An Evolutionary Approach to Leveraging Knowledge into New Technologies and Applications

Transcription

1 An Evolutionary Approach to Leveraging Knowledge into New Technologies and Applications Preeta M. Banerjee Brandeis University International Business School 415 South St. Waltham, MA Douglas J. Miller University of Illinois at Urbana-Champaign Department of Business Administration 1206 S Sixth St Champaign IL djmiller@uiuc.edu DRAFT Please do not cite or quote without permission of the authors 1

2 Abstract: We look at innovation as a diversification decision based on existing knowledge competences. By taking an evolutionary view, that these decisions are not point-in-time but pathdependent, we can better understand how innovation begins at firm start-up and progresses into future development and firm performance. We hypothesize that firms in emerging, technologyintensive industries will pursue multiple paths to growth, and that prior experience and current resource constraints will impact their choice of paths. Also, firms will have more significant innovation the more they leverage their existing knowledge about technology or markets. We study these issues using detailed information on over 100 publicly-traded firms in the U.S. biotechnology industry. Findings have implications for our understanding of both knowledge management in growing firms and corporate strategy decisions. Keywords: path-dependence, innovation, cross-application 2

3 Literature on innovation has highlighted the positive impact of existing skills and knowledge on innovation and firm performance, especially when firms redeploy that experience across new markets and/or industries (e.g. Carroll, Bigelow, Seidel and Tsai, 1996; Holbrook, Cohen, Hounshell, Klepper, 2000; Klepper, 2002; Klepper and Simons, 2000). In particular, firms diversify by building on areas of competence as the underlying resources are fungible, with a range of potential (and profitable) uses (Danneels, 2002; Miller, 2006; Penrose, 1959). In other words, a firm can apply the capabilities learned, the resources earned, and the competences built in one situation to serve a different market or opportunity. However, the decision of how to diversify based on existing competences has been looked at as a point-in-time question given the tradeoffs between exploration and exploitation in innovation (Danneels, 2007; Greve, 2007; Miller, 2006). Exploration and exploitation rely on different organizational routines and capabilities (Benner and Tushman, 2003; Galunic and Eisenhardt, 2001; Lewin, Long, and Carroll, 1999). In innovation, exploitation of existing stocks of knowledge appears to reduce incentives for exploring new knowledge and even the ability to do so in the future (Christenson and Bower, 1996; Leonard-Barton, 1995; Levinthal and March, 1981; Tushman and Anderson, 1986). Yet, theory and empirical evidence suggests that too little of either exploration or exploitation reduces performance (Fagiolo and Dosi, 2003; He and Wong, 2004; Katila and Ahuja, 1002; Levinthal and March, 1993). Moreover, innovation is complex and path dependent in the accumulation of firm-specific technological competencies (Patel and Pavitt, 1997). Thus, it is important to understand how earlier decisions impact later firm decisions and, ultimately, performance outcomes of the firm. We propose a path dependent theory of diversification and evolution of firm technology innovation that outcomes depend on the path of previous outcomes rather than simply on 3

4 current conditions (Lambert and Tikkaken, 2006) - which contributes to extant literature in two main ways. Firstly, we contribute to the studies on diversification that have rarely even attempted to distinguish between firms that moved into new product markets through exploration of technologies new to the firm versus diversifying on the basis of leveraging existing technology. The distinction is important because the two paths to corporate diversification imply different competencies, decision-making processes, timing of entry, and risks. Secondly, we investigate the entire life of innovation in a firm as innovation decisions are more transparent in smaller, younger firms than are often included in studies of corporate diversification. Start-up firms are often closely identified with a single technology and entrepreneurial managers make decisions about how to grow the size and scope of the firm that have long-term implications. These decisions are made under significant uncertainty and time constraints. In this paper, we hypothesize that firms in emerging, technology-intensive industries will pursue multiple paths to growth, and that prior experience and current resource constraints will impact their choice of paths. Also, firms will have more significant innovation the more they leverage their existing knowledge about technology or markets. We study these issues using detailed information on over 100 publicly-traded firms in the U.S. biotechnology industry. Along with defining a measure of leveraging technology into new applications (which we call crossapplication ), we also record when firms pursue alternative paths of technological evolution. The findings have implications for our understanding of both knowledge management in growing firms and corporate strategy decisions. Theory Section 4

5 The earliest statements of resource-based ideas on diversification recognized the inherently dynamic nature of corporate strategy. As Wernerfelt (1984: 176) writes in the paper that introduced the term resource-based view, Although the general idea is to expand your position in a single resource, it is not always optimal to go full force in several markets simultaneously, even with experience curve effects. Quite often, it is better to develop the resource in one market and then to enter other markets from a position of strength sequential entry. Thus, many successful firms expand into new markets based on a core competence (Hamel and Prahalad, 1990), diversifying only after establishing superior efficiency or a differentiation advantage in one industry. Other firms with less competitive assets (i.e., inferior technology) may search across markets to find one that will be a profitable alternative use of those assets (Montgomery and Wernerfelt, 1988). Zollo and Winter (2002) propose that core competences develop out of the activities of the Knowledge Evolution Cycle (Figure 1) adopted from the classic evolutionary paradigm of variation-selection-retention. Managers select from a combination of external stimuli (competitor s initiatives, normative changes, scientific discoveries) with internally generated information derived from the organizations existing routines. These selections become innovation routines when replications of novel knowledge solutions are diffused throughout the firm. As Garud & Nayyar (1994) explicate, there are two main kinds of uncertainty to which firms generate novel knowledge solutions: supply uncertainty and demand uncertainty. Thus, a firm can choose to develop its technology to respond to these uncertainties through exploring or leveraging (exploiting) two types of knowledge: technological knowledge and market knowledge. Technological knowledge is that dealing with the underlying science, engineering 5

6 principles and design; in other words, technological competence. Market knowledge is that of customers and competitors; in other words, customer competence. Thus, a firm can choose to develop its technology to respond to these uncertainties through exploring new knowledge or leveraging (exploiting) existing knowledge. Variation from which managers select is found in the types of innovation that the firm pursues. One type of innovation, established in previous work by the investigators, is the concept of cross-application, which we define as the entrepreneurial effort (Schumpeter, 1934) of leveraging existing firm technological knowledge into new application domains. Crossapplication has two aspects. First, the firm matches fixed technological attributes (versus performance attributes, i.e., Christensen, 1997) to specific customer preferences in the markets and sub-markets by placing different weights on the attributes of the technology. Second, the firm recombines new knowledge elements or complementary assets with the technology elements, without changing the technology itself. We argue that cross-application often occurs very early in the life of a firm, due to predictable drivers, and establishes an evolutionary process that affects the later growth of the firm, its corporate diversification, and its vertical integration. This cross-application 1 of technology leads to greater product market scope; not necessarily in the form of diversified firms, but derivation of profits from multiple markets through internalized activities or licensing. The evolution of the firm s proprietary technological knowledge may also occur along a predictable trajectory: a technological trajectory (Dosi, 1982; Nelson and Winter, 1982). In this case, the available attributes consistently change toward greater efficiency or effectiveness as the technology matures (Fleming, 2001; Sahal, 1985; Foster 1986) and as exogenous developments 1 The name cross-application is taken from the software industry, where the concept references programming code that is utilized across an entire system. 6

7 in science unfold (Dosi 1982; Fleming and Sorenson, 2004; Nelson and Winter, 1982). A firm s growth is not bounded by the existing technological trajectory. A technology may apply to many different products or services, each of which has its own evolving market along a consumer preference trajectory (Adner, 2002; Adner & Levinthal, 2001; Christensen, 1997; Tripsas, 2007). Consumer preferences can determine which performance attributes made possible by a technology are most valuable, and whether there are distinct submarkets, perhaps with different willingness-to-pay for products of different levels of quality. Consumer preference trajectories may be either incremental or discontinuous changes in new functionality or features. Lastly, a firm can pursue pure exploration, in which it is not limited by a technology trajectory or a customer preference trajectory. In such cases, the firm pursues technological and market knowledge that is completely new to the firm. The selection decision of how to diversify based on existing competences has been looked at as a point-in-time question given the tradeoffs between exploration and exploitation in innovation (Danneels, 2007; Greve, 2007; Miller, 2006). Exploration and exploitation rely on different organizational routines and capabilities (Benner and Tushman, 2003; Galunic and Eisenhardt, 2001; Lewin, Long, and Carroll, 1999). In innovation, exploitation of existing stocks of knowledge appears to reduce incentives for exploring new knowledge and even the ability to do so in the future (Christenson and Bower, 1996; Leonard-Barton, 1995; Levinthal and March, 1981; Tushman and Anderson, 1986). Yet, theory and empirical evidence suggests that too little of either exploration or exploitation reduces performance (Fagiolo and Dosi, 2003; He and Wong, 2004; Katila and Ahuja, 1002; Levinthal and March, 1993). As the literature on organizational learning has developed, scholars have noted that exploration and exploitation may coexist, unlike initially argued by March (1991; 1996). An 7

8 ambidextrous firm (Tushman & O Reilly, 1996) can develop dynamic capabilities (Eisenhardt & Martin, 2000) to be able to juggle both the exploration of new technologies and the exploitation of existing ones in real time. He and Wong (2004) have found evidence that manufacturing firms grow more quickly when they match their levels of exploration and exploitation, and that the two processes have a positive interaction effect on growth. Also, a firm that operates in multiple domains may achieve balance by simultaneously exploring in one domain while exploiting in another (Lavie & Rosenkopf, 2006). Similarly, Katila and Ahuja (2002) found that researchers developing inventions could both cite a broad range of patents (search scope, operationalizing exploration) and also cite certain key patents repeatedly (search depth, operationalizing exploitation). The combination of both types of citations had a positive effect on new product development. However, as Gupta, Smith, and Shalley (2006) clarify, exploration and exploitation might be mutually exclusive when the firm is resource-constrained and operating within a single domain. Thus, in small, entrepreneurial firms, exploration and exploitation may be two ends of a continuum, due to significant constraints on financial capital, time, and management attention. Danneels (2002) presents a useful framework of product innovation based on organizational learning theory (e.g., March, 1991) and distinguishing between firm-level competences regarding technology and customers. For example, a firm that leverages customer competence is exploiting a customer competence while exploring a technological competence. As a firm grows, it may move beyond this two-by-two framework of tradeoffs, and into a more complex, multidimensional space. One could draw exploration and exploitation as orthogonal axes in each domain, recognizing that firms may position themselves by doing either or both to different 8

9 degrees, but such a framework becomes easier to consider mathematically rather than graphically. It is also important to understand how managers select upon earlier decisions on knowledge development to later retain and impact firm decisions. Ultimately, performance outcomes of the firm or market feedback guide managers decisions. In the strategic management literature, two types of significance have been used to measure the performance of innovation activities. Firstly, the sheer number of innovations that result from an advance has been used as an indication of firm performance (Rosenkopf and Nerkar, 2001; Nerkar, 2004), relating the number of subsequent innovations to the concepts of technological dynamism (Bresnahan and Trajtenberg, 1995) and importance (Trajtenberg, 1993; Henderson, Jaffe and Trajtenberg, 1995; Jaffe, Fogarty and Banks, 1997; Hall, Jaffe and Trajtenberg, 2001). Technological dynamism represents the potential for continuous innovational efforts and learning as a result of a particular innovation. It has been argued that different technologies have different degrees of technological dynamism technological opportunity simply differs across technologies (e.g. Nelson, 1981; Levin et al., 1985; Dosi, 1988; and Stoneman, 1995). Importance captures the number of descendents of a focal innovation and the impact of those descendents, in number of further descendents (Hall, Jaffe and Trajtenberg, 2001; Jaffe, Fogarty and Banks, 1997; and Henderson, Jaffe and Trajtenberg, 1995). Both approaches used in the management literature highlight the cumulative nature of technology development. Secondly, in the broader economics literature, the breadth of further applications has been used as an additional indication of firm performance. Diversity, or use of technology in a wide range of application domains, can open up new opportunities for both the invention and the inventor. Innovations invented to serve one purpose can eventually find most of their use for 9

10 other unanticipated purposes. For example, while James Watt designed his steam engine to pump water from mines, it soon was supplying power to cotton mills and then, with much greater profit, propelling locomotives and boats (Diamond, 1999). Technologies that possess an extensive range of use are known as enabling technologies (Lipsey and Bekar, 1995) or general-purpose technologies (GPTs) (Bresnahan and Trajtenberg, 1995). In summary, innovation is complex and path dependent in the accumulation of firmspecific technological competencies (Patel and Pavitt, 1997). Doing things in a particular way may yield effects which pre-dispose the firm to do things the same way the next time around. Thus, rather than just investigate how often and why firms are active in a certain sectors of the matrix, we build on Danneels (2002) to explore how a firm chooses between sectors. Either the firm continues to do what it has done or, it moves from one sector to another. The evolution of the firm does not just follow the path laid down by the underlying technology or existing customers. On the one hand, managers may continue the process of creative destruction (Schumpeter, 1934) by incorporating additional knowledge elements into new innovations (e.g., Argyres, 1996; Helfat & Raubitschek, 2000). This is an increase in the technological scope of the firm. On the other hand, there are several reasons why entrepreneurial managers might choose to pause along the technological trajectory to leverage existing technology into additional applications. Thus, there can be significant economic tradeoffs involved in choosing between different methods of technology-based firm growth. Variation, Selection and Retention among Innovation Paths We propose that following the different paths specified by the four types of innovations above will lead firms to different performance outcomes: depth of citations, breadth of citations, and financial performance. We then propose that these innovation paths can be, in some cases, 10

11 complementary and, in other cases, substitutive when simultaneously pursued or juggled. In the final set of hypotheses, we propose that experience in choosing a path truly does constrain the firm. Tests of these hypotheses will provide understanding as to whether the firm s innovation strategy is more than just a point-of-time decision, but rather the outcome of a series of decision points, i.e. path-dependence. Depth When a firm decides to build on existing knowledge rather than explore new knowledge, the firm is hoping to leverage previous successes across new opportunities. In this sense, leveraging involves drawing on an existing competence, while using it as a stepping stone to build a new competence (Danneels, 2002). In terms of impact, it follows that a firm building on existing technology will contribute to the perceived impact of the pioneering innovation. Deeper exploration of the same technology space will facilitate the building on that technology by other firms with similar trajectories. Firms with similar technological experience (i.e., absorptive capacity) can more easily value and utilize similar technologies (Cohen and Levinthal, 1990). Accordingly, we hypothesize that innovations exploring existing technologies will be positively related to more subsequent innovations. In this sense, evolutionary economics argues that the knowledge base of a firm and its routines reinforce and limit the rate of integration of external knowledge or production of radically new knowledge (Coombs and Hull, 1998). However, empirical evidence suggests too little of either exploration or exploitation reduces performance (Levinthal and March, 1993; Katila and Ahuja, 1002; Fagiolo and Dosi, 2003; He and Wong, 2004). In other words, combinations of knowledge elements (technological 11

12 and market-based) that exploit while exploring result in stronger innovations than incremental extensions of existing knowledge or complete exploration of new knowledge. Thus, leveraging knowledge (either technological or market-based) will result in more impact than remaining on a technological trajectory. Pure exploration offers the possibility of radical innovation through the combination of knowledge elements that are new to the firm. However, pure exploration might also have higher risks in diffusion, and might even be a defensive move after the firm s movement down a technological trajectory or customer preference trajectory has slowed. We will investigate the depth of impact of pure exploration, but do not hypothesize about its effect relative to other types of innovation. Hypothesis 1a: Cross-application has a more positive effect on depth of impact than does progression on a technological trajectory. Hypothesis 1b: Progression on a customer preference trajectory has a more positive effect on depth of impact than does progression on a technological trajectory. Breadth The value of deliberately redeploying existing technology in new domains might trump other paths of innovation in creating greater breadth of impact. Abernathy and Clark (1985) find that old technologies can be used to create new market niches. Cattani (2005, 2006) finds that firms can be preadapted and have knowledge without anticipated use that can lead to higher levels of performance as long as this knowledge is consistently leveraged into new markets. By focusing on different customers needs and market requirements, innovation can emerge from redeploying existing knowledge into a new selection environment, i.e. technological speciation (Levinthal, 1998). 12

13 In other words, we propose that exploration of combinations of knowledge elements (technological and market-based) result in more widely cited innovations than incremental extensions (exploitation) of existing knowledge. Generating new technological knowledge (revealed through a customer preference trajectory or exploration) will result in inventions that will be more broadly cited than remaining on a technological trajectory, but leveraging existing technology into additional product markets will lead most strongly to broad impact. Hypothesis 2a: Pure Exploration has a more positive effect on breadth of impact than does progression on a technological trajectory. Hypothesis 2b: Progression on a customer preference trajectory has a more positive effect on breadth of impact than does progression on a technological trajectory. Hypothesis 2c: Cross-application has a more positive effect on breadth of impact than does either progression on a customer preference trajectory or pure exploration. Real-time Tradeoffs Entrepreneurial, technology-intensive firms are constrained, in their finances and their human resources (Aldrich, 1991) and/or in development time (Bourgeois and Eisenhardt, 1988). This third set of hypotheses covers the tradeoffs in real time between different paths. A strong technological trajectory relates well to simultaneous pursuit of leveraging knowledge along one dimension as the underlying technological knowledge will be a firm competence and such competences are argued to be the basis for diversification (cite). However, trying to leverage technology along different dimensions in different projects can be difficult to manage due to bounded rationality of managers (March and Simon, 1958) and the contextual nature of innovation (Tyre and Von Hipple, 1997). So, firms will not be likely to pursue both cross-application and customer preference trajectory at the same time. On the other hand, once 13

14 you have leveraged knowledge along one dimension, it might be easier to add the second dimension due to the development of accompanying routines. Thus, cross-application or customer preference trajectory can be successfully pursued at the same time as exploration. Hypothesis3a: At any given time, cross-application is negatively related to customer preference trajectory. Hypothesis 3b: At any given time, technological trajectory is positively related to customer preference trajectory and cross-application. Hypothesis 3c: At any given time, cross-application and customer preference trajectory are positively related to exploration. Ambidexterity It is possible that any of the four types of innovative paths could be optimal given a particular competitive context. Thus, these methods may be complementary over time, contributing to a punctuated equilibrium of firm evolution (Burgelman, 2002). Entry into an additional market may provide the entrepreneur with feedback from customers or a basis for economies of scale in production that can promote more rapid movement along the technological trajectory. The type of complementary knowledge that enables a firm to respond to fast-moving and complex consumer preference trajectories in its initial market may also contribute to an ability to enter different markets. However, in general, we would expect that extending knowledge brings in revenues and opens up opportunities for future growth that does not cost as much as growing based on acquiring new knowledge. We can also draw a link between the corporate strategy literature on related diversification and the technology management literature on the benefits of exploration versus exploitation. Thus far, our analysis has considered firms to be constrained in resources so 14

15 that they must choose between paths. However, as firms grow, they may well develop ambidexterity the capability to explore new technology while exploiting current technology. We propose that, at the micro level, what related diversification really involves is not just having multiple technologies or being in multiple markets, but being able to leverage knowledge while simultaneously continuing to invest in that knowledge. In terms of technological knowledge, that means pursuing a technological trajectory while also engaging in cross-application. In terms of market knowledge, ambidexterity means pursuing a technological trajectory while also engaging in customer preference trajectory. Finally, ambidexterity across knowledge domains could result in exploration (Tushman and O Reily, 1996). However, our conception of relatedness implies exploration is more valuable when the firm continues to invest in the technological trajectory at the same time. Hypothesis 4a: Ambidexterity in technological knowledge has a more positive relationship to firm financial performance than does solely exploiting or solely exploring technological knowledge. Hypothesis 4b: Ambidexterity in market knowledge has a more positive relationship to firm financial performance than does solely exploiting or solely exploring market knowledge. Hypothesis 4c: Exploration (the simultaneous pursuit of new technological knowledge and new market knowledge) has a more positive relationship to firm financial performance when accompanied by pursuit of a technological trajectory. (Technological trajectory positively moderates the relationship of exploration to financial performance.) Experience The final set of hypotheses considers the effect of prior experience on subsequent behavior. Experience with a particular type of invention will develop capabilities, routines, or strategic 15

16 positioning that suggests firms will benefit from continuing to pursue that path. Experience at leveraging one type of knowledge does not help with leveraging the other type of knowledge; rather, previous cross-application reduces the instance of customer preference trajectory, and vice-versa. Similarly, pursuing a technological trajectory implies either opportunity or capability that precludes much of an attempt at leveraging knowledge. However, moving into new applications provides a basis and reason for continued development of the technology. Crossapplication can help pay for continued R&D on the technology that is now central to more than one revenue stream, and helps the firm that is on a technological trajectory to realize there may be more applications as they continue to improve the technology. Exploration involves acquiring or developing a new technology which will probably require continued development. However, implementing a new technology specifically to help in a customer preference trajectory will likely not lead to future investment in a technological trajectory, because the focus is on finding a technology good enough to satisfy customers now, and the firm will prefer to continue to focus on those customer needs rather than trying to become the best at the technology. Hypothesis 5a: Experience with one type of invention (e.g. cross-application) is positively related to subsequent occurrence of the same type. Hypothesis 5b: Experience with cross-application is negatively related to subsequent occurrence of customer preference trajectory, and vice versa. Hypothesis 5c: Experience with cross-application is positively related to subsequent occurrence of technological trajectory. Hypothesis 5d: Experience with exploration is also positively related to subsequent occurrence of technological trajectory. 16

17 Methodology We create a sample of firms in a research-intensive industry with many different applications of the basic technology. Biotechnology firms are 1) engaged in entrepreneurial activity; 2) use property rights for protection of innovation, especially patents; 3) have many opportunities to leverage their technological knowledge; 4) are active in the market for knowledge (Arora, Fosfuri and Gambardella, 2002) and thus frequently engage in licensing; and 5) are an important industry to the U.S. economy. Our study requires some data on financial performance and growth, so we select all public firms that ever reported an operating segment in SIC 2836 (Biological Products, Except Diagnostic Substances) from 1978 to Eliminating American Deposit Receipts (ADRs), there are 262 public firms in this industry. We track these firms for the full range of years, even years in which they do not report a segment in 2836, until they exit the sample through bankruptcy or acquisition. Additional data on firm subsidiaries, acquisitions, divestitures, and IPOs come from the SDC Mergers & Acquisitions database, Bloomberg, and the Directory of Corporate Affiliations. We use the corporate information to generate a list of all patents awarded to the firms, even prior to going public, as found in the Delphion patent database. Many biotechnology firms do not apply for patents, so we cannot generate our primary variables as described below. Thus, there are 121 patenting firms in the final sample, with 4,198 patents. For models requiring the patents to be submitted in years that the firm is public (e.g., using firm size as a control variable), the sample is 100 firms with 2,226 patents. To test our hypotheses, we identify applications in biotech through content analysis of the exemplary claims section in patent documents. Extensive previous research in strategic management and technology management has used information from patents regarding patent class and subclasses, citations to prior art, reference to scientific journal articles, identity of the 17

18 inventor, geographic location of the assignee, and other data. However, the longer text fields of patents also contain relevant information that has yet to be exploited in large-scale studies. Indicators of the intended use of an invention have been drawn from patent classification or new product introductions. However, the phenomenon of cross-application implies that multiple inventions in the same class may be targeted at different product markets. We develop a finergrained view of biotechnology applications through content analysis, and vary the procedure in robustness checks. From the full text of the exemplary claims, we use WordStat software to identify keywords. We run a frequency of all words in the text, and select ones that we believe indicate the intended use of the invention. In addition to these 78 words, we also retain any words that appear at least 25, but no more than 250 times, and in no more than 25% of the cases. These parameters are selected to generate a list of keywords that can differentiate the patents into categories. With thousands of patents to analyze, the WordStat clustering algorithm is only able to incorporate approximately 100 keywords. 2 For the reported results, we clustered the patents into 100 categories based on the frequency and co-location of these keywords. These categories form our measure of an application. A patent is in a new application for the firm if no previous patent has been in the given category. Similarly, we use patent data to develop a measure of whether a patent represents new technology for the firm. We base the technology distinction on the backward citations of the patent. US patent law requires the applicant to specify any prior art in terms of previously awarded patents. We suggest an invention represents new knowledge if it cites a different set of precursors than those cited in previous patents from the same firm. So, we determine whether the backward citations to prior art of a patent have previously been cited in the firm s earlier patents. 2 Words examples: cholesterol, lung, oral, bovine, cartilage, rheumatoid, hormone, histidine 18

19 For the reported results, we select a cutoff value of 50%, such that if a patent has at least half of its backward citations new to the firm, we consider the patent to be new technology. Dependent Variables The measures of new application and new technology allow us to generate indicator variables for cross-application (new application, old technology), technological trajectory (old application, old technology), customer preference trajectory (old application, new technology), and exploration (new application, new technology). As suggested by the above hypotheses, these indicators are used as dependent variables and independent variables in different models. Firm performance is measured as the market-to-book ratio (MTB) and the current year return on assets (ROA). Our primary measure is an indicator of whether MTB is greater than or equal to 5, which represents approximately the upper quartile of the distribution, which is highly skewed (see Figure 2). Innovation outcomes are measured in terms of depth and breadth of impact on subsequent R&D. The depth of impact is the count of forward citations received by the patent, exclusive of self-citations, consistent with prior research (e.g., Rosenkopf and Nerkar, 2001). The breadth of impact is the adjusted generality measure. Originally defined by Hall, Jaffe and Trajtenberg (2001) and utilized by Jaffe, Fogarty and Banks (1997) and Henderson, Jaffe and Trajtenberg (1995), this measure uses the three-digit U.S. patent classes of the patent s forward citations as follows: G ni 2 1 sij = j where sij denotes the percentage of patent forward citations to the patent i that belong to patent class j out of ni patent classes (Note: the sum is the Herfindahl concentration index). Thus, 19

20 if a patent is cited by subsequent patents that belong to a wide range of fields, the measure will be high (i.e., close to one), whereas if most citations are concentrated in a few fields it will be low (i.e., close to zero). This measure tends to be correlated with the number of patent citations of the patent, so we control for number of patent citations as follows: γ i = Ni Ni G 1 i where Ni is the number of forward patent citations for the patent i and Gi is the value of diversity for patent i. Control Variables Biostate indicates whether the firm has its headquarters in Massachusetts or California. Size is the log of total assets. This variable is only available when the firm is public. Age is the length of time from the year of patent application filing until This variable controls for the fact that some patents have had more years to be cited than others. R&D Expenditure is the current year expense for research and development, as reported in Compustat. Missing data is assumed to equal zero. R&D Intensity is R&D over log of assets. Year dummies control for differences across years in the panel regressions with firm performance as the dependent variable. Estimation For the models explaining Depth of Impact, the dependent variable is a count variable which frequently equals zero. We implement a Zero-Inflated Negative Binomial estimation, which yields coefficients for the independent variables on the dependent variable and on the likelihood that the dependent variable equals zero (the inflation coefficients ). 20

21 Results Analyses are conducted at both the patent level and the firm-year level. Table 3 records the results of models testing the effect of different paths of technology evolution on the depth of impact on subsequent innovation. According to Hypothesis 1a-b, we expect cross-application and customer preference trajectory to have a more positive effect on depth of impact than will technological trajectory. Since our categories are exhaustive and mutually exclusive, there are multiple ways to split the data. Model I includes only the control variables and the indicators for New Application and New Technology in a zero-inflated negative binomial model. Introducing a new technology has a statistically significant, positive effect on forward citation count. Entering a new application also has a positive coefficient, but at a marginally significant level for the main coefficient (p-value=0.069). New application does have a statistically significant effect of reducing the likelihood of a zero value for forward citations (in the logit inflation model). Model II adds an interaction term to demonstrate that pursuing simultaneous new application and new technology, detracts (if anything) from the depth of impact of an invention. However, the effect of exploration is the sum of the interaction and the direct effects, so overall exploration has a stronger positive effect than the excluded other paths. The coefficient on new application becomes statistically significant in the main model, while the coefficient on the interaction term has a p-value of In Models III and IV, we vary the indicator variables and their interaction to make it easier to see the effects of cross-application and customer preference trajectory. Crossapplication is the interaction between new application and old technology. The results suggest that pursuing a new application reduces the likelihood of not receiving any forward citations, but only has a positive direct effect on the number of forward citations when it is combined with old 21

22 technology. Similarly, customer preference trajectory is the interaction between old application and new technology. Having an old application actually decreases the likelihood of a zero in the dependent variable, and increases the count of non-zero observations. Combining new technology and old application has a positive, but non-significant coefficient (p=0.053). New technology in and of itself has no effect on forward citations. A straightforward test of Hypothesis 1 may be best seen in Model V, which reveals that both paths have a positive and significant effect on depth of impact, as compared to the excluded path of technological trajectory. Note that exploration also has a positive effect, and the three coefficients are not different from one another, statistically. The zero-inflated negative binomial estimation computes standard errors with a recognition that some firms have multiple patents (the clustering option in Stata). For robustness, we also estimate the models using a full panel approach, negative binomial with random effects (no inflation model). Model VI repeats the variables in Model V with this panel specification, agreeing with previous findings. In sum, results are consistent with Hypothesis 1a and 1b. In Hypothesis 2, we surmised that cross-application, customer preference trajectory, and exploration would have stronger positive effects on the breadth of impact on subsequent innovation than would technological trajectory and that the strongest effect would be for crossapplication (because it specifically demonstrates how one technology applies to multiple applications). Table 3 summarizes the relevant results, with Models I-V defined as in the previous table. The sample size is smaller because the dependent variable can only be computed when there is at least one forward citation. Note also that the pseudo R-squared is very small, given the fixed effects estimation, although there are numerous significant partial correlations. Model I shows no statistically significant coefficients. Further, including the interaction term in 22

23 Model II reveals that any benefits of new application are only present and are offset by the negative effect of pursuing both types of newness at the same time. In other words, full exploration does not add to the direct benefit of new technology. 3 Similar to the models explaining depth of impact, Models III and IV reveal some benefit of cross-application and customer preference trajectory, but at borderline statistical significance at best. However, Model V shows that each of the three included paths have a stronger positive effect on breadth of impact than the excluded path of technological trajectory. Thus, Hypotheses 2a and 2b find support. The coefficients for exploration and cross-application are nearly identical, consistent with the idea from Model 1 that only new technology matters. Contrary to Hypothesis 2c, although its coefficient is highest (0.130) the effect of cross-application is not strongest, as the three coefficients are indistinguishable, statistically. Hypothesis 3 considered each path of technology evolution as a dependent variable. At any given time, firms (particularly small ones like the majority of our sample) are likely to only pursue one path of technological evolution. We test this set of hypotheses by estimating relationships between the count of each path in a given firm-year, explained by its lagged value and instrumentalized, contemporary values of the alternate paths. Hypothesis 3 focused on whether firms pursue multiple paths at the same time. Hypothesis 3a argued that cross-application and customer preference trajectory would be difficult to pursue together in the short-term. As shown in Table 5, there is a negative coefficient on customer preference trajectory in the model explaining cross-application, and vice-versa. Having more inventions of one type decreases the number of inventions of the other type in the same year, controlling for endogeneity (e.g., because firms can only file a limited number of 3 Recall that the breadth measure is already adjusted for the number of forward citations, so this is not simply repeating the earlier results. 23

24 patents in a given year, given their R&D budget). In Hypothesis 3b, we stated that there should be complementarity between pure exploitation and movement to leverage knowledge along one dimension at a time, i.e. cross-application or customer preference trajectory. As predicted, positive relationships exist in both directions between technological trajectory and crossapplication, and between technological trajectory and customer preference trajectory. Similarly, Hypothesis 3c considered complementarity between leveraging knowledge along one dimension and moving further into full exploration. Both cross-application and customer preference trajectory have positive and significant coefficients in the model explaining exploration. Exploration has a positive effect on the occurrence of customer preference trajectory, but no significant effect in the model explaining cross-application. An alternate test of Hypothesis 3 uses an even shorter time window patents from the sample are actually filed on the same day as at least one other patent from the same firm. We create a measure indicating whether there is any observation (one or more) of each type of invention on a specific date. The pairwise correlations between these dummy variables yield results only partially similar to the regression results in Table 5. The correlation between crossapplication and customer preference trajectory is insignificant, inconsistent with H3a. However, as stated in H3b, there are positive correlations between technological trajectory and both crossapplication (ρ=0.24, p-value<0.001) and customer preference trajectory (ρ=0.17, p- value<0.001). Consistent with H3c, exploration is positively correlated with customer preference trajectory ((ρ=0.28, p-value<0.001). In contrast, the correlation between exploration and crossapplication is negative for patents filed on the same date (ρ=-0.06, p-value<0.05). 4 Patents filed on the same day may represent inventions that were not simultaneously discovered and evaluated, but we assume the development and implementation of the technology is occurring concurrently. 24

25 In Hypothesis 4, we argue that leveraging knowledge will lead to improved short-term returns, but ambidexterity in both exploiting and exploring knowledge (along a single dimension) or in exploring along both dimensions will create greater long-term value. We are able to reject the null hypothesis that ambidexterity is not significant. Table 6 reports results from models with various dependent variables that reflect the financial performance of the firm. The dependent variable is whether the firm achieves a high market-to-book ratio, an indication of its expected future performance or growth prospects. The coefficients for the interactions are positive in Models I to III, and the coefficients on the direct effects are negative. So, in Model I, ambidexterity in technological knowledge is better than pursuing only exploration or exploitation of technological knowledge. In Model II, ambidexterity in market knowledge is better than pursuing only exploration or exploitation in that dimension. In Model III, pursuing exploration across both dimensions has a higher correlation with achieving high financial performance than does pursuing exploration along only one axis at a time. To more specifically test Hypothesis 4c, we interact technological trajectory and exploration in Model IV, finding that exploration itself has a negative effect on financial performance, and technological trajectory has a negligible effect, but the combination has a positive effect. All of these results are consistent with Hypotheses 4a-c. For comparison, customer preference trajectory does not add to performance beyond the direct effect of staying in an old application (in Model V, the positive interaction effect is offset by the negative direct effect of new technology). Likewise, in Model VI, crossapplication does not correlate with high financial performance. When using ROA as a dependent variable, cross-application does show a positive, significant correlation with financial performance, as shown in Table 6, Model VII. No other type of invention or ambidexterity has a significant correlation with ROA (results not reported in the table). 25

26 Finally, we argued that there will be path dependence such that experience with one path over time will enable continued pursuit of that path. We estimate logit models explaining the type of invention for each patent, using as independent variables the firm s past experience with all paths, with summarized in Table 7. In Hypothesis 5a, we stated experience with each type of invention should be positively correlated with subsequent instances of the same type. The results in Table 7 find positive, statistically significant relationships between experience and subsequent occurrence for all types except exploration. Hypothesis 5b considered how experience with one type might preclude innovation of another type: specifically, cross-application and customer preference trajectories as substitutes over time, and the issue of commitment to a technological trajectory precluding other paths. Indeed, the coefficient for number of past innovations that were cross-application has a strong negative coefficient for the model explaining occurrence of customer preference trajectory. However, the reverse relationship is not significant. Hypothesis 5c considered potential complementarity: specifically, that cross-application could lead to further development of a particular technology, which could encourage the firm to continue to pursue more inventions along the technological trajectory, and that exploration might necessitate further pursuit of the technological trajectory for these new technologies. As expected, there is a positive relationship between experience with cross-application and subsequent inventions focusing on development of a technological trajectory. Regarding Hypothesis 5d, the coefficient for exploration is positive and significant in the model explaining technological trajectory. It is worth noting at this point that any experience with other types of inventions is negatively related to exploration. This suggests exploration is the province of new firms or firms that do not frequently patent, possibilities we will examine in the discussion. 26

27 Discussion and Conclusions This paper provides an integrated view that combines a firm s earlier decisions with its later decisions and relates the path to firm outcomes. This paper suggests that an innovation strategy is not just a point-of-time decision, but the outcome of a series of decision points that lead to the current situation. In this context, path dependence means that a firm s previous investments and its repertoire of accumulated routines constrain its future behavior (Teece et al., 1997). From an evolutionary point of view (Nelson and Winter, 1982), path dependence is a necessary condition for resource accumulation and survival. In particular, it seems that selection on innovation performance outcomes alone does not lead to a specific path but that real time tradeoffs and increasing experience lead to a firm s choice to grow along a particular competence, i.e. from exploiting to exploring. When a firm can both explore and exploit to exhibit true ambidexterity along a competence, then it will be rewarded by financial performance. Pure exploration seems a domain for new firms that have not yet developed competences. This paper has implications for the corporate strategy literature as diversification studies have rarely attempted to distinguish between firms that moved into a product market through exploration of technologies new to the firm versus diversifying on the basis of leveraging existing technology to new markets. The distinction is important because the two paths to corporate diversification seem to be mutually exclusive in real time and throughout the firm s accumulated experience. These two paths imply different competencies, decision-making processes, timing of entry, and risks. Firms may employ multiple technologies, enabling them to create products with distinctive features, understand customer needs, or interact with suppliers in a single or multiple product market. The firm itself may not be pursuing development of one of 27