such knowledge? Organizational scholars argue that despite the documented benefits of innovations attempts to implement them sometimes fail

13

(Repenning & Sterman, 2002) and that new technologies often produce different outcomes when introduced in different contexts (Barley, 1986). Prior research on innovation implementation research largely uses qualitative case studies to highlight the recursive relationship among the innovation, organizational structures, managerial beliefs, and behaviors (Barley, 1986; Repenning & Sterman, 2002). However, the process through which firms acquire information and build their experience as they implement innovations remains largely unknown (Klein & Sorra, 1996) A number of studies in innovation diffusion research point to the geographical localization of knowledge flows. Researchers argue that geographic clusters facilitate the transfer of knowledge between firms operating within a region (Jaffee, Trajentberg & Henderson, 1993; Audretsch & Feldman, 1996). The access to information and knowledge occurs as firms collaborate and their members form professional and personal relationships. These overlapping relationships foster interpersonal trust and facilitate higher fidelity transfer of knowledge especially tacit knowledge on how to found, staff, and transact within a given administrative innovation (Saxenian, 1994; Sorenson & Audia, 2000). CVC programs are especially reliant upon tacit knowledge, in part because venture capitalists often resist explicit codification of their practices (Gaba & Meyer, 2008). Therefore, we contend that geographic proximity to the VC population clusters (original source of innovation) enables CVC adopters to enrich their innovation related experience through superior and ongoing access to the VC population and their practices. Close proximity to VC population clusters permits more frequent contact with the private VCs, facilitates exchange of formal and informal information, and assists coordination and sharing of investment activities – all of which are important in implementing CVC programs. It helps CVC adopters to not only minimize

14

implementation errors but also solicit advice in understanding and deploying innovations (Sorenson et. al., 2006). As one Corporate Venture Capitalists commented in our interviews: They key thing in my mind is the physical presence within the VC community. It ought to really help you in understanding the processes that have been developed within the venture capital community over the last 20-30 years, why those processes have been developed, how those processes play a role in mitigating risk. Every single decision that we have made in our program…..to access deal flow, to understand where the new investments are going, to learn how the early stage companies work, to deliver value proposition to your partner ….. has benefited from our presence in Silicon Valley. Geographic proximity to the VC population clusters helps in a number of ways. First, individuals in these firms are likely to have greater chance encounters with VCs that lead to lasting ties. Because geography strongly influences daily activities, being co-located increases the probability that a firm’s employees will meet and form a relationship with a member of the VC population (Sorenson and Audia, 2000). As local bonds form, this facilitates the transfer of knowledge, especially of tacit knowledge which is difficult to transfer except through face-toface interaction. Second, the local nature of labor markets and employee turnover implies a greater likelihood that individuals with a deeper understanding of VC practices enter into a variety of formal relationships with IT firms located close to the VC population clusters. Significant fluidity exists within the VC population clusters as personnel move within high-tech companies, venture capital, investment banking, and consulting among others (Bahrami and Evans, 2000). These individuals may be instrumental in supporting CVC programs (Florida and Kenney, 1998). Third, geographic proximity facilitates an understanding of the cultural context of the VC model (Saxenian, 1994). CVC adopters co-located near the VC population clusters are more likely to be cognizant of the informal norms, language, and basic assumptions guiding the venture capital decision making.

15

To sum up, geographic proximity to VC population cluster enables CVC adopters to build relationships with the VCs, to solicit advice when problems arise with understanding and deploying VC practices, and to become more comfortable with the inherently uncertain nature of venture capital investing. Consequently, such IT adopters will become increasingly skillful and consistent in their use of innovation. Therefore, Hypothesis 2: IT adopters located geographically proximate to the VC population cluster are less likely to abandon their CVC programs

Information acquired from evaluating the innovation. Diffusion scholars argue that innovation related experience also entails evaluating and assessing innovation’s performance. Firms may reject an innovation because by using it they may learn that it is inefficient and unable to deliver on the expected benefits (Abrahamson, 1991; Strang and Macy, 2001). New innovations seldom realize their full productive potential at the time of adoption and firms seek to learn from the coincidence of innovative strategies and the outcomes experienced (Strang and Macy, 2001). Therefore, one of the most persuasive forms of evaluating new innovations is by assessing how well they are meeting their goals for adopting the innovation in the first place (Burns and Wholey, 1993; Tolbert and Zucker, 1996; Rao et al, 2001). Specifically, strong performance decreases the intensity of search and experimentation for alternates, and motivates adopters to exploit and retain the current innovation that has proven successful (Greve, 2003). Weak performance, on the other hand, stimulates search for an alternative and motivates firms to abandon the current innovation. While a number of prior studies have shown that positive performance is associated with organizational persistence, while poor performance is related to changes in the course of action (e.g., Lant, Milliken and Batra, 1992; Audia, Locke and Smith, 2000; Rao et. al, 2000), this relationship has rarely been tested empirically in the context of

16

administrative innovations. As mentioned earlier, one reason is that clear-cut evidence for an innovation’s direct payoff is hard to obtain. Adopters and researchers alike often find it impossible to partial out and accurately assess an innovation’s underlying costs and benefits. For CVC adopters, one of the main strategic goals for adopting these programs is to identify and acquire new technologies that complement or build on their existing set of competencies. Through minority equity investments in entrepreneurial start ups, at a fraction of the risk and cost of an acquisition, organizations have the option to ally themselves with entrepreneurial startups developing proprietary technology or penetrating new markets. IT firms that are able to gain strategic benefits from their CVC programs, are likely to view CVC programs as successful and an effective means for externalizing R&D. Therefore, Hypothesis 3: IT adopters that experience positive outcome with their CVC program are less likely to abandon it.

The Moderating Role of Innovation Related Experience So far, we have argued that the abandonment decision is contingent upon firms’ direct experience with the innovation. Next, we examine how firms’ direct experience with the innovation interacts with the contagion influences emanating from the abandonment decision of other IT adopters. In the post adoption phase, overtime, with experience, firms gain confidence and reduce uncertainties related to implementing the innovation. However, this confidence may be challenged as they observe their peers abandoning the innovation. Both learning and institutional scholars have long argued that firms tend to imitate their peers and that such behavior is particularity ubiquitous under conditions of uncertainty and ambiguity (Cyert & March, 1963; DiMaggio & Powell, 1983; Haunschild & Miner, 1997). While information from direct experience is typically more detailed, salient, and better

17

understood than external information (Levinthal & March, 1993; Schwab, 2007), firms may vary in their ability to learn or they may fail to learn because of paucity or ambiguity of their experiences (Levitt & March, 1988; Levinthal & March, 1993; Argote, 1999). In that case firms may choose to supplement their own experience with those of others (Baum & Ingram, 1998). Indirect evidence suggests as firms reduce uncertainty related to their actions, they tent to rely more on their own experiences. For example, in the context of alliance formation, researchers have found that firms with their better in-house R&D experience are less likely to form alliances with others (Pisano, 1990; Stuart, 1998). Similarly, in the context of entry into international markets, firms with operating experience in a host country are less likely to benefit from the experience of other firms in that country (Shaver, Mitchell & Yeung, 1997). In the context of innovation diffusion, even if the adoption decision is influenced by the observing others, as firms gain direct experience with the innovation they may rely more on their own experience to make incremental adjustments to the innovation (Schwab, 2007). Learning from experience emphasizes careful assessment of the innovation for the specific adopter needs (Levinthal & March, 1993). It allows them to build a causal model specifying how and why the innovation is appropriate given the adopter needs, thereby providing a rationale for the eventual retention of the innovation. It also increases their propensity to engage in independent evaluation of the innovation rather than rely on the experience of others. Conversely, the inability to resolve this uncertainty makes adopters more attentive to the behaviors of others to sort through their own experience and to judge the true value of the innovation (Liberman & Asaba, 2006; Schwab, 2007). In that case, they are more likely to use social comparison as a basis for making decisions (Greve, 1995; DiMaggio and Powell, 1983; Abrahamson and Rosenkopf, 1997) and will be more susceptible to contagion influences.

18

Based on above arguments, we also expect that early adopters of the CVC program, firms geographically proximate to the VC population clusters, and firms who experience efficacious outcomes from their CVC programs will be less susceptible to the contagious influences. Therefore: Hypothesis 4: Early IT adopters, IT adopters geographically proximate to VC population clusters and IT adopters that experience positive outcomes will be less affected by the abandonment behavior of other IT firms.

METHOD Estimation Strategy We use a discrete-time event history methodology to model the abandonment of CVC programs (Allison, 1982, 1995). Our dependent variable Pi ( t ) is the conditional probability that firm i abandons the CVC program at time t, given that it is at risk of abandoning. Pi ( t ) is related to the covariates by the following equation: Pi (t ) = Φ[α + β 1 xi1 (t ) + ... + β k xik (t )] + υ i (t ) (1) where Φ is the cumulative density function and xi’s are covariates that affect the abandonment decision. We assume that the cumulative density function for the error term Φ (.) is normally distributed, so we use a probit model to estimate the probability of an abandonment event in a given year, in a pooled sample consisting of each firm observed during each of the twelve years (1992-2003). However, estimating the above model for only firms that are at risk of abandoning CVC programs may limit the generalizability of our results and bias our estimates. Since the abandonment decision is limited to firms that have adopted CVC programs, our sample would

19

include only adopters of CVC programs. Such a biased sample may not be representative of the entire population. Second, if certain unobserved error components affect both the probability of being in the risk set (the adoption of CVC program decision) and the probability of abandonment then our estimates of the β ’s would be biased. Equation (1) will lead to unbiased estimates only if the correlation between the unobserved errors that affect both the adoption and abandonment decision is zero. Therefore, we use the Heckman selection model, a two-staged procedure that corrects for the sample selection bias and yields unbiased estimates (Heckman, 1979). The Heckman selection model in our context would include two equations: the first equation predicts whether or not a firm is at risk of adopting the CVC program (it estimates the probability of adoption) and the second equation predicts the probability of CVC program abandonment conditional on having adopted it. If the error terms from these equations are significantly correlated, standard techniques applied to the second equation alone can yield biased results (Van de Ven and Van Praag, 1981). Heckman's procedure generates consistent, asymptotically efficient estimates for such models, allowing us to generalize to the larger population of firms (Heckman, 1979). In the original Heckman formulation, the model first estimates the likelihood of CVC adoption with a probit regression. The index function from the probit model is transformed into a hazard rate using the Mills ratio, and the estimated rate is then included as an independent variable in a second-stage bivariate probit model to model the abandonment decisions of firms. However, this two-step method estimator yields heteroscedastic errors and is not as efficient as the maximum-likelihood estimator. Therefore, we use maximum likelihood to simultaneously estimate the following two models, one for adoption and one for abandonment, where we assume that the errors ε i ,υ i are distributed as bivariate normal with correlation ρ.

20

P Adopt i (t ) = Φ[α 0 + α 1 z i1 (t ) + ... + α k z ik (t )] + ε i (t ) Pi

Abandon

(t ) = Φ[β 0 + β1 xi1 (t ) + ... + β k xik (t )] + υ i (t )

For model identification, we rely on exclusion restrictions rather than the nonlinearity of the error terms. This implies that we include variables in the adoption equation that are not in the abandonment equation. We also present a test of ρ = 0 to check for the appropriateness of the Heckman procedure. Sample and Data

To test our hypotheses, we gathered longitudinal data for a sample of U.S. firms in the information technology (IT) sector. We focus exclusively on IT firms to control for unobserved heterogeneity at the industry level, and since many of the venture capital investments during the time period of the study were focused on the IT sector, IT firms were more likely adopters of CVC programs. We followed the National Science Foundation’s definition of the IT sector (NSF, 2000) to include firms in these five industry subsectors: (1) Office, Computing and Accounting Equipment (SIC code 357), (2) Communications Equipment (SIC code 366) (3) Electronic Components (SIC code 367), (4) Communication Services (SIC codes 481-484, 489), (5) Computing and Data Processing Services (SIC code 737). The sample was constructed in two steps. In the first step, we compiled a list of IT firms drawn from the Forbes 500 list, which ranks U.S. firms by sales, profits, assets, and market value. Firms that rank among the top 500 on one or more of these criteria are included in the Forbes list, which is compiled annually1. 270 IT firms ranked among the top 500 on one or more

1

Forbes started ranking firms in 1997, so we were unable to use Forbes rankings prior to 1997 in building our sample. However, to check for bias we selected IT firms listed on the Fortune 500 at the beginning of the time period of our study (1992). Fortune 500’s ranking is based only on revenues, so this sample was much smaller, including only 43 IT firms, 40 of which are present in the larger Forbes 500 sample that we use.

21

of these criteria and these firms form the risk set for the adoption decision. In the second step, we used the Corporate Venturing Yearbook and Directory (2000, 2001, 2002) which reports the year in which firms adopted CVC programs. 92 firms were classified as adopters of CVC programs this information was also confirmed and verified using the VentureXpert database. These firms form our risk set for the abandonment decision; however, the time of entry into the risk set is conditional on the year of adoption of CVC programs so we have an unbalanced panel of observations. Our data covers the time period 1992-2003 where 1992 is the earliest instance of adoption of a CVC program. Variable Definitions Adoption Model

To operationalize the dependent variable for our adoption model, we relied primarily upon the Corporate Venturing Yearbook and Directory (2000, 2001, 2002) which reports the year in which firms adopted CVC programs. We draw upon the innovation diffusion literature to select a set of independent variables used in estimating the probability of CVC program adoption. Since CVC program is an example of an innovation that crossed population boundaries – from the population of private VC firms to the population of IT corporations – we included both cross-population and within-population contagion influences (Gaba and Meyer, 2008). Cross-population contagion arises from IT firms’ proximity to the VC population, their innate familiarity with VC practices, coupled with macro evidence documenting the efficacy of VC practices. Within-population contagion captures the influence of prior IT adopters. The cross-population contagion variables include geographic distance from the VC population (details on this measure is given in the next subsection), a dummy variable for whether a firm was itself venture-backed at the time of founding, and the total number of venture backed IPOs

22

in a given year. IT firms headquartered closer to the VC population are better placed to observe and learn about the VC practices and are better positioned to adopt CVC programs. Venturebacked IT firms are likely to possess a congenital receptivity to VC practices, to retain some ties to the VC community and seem more likely to adopt CVC programs. Finally, increases in the incidence of VC-backed IPOs indicates superior returns for the venture capital model and signals to prospective adopters the potential effectiveness of VC practices in commercializing innovative technologies (Gompers & Lerner, 2001). We obtained data on all three variables from the VentureXpert database. Within-population contagion variables include CVC programs’ popularity (measured by number of prior adopters within the 3 digit SIC industry), proximate prior IT adopters (the number of prior adopters in the same state as the focal firm), and prominent prior IT adopters as measured by the average number of patents by prior adopters (Haunchild & Miner, 1997; Burns & Wholey, 1993; Davis & Greve, 1997). Finally, we include firm-level controls such as size, slack, age, and internal R&D outcome (details on data sources and operationalization are provided in the next subsection). Abandonment Model Dependent Variable

To operationalize the abandonment of CVC programs, we relied primarily upon the VentureXpert database that provides information on CVC investments. In open-ended interviews with corporate venture capitalists, informants told us that when IT firms cease direct investments in technology startups for at least two calendar years, they almost always abandon their CVC program. Following their recommendation, we treated any firm that had not made any investments for at least two years as having abandoned the CVC program. The first year in this

23

interval was coded as the year of abandonment2. Since this is an indirect measure of abandonment decisions, we sought to confirm our coding through Lexis-Nexis searches of business press articles and venture capital newsletters, from websites of IT firms, and through direct contact with Business Development executives. Furthermore, we stratified firms into dual categories of ‘abandoners’ and ‘retainers’ on the basis of the VentureXpert database, and drew a stratified random sample from each category. For every firm in this stratified random sample, we made strenuous efforts to confirm the abandonment of CVC program by getting in touch with the Business Development executives. In all cases, our original coding of the abandonment decision based on the two year interval tallied with the information that we received directly from these firms. Out of the 92 adopters of CVC programs, more than half (48) abandoned it subsequently. Independent Variables

Innovation Related Experience Variables Time of Adoption. We use time of adoption as a proxy to capture depth of understanding of the VC practices at the time of adoption. Instead of relying on an arbitrary time period to classify firms into early vs. late adopters we use the year of adoption as the independent variable. This is simply the year that the firm adopted the CVC programs from the Corporate Venturing Yearbook and Directory (2000, 2001, 2002). The variable ranges from 1992 to 2001.3 Geographic Distance to VC clusters. Silicon Valley houses the world’s dominant cluster of private venture capital firms. In 2000, approximately 40 percent of all U.S. venture capital originated in Silicon Valley. The next two important regions are New York and Route 128 in New England, accounting for 12 and 11 percent of VC investment, respectively. We measured

2

Even though our analysis stops in the year 2003 we analyzed the investments of firms in the risk set in the year 2004 and 2005 in order to code their abandonment decision. 3 We also operationalized ‘late adopters’ as a categorical variable that takes the value one if the year of adoption was after 1998. The results remain unchanged. However, this classification is coarse and somewhat arbitrary.

24

geographic distance as a weighted average of the number of miles from corporate headquarters to each of these three clusters. The weights were chosen as the density of VC funds targeting IT startups in each cluster, lagged by a year. Thus distance of firm i to VCs at time t was calculated as d it −1 =

∑d θ

ij j = SV ,128, NY

jt −1

where dij is the distance between firm i and cluster j (j = Silicon Valley,

Route 128 and New York) and θjt-1 is the proportion of VC funds in cluster j at (t-1) with

∑θ

jt −1 j = SV ,128, NY

= 1 . The zip code for corporate headquarters was obtained from the Forbes lists. We

classified the counties Alameda, Contra Costa, Marin, San Francisco, San Mateo, and Santa Clara as comprising Silicon Valley; Essex, Middlesex, Suffolk, and Norfolk as comprising Route 128; and New York, Bronx, Kings, Queens, and Richmond as comprising New York. Distance to each cluster was calculated as the number of miles from headquarters to the most proximate of the counties that make up a particular cluster, using a spherical geometry formula to calculate distance between zipcodes (Sorenson & Stuart, 2001). Distance (miles) = 3963.0 * arccos(sin(zip1.lat) * sin(zip2.lat) + cos(zip1.lat) * cos(zip2.lat) *cos(zip2.lon - zip1.lon))

CVC Outcome. While VCs realize returns on their investments by taking their portfolio companies public, IT firms arguably, are less concerned with such direct financial returns. Instead, they look more for strategic gains from CVC programs. Through these programs firms can screen private entrepreneurial ventures at the forefront of emerging technologies that can significantly impact the competitive dynamics of their industries. If the entrepreneurial startup turns out to be strategically relevant for the IT firm they are likely to acquire it to reap the benefits of the research and development performed by the startup. The acquisition of private entrepreneurial companies is therefore one of the visible indicators of the health of CVC

25

programs (Gompers and Lerner, 1999). We use the cumulated number of CVC backed acquisitions by each firm (lagged by a year) as a measure of the efficacious results experienced by the adopters of CVC programs4. We collected these data from two sources. First, we used the Securities Data Corporation’s (SDC) Mergers and Acquisitions database to obtain a list of all private acquisitions by all the CVC adopters in our sample for each of the years from 1992-2003. We then matched the acquisitions by each adopter with the VentureXpert database to include only those acquisition targets that the CVC adopters had invested in during the time period of the study. Finally, we cumulated such acquisitions since the year of adoption to obtain a measure of results experienced by the adopters of CVC programs. Controls

Our baseline model includes three contagion variables: Number of Prior Abandoners, Proximate Prior Abandoners and Prominent Prior Abandoners. For the first contagion variable we measure the number of prior abandoners in the same 3-digit industry sector lagged by a year (Burns & Wholey, 1993). By confining this measure to the same 3-digit SIC code, we recognize that potential abandoners tend to pay greater attention to more comparable organizations (Haveman, 1993). Next, we calculated the number of prior abandoners in the same geographic state as the focal organization lagged by one year as a measure of proximate prior abandoners. (Burns and Wholey, 1993; Davis and Greve, 1997). Finally, we used the average number of patents of all prior abandoners in the same industry sector as a measure of prominent prior abandoners. An increase in this measure implies that traditionally innovative organizations are eschewing the CVC route of externalizing innovation.

4

It is true that IT firms also realize a variety of intangible benefits from CVC programs such as a window to new technologies, environmental scanning, etc. However, these benefits are hard to quantify so finding an allencompassing metric for benefits accruing from CVC programs is not possible. Here we trade-off comprehensivenss for ease of measurability.

26

We include four firm level control variables – Size, Slack, Age, and Internal R&D Performance. We measured size as total sales by the firm in the appropriate calendar year, slack as a firm’s current ratio, defined as current assets divided by current liabilities (Bourgeois, 1981) and age as the number of years since founding. Both the size and slack measures are constructed from the Compustat tapes. For organizational age we obtained firms’ founding dates from Standard & Poor’s Million Dollar Directory, with missing observations extracted from the FIS online database compiled by Moody's Financial Information Services. We measure firms’ internal R&D outcome by cumulating the number of patents applied (lagged by a year), since adoption of the CVC program and divided this count by corporate sales to control for any size effects (Cohen, Levin & Mowery, 1987). We use patent application filing dates to assign a patent to a firm in a given year, because firms have a strong incentive to apply for the patent as soon as the innovation is completed (Halle, Jaffe, & Trajtenberg, 2001). These data were obtained from a database compiled by Hall et al. (2001) containing information on all utility patents granted between January 1, 1963 and December 30, 1999. Since this data ends at 1999, we augmented the Hall et al. dataset by collecting primary data on patents granted in the years 2000-2003 directly from the USPTO website (http://www.uspto.gov). We also include all patents granted to the subsidiaries of the IT firms in our sample. We use three main databases (SDC’s Mergers & Acquisition Database, Capital IQ, and Directory of Corporate Affiliations) to identify the subsidiaries for all the IT firms in a given year. After identifying all subsidiaries we assigned the patent applications of these subsidiaries in a given year to the parent firm in our sample. Finally, to account for the boom and busts in public equity markets we controlled for the Return on NASDAQ stock markets. We used the value weighted annual return on NASDAQ (including dividends). Since a disproportionate majority of portfolio investments in the decade of

27

the 1990s were in the IT sector and because the firms in our sample are IT firms as well, the choice of NASDAQ is appropriate. If firms are interested in financial returns then a decline in returns on the NASDAQ should increase the likelihood of abandonments. We also experimented with a couple of other variables. First, we used year dummies for the years 1998-2001. The years 1998-2000 were the years of irrational exuberance, of hot IPO markets, and a boom in venture capital activity. In contrast, the latter half of 2000 saw the bursting of the tech bubble – equity markets crashed and a VC activity experienced a profound slump. Second, we used the number of VC backed IPOs to capture the temporal ebbs and flows in VC activity. All our results are robust to these alternate specifications. ---------------------------------------------------Insert Table 1 here ----------------------------------------------------Table 1 provides summary statistics for the variables used in the abandonment model and shows correlations between the predictor variables. RESULTS

----------------------------------------------------Insert Table 2 here ----------------------------------------------------Table 2 shows the maximum-likelihood estimates of two models predicting likelihood of abandonment of a CVC program, using a Heckman selection model. These models account for the self-selection of the IT firms into the risk set for abandonment. Model 1 is our baseline model and includes the three contagion variables (prior abandonments in the same industry, prominent prior abandoners, and proximate prior abandoners) as well the other control variables (internal R&D outcome, age, slack, size, and the return on NASDAQ). We find that the prior abandonment decisions of firms in the same industry strongly accelerate the abandonment

28

decisions of the focal firm. The coefficient on prior abandonment in the same geographic area is also positive and significant. Finally, prominent prior abandoners who are known to be innovative, exercise a strong contagion influence on the focal firm. Thus, in accordance with the existing research on innovation diffusion, we find that IT firms are influenced by the abandonment decisions of their peers, by those that are prominent and those that are proximate (Greve, 1995; Abrahamson and Fairchild, 1999). To evaluate the relative influence of three contagion variables, we calculate the magnitude of their influences in model 1. The marginal effects are evaluated at the means of the independent variables. We find that the both the prior abandonments in the same industry and prominence of prior abandoners are most influential – a 1% increase in the former raises the probability of abandonment by 0.30% while a 1% increase in the latter raises the probability of abandonment by 0.37%. In contrast, the influence of abandonments in the same geographic area variables is of a considerably smaller magnitude – a 1% increase raises the probability of abandonment by only 0.14%. In addition, firms with internal R&D programs that have been successful in terms of generating patents are more likely to abandon CVC programs. Our results suggest that favorable results from an innovation enhance the likelihood of continuance while favorable results from its close alternates increase the likelihood of discontinuance. However, none of the other firm level controls such as size, slack or age of the firm seems to consistently and significantly affect the abandonment decision5. We also find that the temporal effects as captured by the return NASDAQ play a role. With respect to the return on NASDAQ, we find that a decline in the NASDAQ (which occurred in the year 2000) leads to CVC program abandonment. That a slump in public equity markets would lead VCs to pare back their activities is expected; for them taking the startup public is the most

5

We used assets as a measure of size as well. However, with a correlation between sales and assets of 0.95 it makes very little difference which measure of size we use.

29

desirable outcome. But given that IT adopters are more likely to abandon CVC programs as returns on stock markets dry up suggests that at least for a subset of firms, the financial return on CVC programs looms large in their decision to persist with these programs. Model 2 examines the role of innovation related experience in the abandonment of CVC programs. Our results support Hypotheses 1-3. First, late adopters of CVC programs are more likely to abandon the CVC programs – we obtain a positive and significant coefficient on the year of adoption. As we argued, early adopters of CVC programs have a deeper understanding of the VC model, and are more likely to customize CVC programs to their organizational contexts. These firms retain CVC programs even in the face of a market downturn. On the other hand, late adopters of CVC programs who adopt for faddish or symbolic reasons, lack a basic understanding of how CVC programs lead to valued outcomes. As public markets collapse and observing their peers abandoning these programs, they follow suit. Second, we find that IT adopters geographically distant from VC clusters, who are at a relative disadvantage in terms of learning and implementing the VC practices are more likely to abandon CVC programs. On the other hand, IT adopters embedded in the primary VC clusters, who have greater access to tacit knowledge, and skilled personnel and who are comfortable with the culture, language, and norms of the VC model are more likely to retain their CVC programs. Third, firms who were successful in their CVC programs, in terms of acquiring startup companies in which they had invested in through their CVC program, are less likely to abandon these programs. These firms realize strategic returns from their CVC programs and are more likely to view these programs as a useful way to access externally developed technologies and products. The coefficients on the contagion variables remain positive and significant. In terms of magnitude of effects, distance from the VC clusters exercises the biggest influence. The median firm located outside the

30

counties that form the VC clusters in Silicon Valley, New York and Route 128 is about 1200 miles away from the VC population. The estimated coefficients suggest that this firm is nearly four times as likely to abandon the CVC program. Second, both success with CVC programs and the timing of adoption play a role as well. A 1% increase in acquisitions through CVC programs reduces the probability of abandonment by 0.18%, while a one year delay in the adoption of CVC programs raises the probability of abandonment by about 3.7%. A Wald statistic (=14.48) comparing Models 1 and 2 indicates a significant improvement in model fit. Finally, the last two rows in table 3 show that a) the test of the joint insignificance of all the explanatory variables in both models 1 and 2 is strongly rejected; and b) the hypothesis of independence of the adoption and abandonment decision, that ρ = 0 is consistently rejected. This indicates that the Heckman selection-bias correction procedure is warranted. Next, we examine the role of innovation related experience in moderating the potency of contagion influences. Hypothesis 4 predicted that IT adopters with richer innovation related experience would be less subject to influences arising from the actions of prior abandoners. Accordingly, we test Hypothesis 4 by interacting each of the innovation related experience variables with the contagion variables. ----------------------------------------------------Insert Tables 3 and 4 here ----------------------------------------------------To test Hypothesis H4, first, we interact the year of adoption with each of the three contagion variables. We expect that late adopters are especially susceptible to contagion influences. Models 3-5 support this contention and we find that the interaction term in each model is positive and significant. Late adopters who have had less experience with the CVC programs react more strongly to contagion influences that originate from the abandonments by

31

other IT adopters within their industry, prominent prior abandoners, and prior abandoners in the same state6. Second, we interact geographic distance to the VC clusters with each of the three contagion variables. Models 6-8 also provide support for our Hypothesis H4. The coefficients on the interaction of geographic distance to VC cluster and contagion variables are positive and significant so that IT adopters geographically distant to VC clusters are more susceptible to the actions of their peers. Alternatively, IT adopters proximate to the VC cluster are relatively immune to and discount these contagion influences. In other words, the contagion effect from abandonment of CVC programs by peers is contingent on proximity to the VC cluster, with proximate firms discounting popularity-based contagion influences in their abandonment decisions. Finally, Models 8-11 show that firms who experience beneficial results from the CVC programs discount the contagion influences so that the coefficient on each of the interaction terms is negative. On the other hand, IT firms who have not acquired start-ups through their venture programs, pay more attention to the contagion pressures emanating from elsewhere. Wald tests show that models 3-11 are a significant improvement on model 2 (without the interaction terms) for all but one model. In sum, our results suggest that not only does innovation related experience make firms less likely to abandon CVC programs but also makes them more immune to the contagion influences. While contagion does play an important role in the abandonment decision, the relative strength of its effect varies with firms’ direct experience with the innovation itself. Early adopters of CVC programs, IT adopters geographically proximate to VC clusters, and IT adopters realizing strategic benefits from CVC programs attach greater weight to their own experience with the innovation and discount the abandonment decision of other IT firms. 6

The small coefficient on the interaction term arises because the year of adoption variable takes large values. We could simply scale it by dividing it by say 1000 – this would not affect the magnitude of the effects. However, the summary statistics for this variable in Table 1 is easier to interpret, in the absence of such scaling.

32

DISCUSSION AND CONCLUSION

This study examines the abandonment of an administrative innovation – corporate venture capital programs - by a sample of information technology firms. In 1990s, corporate venture capital programs diffused widely among IT firms only to be abandoned few years later. Prior research examining lifecycle of innovations views innovation adoption and abandonment as fundamentally similar processes where overtime the diffusion of decision to adopt the innovation turn into diffusion of decision to reject the innovation (Abrahamson and Fairchild, 1999). This suggests that like in the context of innovation adoption, contagion influences biases organizations towards abandonment of an innovation as more and more firms jump on the abandonment bandwagon. Using contagion as a baseline expectation we propose that the decision to abandon the innovation is also influenced by firms’ direct experience with the innovation. We conceptualize innovation related experience broadly to include knowledge gained at the time of adoption, during the implementation of the innovation, and from the results experienced with the innovation. We argued that innovation related experience biases IT adopters against abandoning the innovation. We tested hypotheses based on this argument by collecting longitudinal data from 270 IT firms from 1992 until 2003. The empirical analyses provided support for our predictions. Results suggest that early IT adopters, IT adopters headquartered near venture capital clusters, and IT adopters making strategic gains through the CVC programs are less likely to abandon them. We further theorized that variation in IT adopters’ innovation related experience also leads to heterogeneity in their susceptibility to contagion influences. We reasoned that IT adopters with deeper insights about the venture capital practices are less susceptible to contagion

33

influences. Our results show that early adopters of the innovation, who are less likely to have adopted the CVC program for purely faddish reasons, are also less likely to abandon it. Even as equity markets collapse and peers abandon, these IT adopters continue to implement the innovation and put less weight on the abandonment decisions of their peers. We also find that proximity to the VC population clusters confers an informational advantage that manifests itself in two ways: it makes IT adopters more likely to retain the CVC program and it makes them less susceptible to the contagion influences. Next, IT adopters making large number of acquisitions through their CVC programs are less likely to abandon these programs. These organizations are relatively successful in realizing strategic benefits of the CVC programs and are less concerned with the abandonment decision of their peers. Finally, we find that IT adopters that have been historically successful in generating new ideas, products and technologies through their internal R&D, see less value in CVC programs as a source of innovative ideas. They are more likely to abandon the CVC programs. Our investigation of CVC program abandonment has implications for research on innovation diffusion. We take the pro-innovation bias critique seriously by focusing on the abandonment of an innovation. Much of the diffusion literature focuses on the initial adoption of the innovation and rarely examines what happens afterwards (Abrahamson, 1991; Strang and Macy, 2001). Innovation abandonment is an understudied phase in an innovation’s lifecycle. It is also important because conceptually it helps us understand the relative importance of different mechanisms in the distinct phases of an innovation’s lifecycle. Our study highlights the important role of firms’ direct experience with the innovation in the abandonment phase of the lifecycle. We find that not only do firms pay close attention to their own experience with the

34

innovation, but the weight they attach to the abandonment decision of others is itself a function of this experience. Our study contributes to the current research on organizational learning. Recent studies have shown that firms draw inferences from the behavior of other firms as well as from their own experiences in making a variety of strategic decisions (e.g., Baum, Li & Usher, 2000). While both may be important, we present a more nuanced perspective by showing that an individual firm’s direct experience also moderates its susceptibility to learn from the behavior of its peers. This interaction between experiential and mimetic learning has seldom been investigated in prior theoretical or empirical work (Lieberman and Asaba, 2006). In the postadoption phase, firms continue to make sense of their experience with the innovation through the theorization process. This theorization process renders the innovation understandable, justifiable, and useful to the adopters. The ability to do so also makes them rely more on what they know internally. Poor theorization, on the other hand, makes it more appealing to imitate the actions and behaviors of other firms. Our research also has implications for institutional theory. First, institutional accounts of innovation diffusion implicitly imply that innovations diffuse to the fullest extent (Abrahamson and Rosenkopf, 1997). All potential adopters eventually adopt as they get pulled in by the contagion influences. However, this prediction does not square with observed cyclical patterns of adoption and abandonment (Strang and Macy, 2001). In our study, abandonment decisions by poorly informed organizations, implies that full diffusion even if it does take place, is likely to be reversed. Even as the contagion pressures for abandoning an innovation build up, we are unlikely to see the innovation completely disappear since a subset of organizations will resist these contagion influences. So we are unlikely to observe either a complete institutionalization or a

35

complete deinstitutionalization of an innovation (Oliver, 1992). Furthermore, if a certain set of organizations retain the innovation and experience stellar performance, then a potentially new cycle of adoption (re-adoption to be more accurate) may start. Future research should examine the re-diffusion of an innovation. It is plausible that organizations abandoning the innovation may experience post-decision regret and may want to re-adopt the innovation. If so, do such firms draw upon their prior experience or do they suffer from a memory loss after having abandoned the innovation. Second, institutional theorists have often used contagion as one of the main mechanisms explaining the spread of innovations (Strang and Soule, 1998). Contagion occurs because the innovation is judged to be technically efficient and appropriate by the use of others. Even so, for the innovation to sustain in the long term, it needs to be interpreted, translated, and presented in an understandable and compelling format. Firms that are able to theorize the innovation by justifying its rationale in an appealing way not only have a better chance of retaining the innovations but are also less likely to be swept by fads and fashions characterizing the business world. Future research might further explore the role of theorization in the implementation, retention or abandonment of innovations. Finally, the limitations of our study point to the need for future research. We examine a complex administrative innovation that originates among a distinctive set of firms (private venture capital firms) who agglomerate in a well-defined geographic space (e.g., Silicon Valley). As such the ambiguity and uncertainty associated with adopting this innovation persists even in the post adoption phase so that IT firms continue to draw inferences from the behavior of their peers in addition to their own experience with the innovation. Perhaps the role of contagion influences is conditional on the speed with which the ambiguity surrounding the innovation can be resolved in the post adoption phase. This merits further investigation. Our results also show

36

that firms’ own experience and those of others act as substitutes for one another. It would be interesting to examine if our findings can be extended in the context of innovations that are administratively or technically less complex.

37

References:

Abrahamson, E. 1991. Managerial fads and fashions: The diffusion and rejection of innovation. Academy of Management Review., 16 (3) 586-613. Abrahamson, E., L. Rosenkopf. 1997. Social network effects on the extent of innovation diffusion: A computer simulation. Organization Science, 8(3) 289-309. Abrahamson, E., Fairchild, G. 1999. Management fashion: Lifecycles, triggers, and collective processes. Administrative Science Quarterly, 44: 708-740. Allison, P. D. 1982. Discrete time methods for the analysis of event histories. In S. Leinhardt (Ed.) Sociological Methodology: 61-97. San Francisco, Jossey-Bass. Argote, L., Beckman, S. L. & Epple, D. 1990. The persistence and transfer of learning in industrial settings. Management Science, 36: 140-154. Audia, P.G., E.A. Locke, K.G. Smith. 2000. The paradox of success: An archival and a laboratory study of strategic persistence following radical environmental change. Academy of Management Journal., 43(5) 837-853. Bahrami, H., S. Evans. 2000. Flexible recycling and high-technology entrepreneurship. M. Kenney, ed. Understanding Silicon Valley. Stanford University Press, Stanford, CA. Baum, J.A.C., S.X. Li, J.M. Usher. 2000. Making the next move: How experiential and vicarious leanring shape the locations of chain’s acquisition. Administrative Science Quarterly,, 45 766801. Bourgeois, L.J. 1981. On the measurement of organizational slack. Academy of Management Review, 6: 29-39. Burns, L. R., D. R. Wholey. 1993. Adoption and abandonment of matrix management programs: Effects of organizational characteristics and interorganizational networks. Academy of Management Journal, 36(1) 106-138. Chesbrough, H. W. 2000. Designing corporate ventures in the shadow of private venture capital. California Management Review, 42: 31-49. Cohen, W. M., Levin, R.C. & Mowery, D. C. 1987. Firm size and R&D intensity: A reexamination. Journal of Industrial Economics, 35: 543-565. Cole, R. E.1985. The macropolitics of organizational change: A comparative analysis of the spread of small-group activities. Administrative Science Quarterly, 30: 560-585. Corporate Venturing Directory and Yearbook. 2000, 2001, 2002. Mass: Asset Alternative.

38

Davis, G. F. & Greve, H. R. 1997. Corporate elite networks and governance changes in the 1980s. American Journal of Sociology, 103: 1-37. DiMaggio, P. J. & Powell, W. W. 1983. The iron cage revisited: Institutional isomorphism and collective rationality in organizational fields. American Sociological Review, 48: 147-160. Doerflinger, T. M. & Rivkin, J. L. 1987. Risk and reward: Venture capital and the making of America's great industries. New York, NY: Random House. Florida, R.L & Kenney, M. 1988. Venture capital high technology and regional development. Regional Studies, 221: 33-48. Gaba, V., A.D. Meyer. 2008. Corporate Venture Capital: The Cross-Population Contagion of an Endemic Innovation. Academy of Management Journal, Forthcoming. Gomper, P. & Lerner, J. 1999. The venture capital cycle. Cambridge, MA: MIT Press. Gomper, P. & Lerner, J. 2001a. The money of invention. Boston, MA: Harvard University Press. Greve, H.R. 1995. Jumping ship: The diffusion of strategy abandonment. Administrative Science Quarterly, 40: 444-473. Greve, H. R. 1998. Managerial cognition and the mimetic adoption of market positions: What you see is what you do. Strategic Management Journal, 19: 967-988. Greve H. R. 2003. Organizational Learning from Performance Feedback: A Behavioral Perspective on Innovation and Change. Cambridge University Press, Cambridge, MA. Hall, B. H, Jaffe, A.B. & Trajtenberg. M. 2001. The NBER patent citation data file: Lessons, insights, and methodological tools. NBER Working Paper, 8498. Hamel, G. 1999. Bringing Silicon Valley inside. Harvard Business Review. 77(5): 70-84 Haunschild, P. R. & Miner, A. S. 1997. Modes of imitation: The effects of outcome salience and uncertainty.” Administrative Science Quarterly, 42: 472-500. Haveman, H. 1993. Follow the leader: Mimetic isomorphism and entry into new markets. Administrative Science Quarterly, 38: 593-627. Heckman, J., Sample selection bias as a specification error. Econometrica, 47: 153-161. Kenney, M. & Florida, R. 2000. Venture capital in Silicon Valley: Fueling new firm formation. In M. Kenney (Ed.), Understanding Silicon Valley: Anatomy of an entrepreneurial region: 98123. Stanford: Stanford University Press.

39

Lant, T. K., F. J. Milliken, B. Batra. 1992. The role of managerial learning and interpretation in strategic persistence and reorientation: An empirical investigation. Strategic Management Journal, 13 585-608. Levitt, B., J.G. March. 1988. Organizational learning. In Annual Review of Sociology. 14 319340. Palo Alto, CA Lieberman, M.B., S. Asaba. 2006. Why do firms imitate each other? Academy of Management Review, 31(2): 366-385. Oliver, C. 1991. The Antecedents of Deinstitutionalization. Organization Sci., 13 563-588. Pisano, G. P. 1990. The R&D boundaries of the firm: An empirical analysis. Administrative Science Quarterly, 35 153-176. Rogers, E. M. 1995. Diffusion of Innovations. 4th ed. New York: Free Press. Rao, H., Greve, H. R. & Davis, G. F. 2001. Fool's gold: Social proof in the initiation and abandonment of coverage by Wall Street analysts. Administrative Science Quarterly, 46(September) 502-526. Shaver, J.M., Mitchell, W. & Yeung, B. 1997. The effect of own-firm and other-firm experience on foreign direct investment survival in the United States 1981-1992. Strategic Management Journal., 18: 811-824. Sorenson, O. & Audia, P.G. 2000. The social structure of entrepreneurial activity: Geographic concentration of footwear production in the United States, 1940-1989, American Journal of Sociology, 106: 424-462. Sorenson, O. & Stuart, T. E. 2001. Syndication networks and the spatial distribution of venture capital financing. American Journal of Sociology, 106: 1546-1588. Sorenson, O., Rivkin, J.W. & Fleming, L. 2006. Complexity, networks and knowledge flow. Research Policy, 35: 994- 1017. Strang, D. & Meyer, J. W. 1993. Institutional conditions for diffusion. Theory and Society, 22: 487-512. Strang, D. & Soule, S.A. 1998. Diffusion in organizations and social movements: From hybrid corn to poison pills. Annual Review of Sociology, 24: 265-290. Strang, D., & Macy, M. W. 2001. In search of excellence: Fads, success stories, and adaptive emulation. American Journal of Sociology, 107: 147-82. Tolbert, P. S. & Zucker, L. G. 1983 Institutional sources of change in the formal structure of organizations: The diffusion of civil service reform, 1880-1935. Administrative Science Quarterly, 28: 22-39.

40

Tolbert, P. S. & Zucker, L. G. 1996. Institutionalization of institutional theory. In s. Clegg, C. Hardy, W. Nord (eds.) The Handbook of Organization Studies, 175-190. Thousand Oaks, CA: Sage. Van De Ven, W.P.M.M. & Van Praag, B.M.S. 1981. The demand for deductibles in private health insurance: A probit model with sample selection. Journal of Economoetrics, 17: 229-252. Venture Economics 2005. Thompson Financial Securities Data Company (SDC), New York: Westphal, J. D., Gulati, R. & Shortell, S.M. 1997. Customization or Conformity: An Institutional and Network Perspecitve on the Content and Consequences of TQM Adoption. Administrative Science Quarterly, 42: 366-394. Zucker, L. G. 1988. Where do institutional patterns come from? Organizations as actors in social systems. In L.G. Zucker (ed.) Institutional Patterns and Organizations. Ballinger Publishing Co. Cambridge, MA, 23-52.

41

Figure 1

Corporate Venture Capital 180

7000

160 6000

120 4000

100

80

3000

60 2000 40 1000 20

0 1992

1993

1994

1995

1996

1997

1998

1999

Year Total Investments

42

No. of firms

2000

2001

2002

2003

Number of Firms

Corporate Investments ($$)

140 5000

Figure 2

Innovation Related Experience Depth of learning at the time of adoption

Information acquired during implementation

Adopter

Outcome experienced

43

Table 1: Descriptive Statistics and Correlations (Abandonment Model) Variable 1. Distance to VC clusters it-1a 2. Year of adoption i 3. CVC outcome it-1a 4. Internal R&D outcome it-1a 5. Prior abandoners (in 3 digit industry)it-1 6. Proximate prior abandoners it-1

Mean 4.89 1997.02 0.28 -4.78

S.D. 2.39 2.7 0.61 2.93

1

2

3

4

1 -0.14 0.06 -0.16

1 -0.01 -0.02

1 0.13

1

2.39

4.21

-0.11

0.26

0.01

0.01

1

2.26

3.65

-0.4

0.32

0.09

0.2

0.51

1

7. Prominent prior abandoners it-1

16.39

29.59

-0.06

0.14

0.13

0.15

-0.15

-0.02

1

8. Sales ita 9. Slackita 10. Ageita 11. Return on NASDAQit

7.84 0.71 3 0.09

1.98 0.68 0.86 0.43

0.29 -0.26 0.21 0.03

-0.16 0.17 -0.22 -0.22

0.29 0 0.12 -0.02

0.16 0.23 0.38 -0.05

-0.19 0.09 -0.18 -0.34

-0.17 0.18 -0.17 -0.34

0.05 0.05 0.08 0.41

a: these variables are logged

44

5

6

7

8

9

10

11

1 -0.51 0.51 0.02

1 -0.29 0

1 0.03

1

Table 2: Analysis of Innovation Abandonment – Heckman Selection Model (1) Year of adoption i Distance to VC clusters it-1 CVC outcome it-1 Prior abandoners (by industry) it-1

0.094*** (0.017)

Proximate prior abandoners it-1

0.044** (0.021)

Prominent prior abandoners it-1

0.010*** (0.003)

Internal R&D outcome it-1 Sales it

0.058 (0.073)

Slack it

0.091 (0.136)

Age it

0.012 (0.108)

Return on NASDAQ it

-0.749** (0.305)

Constant

-3.013*** (0.708)

(2)

0.037** (0.018) 0.095*** (0.024) -0.183* (0.128) 0.097*** (0.017) 0.052*** (0.006) 0.009*** (0.003) 0.078*** (0.028) 0.060* (0.035) 0.012 (0.049) -0.089 (0.086) -0.579* (0.318) -76.481** (35.508)

No. of observations

2273

2273

No of observations at risk

413

413

No of firms at risk

92

92

No of abandonments

48

48

0.767**

0.707***

105.96***

115.5***

Correction for self-selection(ρ) χ2 test of covariates all zero [7 (11) d.f for model 1(2)]

Standard errors in parentheses adjusted for clustering on 3 digit SIC codes; * significant at 10%; ** significant at 5%; *** significant at 1% (one-tailed for hypothesized variables, 2 tailed for controls)

45

Table 3: Analysis of Innovation Abandonment – Heckman Selection Model (3) 0.046*** (0.012) 0.093*** (0.016) -0.200* (0.140) 0.00001***

Year of adoption i Distance to VC clustersit-1 CVC outcome it-1 Year of adoption i * Prior abandoners (in 3 digit industry) it-1

(4) 0.047*** (0.014) 0.100*** (0.016) -0.180* (0.132)

(5) 0.047*** (0.013) 0.115*** (0.028) -0.233* (0.171)

(6) 0.025* (0.017) 0.104*** (0.030) -0.198* (0.131)

(7) 0.036* (0.028) 0.060* (0.041) -0.171 (0.167)

(8) 0.020 (0.021) 0.102*** (0.031) -0.234* (0.167)

(0.000003) Year of adoption i * Proximate prior abandoners it-1

0.00002* (0.00001)

Year of adoption i * Prominent prior abandoners it-1

0.000004 ** (0.000002)

Distance it-1* Prior abandoners (in 3 digit industry) it-1

0.006*** (0.002)

Distance it-1* Proximate prior abandoners it-1

0.005* (0.003)

Distance it-1* Prominent prior abandoners it-1 Prior abandoners (in 3 digit industry) Proximate prior abandoners

it-1

Prominent prior abandoners

it-1

Internal R&D outcome it-1 Sales it Slack it Age it Return on NASDAQ it Constant No of observations No of observations at risk No of firms at risk No of abandonments Correction for self-selection(ρ) χ2 test of covariates all zero [12 d.f]

0.009**

it-1

0.104*** (0.017) 0.054*** (0.010) 0.007* (0.005) 0.081*** (0.029) 0.090*** (0.025) -0.059 (0.082) 0.056 (0.068) -0.410 (0.274) -95.545*** (23.387) 2273 413 92 48 0.78*** 142.78***

0.114*** (0.017) 0.037** (0.015) 0.005* (0.004) 0.073** (0.032) 0.077** (0.037) -0.063 (0.091) 0.028 (0.037) -0.439* (0.232) -95.84*** (27.784) 2273 413 92 48 0.73*** 136.25***

0.105*** (0.021) 0.053*** (0.008) 0.044** (0.018) 0.073** (0.034) 0.011 (0.054) -0.054 (0.108) -0.001 (0.028) -0.763*** (0.190) -95.83*** (26.768) 2273 413 92 48 0.46* 120.74***

0.082*** (0.014) 0.059*** (0.010) 0.005** (0.003) 0.082*** (0.028) 0.072** (0.034) -0.091 (0.079) 0.037 (0.055) -0.274 (0.358) -51.779 (34.045) 2273 413 92 48 0.758*** 157.93***

0.100*** (0.012) 0.030** (0.015) 0.006*** (0.001) 0.074** (0.033) 0.079** (0.039) -0.082 (0.066) 0.012 (0.047) -0.534* (0.299) -74.695 (55.609) 2273 413 92 48 0.79** 120.99***

Standard errors in parentheses adjusted for clustering on 3 digit SIC codes; * significant at 10%; ** significant at 5%; *** significant at 1% (one-tailed for hypothesized variables, 2 tailed for controls)

46

(0.004) 0.084*** (0.022) 0.040*** (0.007) 0.008** (0.004) 0.064** (0.031) 0.021 (0.056) -0.080 (0.106) 0.007 (0.044) -0.931*** (0.302) -41.965 (41.938) 2273 413 92 48 0.57* 127.28***

Table 4: Analysis of Innovation Abandonment – Heckman Selection Model (9) 0.038** (0.019) 0.095*** (0.030) -0.182* (0.132) -0.194* (0.132)

Year of adoption i Distance to VC clusters it-1 CVC outcome it-1 CVC outcome it-1* Prior abandoners it-1 CVC outcome it-1 * Proximate prior abandoners

it-1

CVC outcome it-1 * Prominent prior abandoners

it-1

Prior abandoners (by industry) Proximate prior abandoners

it-1

Prominent prior abandoners

it-1

Internal R&D outcome it-1 Sales it Slack it Age it Return on NASDAQ it Constant No of observations No of observations at risk No of firms at risk No of abandonments Correction for self-selection(ρ)

χ2 test of covariates all zero [12 d.f]

(11) 0.037** (0.016) 0.096*** (0.011) -0.156*** (0.034)

-0.111* (0.086)

0.099*** (0.017) 0.048*** (0.006) 0.009*** (0.003) 0.081*** (0.027) 0.060 (0.048) 0.005 (0.040) -0.107 (0.091) -0.617* (0.319) -79.380** (38.380) 2273 413 92 48 0.686** 126.52***

it-1

(10) 0.040* (0.030) 0.070*** (0.019) -0.180 (0.167)

0.108*** (0.014) 0.038*** (0.010) 0.004** (0.002) 0.070* (0.036) 0.088* (0.047) 0.040 (0.033) -0.086 (0.064) -0.547* (0.302) -83.223 (58.990) 2273 413 92 48 0.781** 116.54***

-0.002** (0.001) 0.095*** (0.000) 0.052*** (0.012) 0.010*** (0.002) 0.076*** (0.005) 0.055 (0.034) 0.003 (0.072) -0.093*** (0.002) -0.623*** (0.012) -76.771** (33.180) 2273 413 92 48 0.681*** 118.56***

Standard errors in parentheses adjusted for clustering on 3 digit SIC codes; * significant at 10%; ** significant at 5%; *** significant at 1% (one-tailed for hypothesized variables, two-tailed for controls)

47