Dissertation. Ting Xiao, M.A., M.S. Graduate Program in Business Administration. The Ohio State University. July Dissertation Committee:

Transcription

1 Highlighting the Role of Knowledge Linkages in Knowledge Recombination Dissertation Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Ting Xiao, M.A., M.S. Graduate Program in Business Administration The Ohio State University July 2015 Dissertation Committee: Jaideep Anand, Co-Advisor Mona Makhija, Co-Advisor Ashton Hawk Richard Bettis

2 Copyrighted by Ting Xiao July 2015

3 Abstract Innovation comes about through manipulation of knowledge components as well as the linkages of these components by firms. Prior research has concentrated primarily on how adding of new knowledge components to those already known to the firm gives rise to innovation, but little attention has been devoted to the role of linkages in this regard. Thus, the literature lacks a systematic explanation for how variations in linkages between components influence innovation outcomes. The purpose of this dissertation is to help address this problem. In this research, I conceptualize knowledge linkages as a reflection of the organizational routines that motivate the integration of individual knowledge components, and in doing so, enrich their joint meaning and function. That is, disparate and organizationally dispersed knowledge components can only be linked through the institution of appropriate routines. Drawing on this argument of linkages as reflections of routines, I highlight how changes in two key characteristics of linkages extent and local embeddedness influence subsequent innovation novelty and use in three differing types of knowledge recombination that involve, from a previous innovation to a subsequent one (a) no change in components, (b) adding of new knowledge components, and (c) subtracting of components. I hypothesize that, in each case, change in linkages will play a major and systematic role in influencing the novelty and use of the resulting innovation. I use detailed pharmaceutical industry patent data from 1976 to 2014 to empirically test the hypothesized relationships. The findings demonstrate that linkages play an essential role in innovation, largely in the manner ii

4 hypothesized. In doing so, this research uncovers a key feature of knowledge recombination that furthers our understanding of how innovation actually comes about and is utilized by firms. iii

5 Acknowledgments The entire dissertation work has significantly benefited from the guidance of my graduation committee members, Dr. Jaideep Anand (co-chair), Dr. Mona Makhiia (co-chair), Dr. Ashton Hawk, and Dr. Richard Bettis. They have devoted significant amounts of time working with me to help push forward this dissertation. In particular, Dr. Mona Makhija even sacrifices her weekends to help me with this dissertation. I am greatly indebt for her invaluable dedication. I want to thank Dr. David Teece (University of California, Berkeley), Dr. Lee Fleming (University of California, Berkeley), and Dr. Jay Barney (University of Utah) for devoting their precious time to provide me guidance and encouragement. I also want to thank all my colleagues and friends, who have provided me valuable supports and suggestions. Finally, I want to thank Dr. Han Jiang (University of Arizona), my best friend, who constantly challenges every potential loophole in this study and pushes me to polish the dissertation. iv

6 Vita 2007 B.S. Computer Science, Xi an Jiaotong University, Xi an, China B.S. Mathematics, Xi an Jiaotong University, Xi an, China 2010 M.A. Management, Xi an Jiaotong University, Xi an, China 2010 to present M.A. Economics, Department of Economics, The Ohio State University Fields of Study Major Field: Business Administration Minor: Statistics v

7 Table of Contents Abstract ii Acknowledgement iv Vita v Fields of Study v Table of Contents vi List of Tables..... vii List of Figures viii Chapter 1: Introduction Chapter 2: Review of the Literature on Knowledge Recombination and Innovation..10 Chapter 3: Isolating the Effects of Changed Linkages on Innovation Outcomes by Holding Knowledge Components Constant Chapter 4: The Interactive Effects of Changed Linkages and Added Knowledge Components on Innovation Outcomes Chapter 5: The Effects of Changed Linkages in Subtracted Knowledge Components on Innovation Outcomes Conclusions and Future Research Directions References vi

8 List of Tables Table 3.1. Number of Paired Patents for a Sample of Pharmaceutical Firms..61 Table 3.2. Variations in Names by which Hoechst s Patents are Filed Table 3.3. Descriptive Statistics Table 3.4. Relationship between Linkage Change and Innovation Novelty Table 3.5. Relationship between Linkage Change and External Use...66 Table 3.6. Relationships between Linkage Change and Internal Versus External Use...67 Table 4.1. Number of Paired Patents for Firms in the Sample Table 4.2. Descriptive Statistics Table 4.3. Relationships among Added Components, Changed routines and Innovation Use Table 4.4. The Difference between Internal and External Use Table 5.1. Number of Paired Patents for Firms in the Sample Table 5.2. Descriptive Statistics Table 5.3. Relationships among Subtracted Components, Routines and Novelty Table 5.4. Relationships among Subtracted Knowledge Components, Routines and Internal versus External Use vii

9 List of Figures Figure 3.1. Hypothesized Relationships between Change in Architectural Characteristics and Innovation Outcomes Figure 3.2. An Example of Knowledge Linkage Change viii

10 Chapter 1: Introduction It was in 1939 that Schumpeter first suggested that innovation combines components in a new way, or that it consists in carrying out new combinations (p.88). Since then, knowledge recombination has developed into a useful and important perspective for studying innovation. It has been widely acknowledged that knowledge components are the basic building blocks of an innovation (Fleming, 2001; Rosenkopf and Nerkar, 2001). In this regard, Fleming (2001) argues that the choices of knowledge components come from social construction or previous association (p.118), while Nelson and Winter note, the creation of any sort of novelty in art, science or practical life consists to a substantial exert of a recombination of conceptual and physical materials that were previous in existence (p.130). It stands to reason, then, that an innovation can be achieved through change in knowledge components (Fleming, 2001; Katila and Ahuja, 2002; Carnabuci and Operti, 2013), as well as by linking existing and new components in different ways (Henderson and Clark, 1990). The arguments above imply that knowledge recombination can be accomplished in multiple ways, including adding of new knowledge components to existing ones, subtracting of existing knowledge components from others, and holding knowledge components constant but changing the ways in which they are linked with each other. Nonetheless, prior literature lacks a comprehensive understanding of the mechanisms by which different forms of knowledge recombination influence 1

11 innovation outcomes. Specifically, the role of changed knowledge linkages has not been adequately considered in studies on knowledge recombination. Knowledge linkages of an innovation are the structural pattern of connections through which all knowledge components are associated and function with each other to deliver a welldefined function. The majority of the recombination literature has focused solely on understanding the role of knowledge components rather than the linkages that integrate them. However, as, Carnabuci and Operti (2013) point out, these do not exist in a vacuum the manner in which components are linked plays a fundamental role in determining their meaning. Indeed, in their path-breaking research, Henderson and Clark (1990) show how significantly changed linkages between knowledge components give rise to new innovations that other firms are unable to copy. However, change in knowledge linkages can occur within a continuum between the two extremes of no change and complete change. We lack a systematic explanation for how variations in changed linkages among knowledge components systematically influence innovation novelty and value. Even so, in the decades following Henderson and Clark s (1990) study, no research has attempted a deeper examination of the role played by linkages in the innovation process. The purpose of this research is to address this gap. In particular, three essays systematically develop the critical role of linkages in the knowledge recombination process. These essays are discussed below in turn. 2

12 Essay 1 Isolating the Effects of Changed Linkages on Innovation Outcomes: Holding Knowledge Components Constant in Knowledge Recombination In this first essay, our objective is to isolate the effects of knowledge linkages among components on innovation outcomes. To accomplish this, we hold knowledge components constant in a pair of prior and subsequent innovations and consider only the changes in linkages. We consider (a) how change in linkages alone affects innovation novelty, and (b) the effect of these changed linkages on the relative abilities of the innovating firm and other non-innovating firms in using the resulting innovation. In doing so, we highlight change in two key features of knowledge linkages extent (the overall connectedness of knowledge components) and local embeddedness (the degree to which a focal linkage operates with other closely intertwined linkages). We expect that (a) greater change in these features of routines enhances the novelty of a firm s innovation by creating new insights from altered combinations of existing knowledge components, (b) difficulty in appreciating the changed routines impedes external firms ability to effectively use the innovation, and (c) greater change in routines affords a comparative advantage to the innovating firm over external firms in using the innovation. Six hypotheses are developed in Essay 1 in accordance with these relationships. The hypotheses are tested using patent data for 33 of the largest firms in the pharmaceutical industry between 1976 and We identify 4,014 pairs of prior and subsequent patents from a single firm that have the same exact knowledge subclasses for Essay 1; We constructed a network for each of these patents based on the pairwise coupling relationships or ties between every two components within the patent, with subclasses serving as nodes and coupling relationships as ties (Yayavaram and Ahuja, 3

13 2008). To measure the strength of a linkage, we adapt Fleming and Sorenson s (2004) method for measuring coupling to account for the history of how the two components have been combined with each other in all of the firm s previous patents. Knowledge components exhibiting closer interdependent relationships over time are reasoned to have stronger underlying routines. This dynamic approach, which takes into account ongoing changes in the linkages associated with their innovations, allows us to capture the evolutionary process by which firms develop their sense-making routines. We identify changes in extent and local embeddedness of linkages for each pair of patents. These data form the basis for testing the hypotheses. Indeed, our findings largely support the hypothesized relationships. After establishing the relevance of changed linkages for innovation in Essay 1, we consider the interactive effects of changed components and linkages in Essays 2 and 3. Essay 2 The Interactive Effects of Changed Linkages and Added Knowledge Components on Innovation Outcomes The value of adding new knowledge components to those already known to the firm has already been strongly stressed in the recombination literature, due to the ability of new elements to refresh the meanings and functions of the firm s existing knowledge, and introduce novel insights that lead to innovation (Fleming, 2001; Galunic and Rodan, 1996; Katila and Ahuja, 2002). Nonetheless, a subtle but critical fact is that the addition of new knowledge components will also alter the structure of knowledge linkages, i.e., the way in which the new and existing components are connected with each other (Henderson and Clark, 1990). However, prior studies 4

14 examining the addition of new knowledge components in recombination have not accounted for the role of changed linkages in this regard. Ignoring the effects of changed linkages associated with newly added knowledge reduces understanding of how firms accomplish the integration of differing knowledge components associated with an innovation. The goal of Essay 2 is to address this problem. We investigate how change in linkages interplays with the addition of knowledge components in determining the ability of firms, including the innovating firm as well as firms that did not come up with the innovation, to institute and use the innovation. Treating linkages as a reflection of requisite routines, we continue to focus on the two central features overall extent and local embeddedness that capture both general and local contexts of routines. Since a change in knowledge components is relatively explicit in a patent, and thus identifiable, we predict that both innovating and non-innovating firms are equally capable of utilizing an innovation created from changed knowledge components. In contrast, change in knowledge linkages and their corresponding routines is much more difficult to identify or understand due to their tacit nature. The need to change already established routines is also problematic (Teece, 1982; Feldman and Pentland, 2003). Thus, change in knowledge linkages is argued to create difficulties for both innovating and noninnovating firms in utilizing the new innovation. Nonetheless, we expect that changed routines will create comparatively more problems for non-innovating firms, as they will find the diagnosis of appropriate routines particularly difficult in comparison to the innovating firm that had developed superior understanding in this regard from the process of creating the innovation. Six sets of testable hypotheses are developed in line with these predictions. 5

15 To test the proposed hypotheses, 6,707 pairs of pharmaceutical industry patents were identified in which a subsequent patent is derived by adding knowledge components to those of a prior patent. Other constructs were measured in a manner similar to those in Essay 1. Based on the findings using these data, the predictions above are largely upheld. Essay 3 The Effects of Changed Linkages in Subtracted Knowledge Components on Innovation Outcomes While the interactive effects of adding knowledge components and changed linkages were studied in Essay 2, the effects of changed linkages stemming from subtracted knowledge components on innovation outcomes are considered in Essay 3. Note that the subtracting of knowledge components is not simply the opposite of adding knowledge components in terms of knowledge recombination it has its own unique features. A key distinction between the addition of new components and the subtraction of existing components is the different means by which novelty is created. In the case of adding components, novelty is created through the fresh insights gained from new knowledge. Enlarging the pool of knowledge pieces rejuvenates the functioning of the entire set of components. Since new knowledge components are clearly associated with new ideas and insights, the rationale for adding new knowledge components is relatively obvious. The benefits derived from subtracting knowledge components is not as easy to figure out, especially for outsiders that did not come up with the innovation. In the case of subtracting knowledge components, novelty is created through alteration in the way in which the remaining knowledge components are coordinated 6

16 with each other. That is, rather than incorporating external knowledge to create a new innovation, subtraction requires a reworked set of connections among the rest of the components, In this way, as noted earlier, the value of subtracting knowledge components lies in a new logic for connecting the reduced set of components. The aim here is to extract a hidden meaning underlying the subset of knowledge components. In contrast to the adding of new knowledge components, in which the new components play as important a role as the new linkages that integrate them with existing components, for knowledge subtraction the role of changed linkages plays a particularly critical role in creating novelty. That is, once components are subtracted, changed linkages among remaining components are a key source of novelty of the resulting innovation. The new meanings afforded by the changed linkages among established components are vital in bringing about innovation. Therefore, subtraction of knowledge components indeed deserves a significant attention in knowledge recombination studies. Interestingly, however, there has been virtually no prior work on this issue. To address this void, Essay 3 explores the mechanism and influence of subtracting knowledge components on innovation novelty and use. We argue that, the more components that are extracted, the stronger the positive relationship between change in both extent and local embeddedness of linkages and novelty of the newly created innovation. Since the way through which knowledge components are connected represents how organizational members conceptualize, interpret, make sense and reason the relationships among components (Dosi, Nelson and Winter, 2000), and involves a path-dependence nature (Nelson and Winter), we expect change in linkages to be hard to spot and accomplish, resulting in difficulty in using the resulting innovation. However, the understanding and experience accumulated in the 7

17 creation process helps the innovating firm to cope with this difficulty. As a result, the innovating firm is expected to have a relative advantage in using the innovation compared to other firms that do not directly participate the innovation creation in this regard. Four sets of hypotheses are proposed in this regard. To test these hypotheses, we identified 7,476 pairs of patents in the pharmaceutical industry in which a subsequent patent is created by subtracting knowledge components used in a prior patent. All other constructs are measured in a manner similar to Essay 1. Again, the findings confirm the expected theoretical relationships. Contributions of the Research The three studies contribute to the knowledge recombination literature in a number of ways. First of all, we address an important gap in this literature the role of knowledge linkages. Our treatment of knowledge linkages as routines provides a theoretical rationale for how linkages among knowledge components influence the combined conceptualization of knowledge components and give rise to larger meaning. This is derived from our expectation that knowledge components are not combined in random or varying ways to give rise to the same innovation outcomes. Instead, the manner in which knowledge components are combined is purposeful and intentional. At the same time, due to their diverse and dispersed nature, individual knowledge components are unlikely to be recombined in the desired manner on their own. A purposeful recombination requires accurate routines to ensure that it materializes. This research demonstrates that, above and beyond change in knowledge components that has been the primary focus of prior literature, changing the routines 8

18 that combine these components alone can bring about new and unique insights that allow for the creation of novel innovations. Routines are evolutionary, developing over time in a path-dependent manner (Nelson and Winter, 1982). This evolutionary attribute ensures that routines are difficult to establish. However, once established, they are also difficult to change. This research shows that linkages exhibit likewise evolutionary characteristics within a firm, relationships between knowledge components tend to persist over time. However, a change in these linkages and their associated routines elicits novel insights in relation to the knowledge components. This research is the first to account for the evolutionary nature of routines associated with innovation. By capturing the dynamic development of linkages between knowledge components in successive patents over time, we create a means to identify features of routines that have thus far been invisible to researchers. This study also considers the interactive effects of knowledge linkages and changed knowledge components, either through addition or subtraction, thus providing a more clear and complete picture of how knowledge recombination works. We demonstrate that, while the addition of new knowledge components allows both innovating and non-innovating firms to successfully use the resulting innovation due to their ability to identify this change, change in extent and non-decomposability of routines that link these new components detract from its successful use by firms. Changed linkages and their associated routines are difficult for firms to identify and understand due to their opaque nature, and thus, provide a comparative advantage to the innovating firm in using the innovation. This insight regarding knowledge linkages is new to the literature. 9

19 While the literature has focused heavily on the adding of knowledge components for recombination, this research has also considered the linkage effects of subtracting knowledge on innovation outcomes. In this case, we argue that novelty stemming from knowledge subtraction comes primarily from altered knowledge linkages among the reduced set of components. Indeed, we demonstrate that innovation can come about from reduction of components, but only through reworked connections among them. The focus on the knowledge subtraction is also new to the recombination literature. In all, this study not only distinguishes and examines different forms of knowledge recombination, but also differentiates among mechanisms and influences of knowledge linkages in these different types of recombination. In doing so, this research advances our understanding of knowledge recombination in new directions. 10

20 Chapter 2: Review of the Literature on Knowledge Recombination and Innovation In this section, we will discuss how knowledge recombination contributes to the understanding of innovation, and identify key assumptions associated with this literature. As early as 1939, Schumpeter has argued that, the making of invention and the carrying out of the corresponding innovation are, economically and sociologically, two entirely different things (Schumpeter, 1939: 88). Continuing in this vein, he notes that innovation combines components in a new way, or that it consists in carrying out new combinations. He described new innovation as derived from recognizing the value in the underlying parts of diverse systems and discerning that these parts can be recombined in a novel fashion. He stresses that, to produce means to combine materials and forces within our reach to produce other things means to combine these materials and forces differently. In line with these points from Schumpeter is Penrose s (1959) assertion that the services yielded by resources are a function of the way in which they are used exactly the same resources when used for different purposes or in different ways and in combination with different types or amounts of other resources provides a different service or set of services (Penrose, 1959: 25). By the same token, Nelson and Winter (1982) state that the creation of any sort of novelty in art, science or practical life consists to a substantial exert of a recombination of conceptual and physical materials that were previous in existence (Nelson and Winter, 1982: 30 11

21 bolding incerted). Note that such an approach is consistent with what many scholars have referred to as combinative capabilities (Kogut and Zander, 1992) or reconfiguring competence (Galunic and Rodan, 1998). For example, Kogut and Zander (1992) argue that firm depends on its combinative capability to be able to synthesize existing knowledge. Similarly, Carnabuci and Operti (2013) suggest a firm s innovations are driven by its recombinant capabilities. Drawing on these early classic works, we highlight three key assumptions that are important for the recombination process. The first is that recombination has an underlying purpose, and doesn t happen in a vacuum. Firms recombine knowledge to achieve different goals; it is the purpose of recombination that motivates firms to initiate the process in the first place. The second assumption that can be drawn from this work is that the same building blocks of firm resources and/or knowledge components have the potential to be recombined in alternative ways. Finally, the previous point suggests that the specific way or exact manner in this paper we refer to knowledge configuration in which components are combined or integrated is what yields their new services. Below we will discuss the effect of each of these three assumptions on innovation creation. The literature has well elaborated the first two assumptions and leaves the third assumption as a hole. This paper is going to fill it. Effects of Knowledge Components on Innovation Firms hope to find successful recombination by searching out and integrating different sets of components that are not familiar to other firms (Fleming, 2001). According to Mead and Carver (1980), the value of reuse and refinement of components has been long recognized Firms add new components to their ongoing innovation by introducing new components or adopting scientific findings to guide how to add new components (Fleming, 2001). It is for this reason that studies have 12

22 differentiated the knowledge components from old and new domains, and suggest that they have different impact on the knowledge recombination process. Fleming (2001) argues that knowledge components from old domains that are used more recently and frequently reflect greater knowledge and familiarity. Innovations that combine familiar components to deliver new innovations could also be more valuable stemming from the fact that the components selected are better understood and more likely to be appropriate. The performance of the innovation may also be more predictable due to the incorporation of familiar components. On the other hand, Rosenkopt and Nerkar (2001) stress the value of knowledge components from new domains, finding that most innovative technologies come from a recombination of diverse knowledge components that span technological categories and organizational boundaries. Similarly, Carnabuci and Operti (2013) indicate that the diversity of knowledge base increases the probability of radical innovation (Ahuja and Lampert, 2001). There are also studies simultaneously study both categories. Taylor and Greve (2006) find that novel combinations increase the variance of product performance and associated extensive experience produce output with high average performance. They confirm that incremental innovation will have a high mean performance but low significant inordinate profits or losses, and that radical innovation will offer high potential for extreme profits or losses but low expected mean owing to high likelihood of failure. A number of studies also notice that too-diverse knowledge can lead to unwieldy and impractical outputs and too-focused knowledge can result in competence trap in which organizations are locked in their old activities and cannot absorb new knowledge and information (Arthur, 1989; March and Simon, 1958; Taylor and Greve, 2006). In summary, the more diverse the knowledge and 13

23 information that are used, the more novelty the output has, while the deeper the use of existing knowledge, the less novelty characterizing the output. After reviewing how components influence innovation, we will consider how their characteristics and organizational features impact innovation. Effects of knowledge component characteristics and organization features on innovation creation Previous studies investigate the knowledge components characteristics together with organizational features in order to better illustrate the impact of knowledge components on innovation creation process. As suggested by Penrose (1959), not only do the components matter, the characteristics of components also influence innovation creation. In this regard, Galunic and Rodan (1998) study how characteristics of knowledge and its social organization in the firm impact resource recombination. Nonaka (1994) proposes four modes of knowledge creation taking into account the tacit and explicit dimensions of knowledge, including modes that take tacit knowledge to tacit knowledge, explicit to explicit knowledge, tacit to explicit knowledge, and explicit to tacit knowledge. Nonaka acknowledges the role of individuals as essential actors in organizational knowledge creation and explains the process by describing the interactions between team members in organizational project teams. Similarly, Sabherwal and Becerra-Fernandez (2005) develop a classification of types of specific knowledge (context-specific, technology-specific, and context- and-technology-specific) to study the knowledge recombination process. Katila (2002) investigate how two characteristics of knowledge, age of knowledge and its location (intra-industry or extra-industry), influence innovation. Gardner, Gino 14

24 and Staats (2012) study how three different types of resources which are relational, experiential and structural influence team s knowledge integration capability. Other research focuses on organizational factors from different levels to study how these factors influence knowledge recombination. Fleming et al. (2007) study the influence of the inventor s network cohesion, breadth of experience, change of employers and external ties on innovation creation. Gruber, Harhoff, and Hoisl (2013) investigate how individual-level characteristics of inventors such as their educational differences impact knowledge recombination. Smith (2006) studies how human resource practices affect the organizational social climate conditions and how these social climate conditions promote knowledge combination process. Guler and Nerkar (2011) examine how the intraorganizational network especially cohesion among members of an organization influence firm s innovation. Davis and Eisenhardt (2011) study how rotating leadership facilitates knowledge recombination and thus enhance collaborative innovation. Previous studies on how characteristics of knowledge impact knowledge recombination are consistent with the literature of knowledge integration (Luca and Atuahene-Gima, 2007) In this literature, knowledge integration refers to the formal processes and structures that ensure the capture, analysis, interpretation and integration of market and other types of knowledge among different functional units within the firm (Luca and Atuahene-Gima, 2007; Olson, Walker and Ruekert, 1995; Zahra, Ireland, and Hitt, 2000). So far, current literature has done a good job in elaborating how components, components characteristics and organizational features facilitate or impede innovation process. However, the effects of knowledge configuration on innovation have not been considered. 15

25 Effects of knowledge configuration on innovation creation Even though the significance of knowledge configuration has long been recognized in literature (Schumpeter, 1939; Penrose, 1959;Nelson and Winter, 1982; Henderson and Clark, 1990), it is strange that the subsequent research has assumed that the exact form of recombination is homogeneous across subjects. Consider the following statement by Galunic and Rodan (1998): Regardless of the exact form recombinations come in, however we hold that recombinations depend upon knowledge flows. Their rationale for making this argument was to focus on how characteristics of knowledge and its social organization influence resource recombination (Galunic and Rodan, 1998). However, as noted by Carnabuci and Operti (2013), creating innovation through knowledge recombination is neither a matter of chance nor mindless bricolage (page?). Instead, it evolves sophisticate design and planning. More importantly, different firms exhibit different ways of recombining knowledge components to create innovation (Carnabuci and Operti, 2013). Thus, researchers that hold different firms configurations the same will miss the opportunity to explore this essential element relating to knowledge recombination. It is vital that future research also takes into account the issue of how knowledge is recombined. If novelty of knowledge recombination does not depend solely on the new diversity of knowledge components, as suggested in prior research, and instead also relies significantly on how existing knowledge components are linked, it remains unclear how firms create innovation efficiently through knowledge configuration rather than adding new components. The majority of current research focuses on the mechanisms by which diverse knowledge components are gained. We argue, however, that knowledge configuration, however, is conceptually distinct and inherently more 16

26 challenging than knowledge components acquisition (Tiwana, 2008). Unfortunately, the research of configuration of knowledge components is silent on the form of knowledge recombination. Although we recognize that different configurations among knowledge components does matter, as Henderson and Clark (1990) argue even the mere rearrangement of previous components may lead to destabilizing technological change, we do not have a good understanding what different configuration forms are and how they impact innovation creation. Without understanding different ways to configure knowledge, there will be a misleading in both theoretical and managerial development. Current works put emphasis on knowledge components to study innovation creation (Rosenkopt and Nerkar, 2001; Ahuja and Lampert, 2001; Vermeulen and Barkema, 2001; Baum, Li, and Usher, 2002; Benner and Tushman, 2002; He and Wong, 2004; Taylor and Greve, 2006; Carnabuci and Operti, 2013) and exploring recombination among existing knowledge components allow organizations to efficiently avoid useless or unvalued innovations due to the familiarity of existing knowledge components (Davis and Eisenhardt, 2011).. Unfortunately, as noted by Kim and Kogut (1996), the repeated application of a particular set of technologies or organizing principles eventually exhausts the set of potential combinations. Also, as suggested by Fleming (2001), too many combinations of existing knowledge components yield diminishing returns due to exhaustion if we hold the configuration (the way to recombine them) the same. Following suggestions from current literature, firms often turn to external sources such as new knowledge acquisitions to gain new ideas and insights (Rosenkopf and Almeida, 2003). However, the greater the extent of new knowledge components from external sources, the more difficulty there will be to recombine or integrate them (Tiwana, 2008). Knowledge recombination of diverse knowledge 17

27 components involves uncertainty of the value, potentially long time frames and difficult technology processes or trajectories that are hard to evaluate ex ante or even ex post (Davis and Eisenhardt, 2011; Dosi, 1982; Dougherty, 1990; Henderson, 1995; Tripsas, 1997). Too-diverse knowledge can lead to unwieldy and impractical outputs and two-focused knowledge can result in competence trap in which organizations are locked in their old activities and cannot absorb new knowledge and information (Arthur, 1989; March and Simon, 1958; Taylor and Greve, 2006). Thus depending on exploring knowledge components in new knowledge domains make it very difficult for organizations to manage. As a result, it is necessary and urgent to study the role of knowledge configuration in current literature. In the section below, we will illustrate how knowledge configuration influences innovation creation by providing some useful examples. 18

28 Chapter 3: Isolating the Effects of Changed Linkages on Innovation Outcomes: Holding Knowledge Components Constant in Knowledge Recombination Innovation has been characterized as a knowledge recombination process involving at least two distinct dimensions knowledge components, referring to the portion of an innovation that embodies a core design concept and delivers a welldefined function (Fleming, 2001; Fleming and Sorenson, 2004), and the linkages between these knowledge components, defined as the connections affecting the meanings of combined knowledge components (Schumpeter, 1934; Henderson and Clark, 1990). While the majority of the recombination literature has focused solely on understanding the role of knowledge components, Carnabuci and Operti (2013) point out that these do not exist in a vacuum the manner in which components are linked plays a fundamental role in determining their meaning. Indeed, in their pathbreaking research, Henderson and Clark (1990) show how significantly changed linkages between knowledge components give rise to new innovations that other firms are unable to copy. However, change in knowledge linkages can occur within a continuum between the two extremes of no change and complete change. We lack a systematic explanation for how variations in changed linkages among knowledge components influence innovation novelty and value. Nonetheless, in the decades following Henderson and Clark s (1990) study, no research has attempted a deeper examination of the role played by linkages in the innovation process. 19

29 The purpose of this research is to address this issue. Holding knowledge components constant in order to highlight the effects of changed linkages, we examine (a) the relationship between linkage change and innovation novelty, and (b) the connection between changes in linkages and the relative abilities of the innovating firm and other firms in using the resulting innovation. Establishing linkages among otherwise isolated knowledge components enhances their joint interpretation and meaning, generating new knowledge. By connecting such knowledge components, linkages allow a larger meaning to be created than is possible from each one operating individually. It should be noted, however, that the knowledge recombination perspective does not provide insight into how the linking of knowledge components is actually accomplished by organizational members, establishing only that they are in fact linked. This is problematic since, within an organizational setting, the linking of components does not happen automatically rather, members must be motivated to integrate knowledge that is otherwise isolated or dispersed in a manner that elicits the desired outcome. Their ability to adhere to a pre-established template or program for linking such knowledge depends on the institution of appropriate organizational routines. Thus, we argue that, in an organizational setting, linkages operate through organizational routines, which provide imperatives for organizational members to draw upon and integrate the necessary knowledge and develop appropriate skills for the accomplishment of organizational goals. As Nelson and Winter (1982) note, tacit connections between disparate knowledge is particularly crucial for the creation of skills, resulting in coordinated information processing that is smoothly accomplished while difficult to articulate. However, making connections among such knowledge in a way that is meaningful for organizational purposes does not happen automatically it has to be 20

30 made to happen. This crucial role is performed by organization routines. Consistent with this, we conceptualize linkages connecting knowledge components, hereafter linkages, as routines. (Since a given individual alone may not be able to link all the necessary knowledge resulting in desired organizational outcomes, multiple individuals may be required to participate in this regard, each taking responsibility for different parts of the process. Thus, linkages reflect the multi-actor and distributed nature of routines (Cohen and Bacdayan, 1994).) Drawing on this conceptualization of linkages as a reflection of routines, we consider how, after holding knowledge components constant, change in two features of linkages extent (the overall connectedness of knowledge components) and local embeddedness (the degree to which a focal linkage operates with other closely intertwined linkages) influences innovation outcomes. We expect that (a) greater change in these features of routines enhances the novelty of a firm s innovation by creating new insights from altered combinations of existing knowledge components, (b) difficulty in appreciating the changed routines impedes external firms ability to effectively use the innovation, and (c) greater change in routines affords a comparative advantage to the innovating firm over external firms in using the innovation. Six hypotheses are developed in relation to these relationships. The hypotheses are tested using patent data for 33 of the largest firms in the pharmaceutical industry between 1976 and In order to isolate the effects of linkages, we identify 4,014 pairs of prior and subsequent patents from a single firm that have the same exact knowledge subclasses. We construct a network for each of these patents based on the pairwise coupling relationships or ties between every two components within the patent, with subclasses serving as nodes and coupling relationships as ties (Yayavaram and Ahuja, 2008). To measure the strength of a 21

31 linkage, we adapt Fleming and Sorenson s (2004) method for measuring coupling in a manner that accounts for the history of how the two components have been combined with each other in all of the firm s previous patents. Knowledge components exhibiting closer interdependent relationships over time are reasoned to have stronger underlying routines. This dynamic approach, which takes into account ongoing changes in the linkages associated with their innovations, allows us to capture the evolutionary process by which firms develop their sense-making routines. We identify changes in extent and local embeddedness of linkages for each pair of patents. These data form the basis for testing the hypotheses. The findings of this research indicate that, after holding components constant, changes in extent and local embeddedness of linkages do indeed enhance innovation novelty. At the same time, such changes are seen to impede the ability of external firms to utilize the resulting innovation, giving the innovating firms an advantage in this regard. These findings generally uphold our argument that change in linkages represents alteration of underlying routines that connect the firm s individual knowledge components, creating important new insights that give rise to innovation. Since linkages are manifested in opaque routines, they are difficult for other firms to observe or copy, and create comparative advantages for the innovating firm. This study contributes to the knowledge recombination literature in a number of ways. Our treatment of knowledge linkages as routines provides a theoretical rationale for how linkages among knowledge components influence the combined conceptualization of knowledge components and give rise to a larger meaning. This research demonstrates that, above and beyond change in knowledge components that has been the primary focus of prior literature, changing the routines that combine these components can bring about new and unique insights that allow for the creation 22

32 of novel innovations. This research is also the first to account for the evolutionary nature of routines associated with innovation. By capturing the dynamic development of linkages between knowledge components in successive patents over time, we create a means to identify features of routines that have thus far been invisible to researchers. Finally, while prior literature has often equated the novelty of an innovation with its value, this research shows that the two are not interchangeable. Instead, our findings suggest that the innovating firm has an advantage in using the novel innovation stemming from the changing of linkages over external firms. The difficulty of external firms in utilizing novel innovations derived from the innovating firm s changed routines validates the role of tacit firm-specific knowledge in creating competitive advantage from innovation. LINKAGES IN KNOWLEDGE RECOMBINATION The example of a desktop computer and a laptop computer helps us to understand the importance of changed linkages for innovation. Both desktops and laptops have five major knowledge components that are the same, including the central processing unit (CPU), a control unit, memory that stores both data and instructions (i.e., random-access memory (RAM), read-only memory (ROM), and hard disk), input mechanisms (i.e., keyboard or mouse), and an output mechanism (monitor). Note that, while the knowledge for these components in both desktops and laptops remains largely the same, the manner in which they are linked differs, through new connections that facilitate the coordination among multiple components. The linkages among different pieces of knowledge in a laptop are relatively more closely related than those in a desktop. For example, the strength of the connection to link a keyboard and a screen together on a laptop is closer and stronger than that on a 23

33 desktop, since a laptop link these two pieces of knowledge by using a more direct and short means. Thus change in how knowledge is thus linked gives rise to new innovation. 1 Innovation is a process of knowledge recombination (Fleming, 2001; Katila, 2002; Katila and Ahuja, 2002), in which firms integrate knowledge components from varying domains in order to create new knowledge. As early as 1939, Schumpeter noted that, innovation combines factors in a new way, or that it consists in carrying out new combinations (p. 84). He describes new innovation as derived from recognizing the value in the underlying parts of diverse systems and recombining these parts in a novel fashion. Similarly, Penrose (1959) asserted that the services yielded by resources are a function of the way in which they are used exactly the same resources when used for different purposes or in different ways and in combination with different types or amounts of other resources provides a different service or set of services (p. 25). Consistent with this, Nelson and Winter (1982) note that the creation of any sort of novelty in art, science or practical life consists to a substantial exert of a recombination of conceptual and physical materials that were previously in existence (p. 30). 2 From this prior work emerge three key points regarding the role of linkages in knowledge recombination. The first is that the same building blocks of firm resources and knowledge components have the potential to be recombined in alternative ways, rather than in only one way. The second point is that recombination requires effort, undertaken to achieve an underlying purpose rather 1 Another example of two generations of portable phones also illustrate the effect of knowledge linkages on innovation outcome. While the earlier generation ( nonsmart ) cell phone had a separate screen and keyboard, the later generation smart phone, such as the Apple Iphone, has a keyboard incorporated into the screen. The screen- and keyboard-related knowledge for both phones play the same function, but different linking mechanisms are used to combine these two pieces of knowledge. 2 Scholars have characterized the firm s ability to recombine its knowledge for innovation as combinative capabilities (Kogut and Zander, 1992), reconfiguring competence (Galunic and Rodan, 1998), or recombinant capabilities (Carnabuci and Operti, 2013). 24

34 than for no particular reason. Finally, the manner in which components are recombined determines the nature of the innovation achieved. Despite the recognition that recombination of existing knowledge components can result in innovations, the vast majority of studies on knowledge recombination have concentrated on understanding how characteristics of knowledge components explain innovation (Fleming, 2001; Galunic and Rodan, 1998; Rosenkopf and Nerkar, 2001; Carnabuci and Operti, 2013; Taylor and Greve, 2006; Arthur, 1989; March and Simon, 1958). For example, Fleming (2001) has shown that familiar components are better understood and therefore result in more valuable innovations, while Rosenkopt and Nerkar (2001) stress the value of diverse new components spanning technological categories and organizational boundaries in this regard. Ahuja and Lampert (2001) and Carnabuci and Operti (2013) find that diversity of knowledge increases the probability of radical innovation. In examining the age of knowledge, Katila (2002) finds that older intra-industry knowledge hurts innovation but older extra-industry knowledge actually helps innovation. While much of the literature has concentrated on the attributes of knowledge components, it should be noted that components do not exist in a vacuum their meaning is determined by how they are linked or connected with each other. As noted by Carnabuci and Operti (2013), creating innovation through knowledge recombination is neither a matter of chance nor mindless bricolage (p. 1593), it involves sophisticated design and mindful planning. This helps to explain why firms exhibit contrasting approaches for recombining knowledge components. Henderson and Clark s (1990) influential work opened the door for studying how linkages among knowledge components influences innovation, showing how mere rearrangement of existing components can lead to innovation that another firm is 25

35 unable to copy. They envision linkages or in their words, architecture as an embodiment of communication channels and information filters through which a larger sense is made of the components. Linkages contains tacit understanding about components that is difficult for others to identify or transfer, creating impediments in their ability to duplicate the innovation. They note, Since the core concepts of the design remain untouched, [another] organization may mistakenly believe that it understands the new technology (Henderson and Clark, 1990: 17). Interestingly, since Henderson and Clark s (1990) original study, there has been relatively little additional insight into the role of linkages in knowledge recombination (Fleming 2001, Rosenkopt and Nerkar, 2001). 3 In particular, we continue to lack a detailed and systematic explanation for how change in differing attributes of linkages among knowledge components affect innovation outcomes. Linkages can be made to vary in a number of ways for example, their relative number, strength or embeddedness in a set of other linkages each of which can have its own effect on a subsequent innovation. That is, less drastic change in linkages than suggested by Henderson and Clark (1990) may in fact engender novel innovation. There is a need for a more in-depth understanding of how changes in linkages can vary, and how these varying attributes of linkages in turn influence innovation outcomes. Tracking even minor changes in linkages would allow us to zoom in on their influence on innovation. Furthermore, while Henderson and Clark demonstrate the difficulty of firms to copy architectural innovations of new entrants, there remains 3 Since Henderson and Clark (1990), the role of linkages for innovation has been much appreciated by scholars (e.g., Hill and Rothaermel, 2003; Rindova and Petkova, 2007; Kaplan, 2008; Taylor, 2010; Foss, Laursen and Pederson, 2011), yet the majority of research continues to focus on attributes of components (Fleming, 2001; Brusoni, Prencipe and Pavitt, 2001; McEvily and Chakravarthy, 2002; Furlan, Cabigiosu and Camuffo, 2014). An exception is Yayavaram and Ahuja (2008), who examine linkages in innovation-related knowledge at the firm level. 26

36 a need to understand how changes in the firm s own knowledge linkages affect their own ability to innovate. In order to address this void in the literature, it is necessary to reconsider the role played by linkages in the knowledge recombination process. In the section below, we develop the conceptual relationship between organizational routines and knowledge linkages, highlighting the role of routines in combining and recombining knowledge within the firm. We argue that knowledge linkages reflect the organizational routines that facilitate the combination of diverse and distributed knowledge by organizational members, creating new and enhanced meanings that help them to accomplish organizational goals. It follows that changes in linkages among such knowledge occur through changed routines. THE ROLE OF ROUTINES IN KNOWLEDGE RECOMBINATION Researchers have noted that organizational routines are patterned sequences of learned behaviors (Cyert and March, 1963; Teece, Pisano and Shuen, 1997), coordinated across multiple organizational members in a way that accomplishes complex organizational goals (Cohen and Bacdayan, 1994; Pentland and Reuter, 1994). Routines ensure that members acquire, integrate and deploy knowledge relevant to these goals (Nelson and Winter, 1982). Since a single individual would be unable to efficiently process all the knowledge necessary for this purpose, multiple individuals are typically involved, each with responsibility for different parts of a routine. Thus, as Cohen and Bacdayan (1994) note, a critical feature of routines is their multi-actor and distributed nature. The primary role played by knowledge in this regard reflects the cognitive basis of routines not only is the knowledge necessary for enacting routines committed to and drawn from memory, the development of tacit 27

37 understanding and skill associated with routinized activity occurs through experiential learning and ongoing fine-tuning of knowledge connections (Nelson and Winter, 1982). As Cohen and Bacdayan (1994) suggest, such emergent, historical qualities of routines imply that sorting out the real functions of the different components of a routine can be a very challenging task (p. 556). How do routines influence the use of knowledge for accomplishing tasks? Routines facilitate the effective combining of different or disparate pieces of knowledge that together create new or greater meaning than the individual parts. Their disparate nature ensures that, in the absence of some organizational imperative to combine them, they would be isolated from each other. When linkages are created among such knowledge components, they impose new conditions for their interpretation and meaning. The connections between differing knowledge domains allows organizational members to develop newer inferences and more original insights than would otherwise be possible (Nelson and Winter, 1982; Dosi, Nelson and Winter, 2000). Since these combinations are unlikely to come about without organizational intervention, the manner in which knowledge components are linked to each other reflects the nature of routines required to create their combined meaning. Thus, the unique firm-specific knowledge that sets a firm apart from others is accomplished through routines that facilitate organizational members ability to create, appreciate and use it. A firm s established set of routines serves as a blueprint for how it organizes its activities (Nelson and Winter, 1982), exercised as a programmed gene or habit embedded in organizational members minds (Stene, 1940, Simon, 1976; March and Simon, 1958; Cyert and March, 1963; Gioia and Poole, 1984; Carley, 1996; Carley and Lin, 1997). As such, it reflects ongoing or persistent patterns of activities that are 28

38 difficult to change (Cohen and Bacdayan, 1994; Galunic and Rodan, 1998). For example, Teece (1997:1335) notes that, departure from routines will lead to heightened anxiety within the organization, while Nelson and Winter (1982: 134) point out that attempts to change routines often provoke a renewal of conflict. The embeddedness, persistence and inertia of an organization s routines create obstacles for change. Since routines are instituted over time in a path-dependent manner, a significant investment of time and resources may be required to change the ways in which activities are accomplished. In all, once established, changing of routines is very difficult for firms (Nelson and Winter, 1982; Teece, 1982; Edmondson, Bohmer and Pisano, 2001; Feldman and Pentland, 2003; Bresman, 2013; Hargadon and Sutton, 1997). Even though difficult to accomplish, a change in the routines that combine or link knowledge components plays an important role in knowledge recombination. Changed linkages provide a new conceptual template for reasoning the role of the knowledge components (Nelson and Winter, 1982; Walsh, 1995; Elsbach, Barr and Hargadon, 2005), creating fundamentally new knowledge that can give rise to innovation (Schumpeter 1939; Henderson and Clark, 1990). By changing linkages, new routines alter previously established logic, create new cause-effect relationships among knowledge components, change interpretation of knowledge components, and generate new insights. Thus, a change in the routines gives rise to new innovations (Galunic and Rodan, 1990; Teece, Pisano and Shuen, 1997). Drawing on the theoretical relationship between routines and knowledge linkages outlined above, we develop testable hypotheses between characteristics of linkages and innovation outcomes in the next section. 29

39 HYPOTHESES DEVELOPMENT In this section, we develop a theoretical model that draws on the evolutionary theory of routines to explain how linkage changes among knowledge components influence innovation outcomes such as novelty and use. Holding knowledge components constant in consecutive innovations, we consider the effects of change in two key characteristics of linkages: (1) extent of linkages, and (2) local embeddedness of linkages. We argue that these characteristics of linkages reflect routines by which organizational members can make sense out of the knowledge components. Changes in these routines alter the meaning associated with the components, influencing innovation novelty. Nonetheless, we expect that the opaqueness of changed routines will reduce external firms ability to adopt the innovation relative to the innovating firm. Below we develop six hypotheses corresponding to these relationships, as shown in Figure 3.1. Change in extent of linkages and innovation novelty As discussed earlier, linkages among knowledge components reflect organizational routines that create heightened meaning from their combination. The linking of two otherwise independent knowledge components creates enriched knowledge flows between them that enable distinctive activities to be performed (Teece, Pisano and Shuen, 1997). The linkages influence how different pieces of knowledge are understood and reasoned in relation to each other. The way in which a firm configures a set of knowledge components that is, the extent to which components are isolated or integrated, and the strength of these connections reflects 30

40 the nature of the routines required for making sense of them. Thus, different types of linkages represent different types of routines. When the linkages among knowledge components are changed, it indicates an alteration in the routines through which organizational members conceptualize the underlying issues that lead to an outcome. The way in which a particular combination of knowledge components is interpreted and reasoned is no longer the same as it was originally. The change in extent of linkages indicates that firm has changed its understanding of how knowledge components work together (Dosi, Nelson and Winter, 2000). As noted earlier, changing routines is not easy since, once established, they become part of the habitual ways of conceptualizing and accomplishing tasks within the organization. Patterns of activities become ingrained in perceptions and embedded in organizational members minds of how things ought to be done (Cohen and Bacdayan, 1994). However, when a firm is able to actually change linkages among knowledge components, leading to new routines for making sense of them, the insights gained from this change in routines will be novel. The previously accepted logical connections between knowledge components are now altered, and instead, a new logic is created for how a given set of knowledge components work together. This different logic sets a new way for the firm to combine knowledge components and new outcomes will be achieved due to the departure from its old routine framework. As we showed earlier, novelty is generated through reconfiguring the ways in which competencies are linked to jointly achieve some broader purpose (Galunic and Rodan, 1998: 1195). 31

41 Thus, we propose the following hypothesis: H1: Holding components constant, change in the extent of linkages is positively related to the novelty of the resulting innovation. Change in local embeddedness of linkages and innovation novelty The meaning of a linkage among knowledge components is also influenced by the context or neighborhood within which it functions, formed by other linkages that operate in close proximity and with which it is also connected (Nelson and Winter, 1982; Teece, 1982, 2007). In order to function coherently, a particular linkage may have to coordinate with other closely related linkages (Simon, 1962; Kogut and Zander, 1992; Galunic and Rodan, 1998). The simultaneous operation of the routines underlying these linkages facilitates highly tacit and complex knowledge flows among components. The close and interdependent relationships among these routines create a context that enhances the functioning of any particular linkage operating within it. It would be difficult to correctly perceive, assess and interpret the meaning of such a linkage without considering the others operating along with it, since organizational routines containing tacit knowledge cannot effectively stand alone (Nelson and Winter, 1982; Teece, 1982). These closely related linkages reflect a highly coherent set of underlying routines that generate a larger meaning than their simple aggregation. In this way, the context within which a linkage is embedded can create greater meaning than its sole operation, since the effectiveness of the knowledge transferred is enhanced through other closely intertwined routines. As a result, a given linkage has different meanings in different contexts. When local embeddedness of linkages changes, it indicates that the context of organizational routines has also changed. Since the context of routines within which a 32

42 particular routine is embedded plays an important role in its interpretation, changes to this context in turn changes its meaning. These new meanings generate new insights and functions, which give rise to novelty. Accordingly, we propose the following hypothesis: H2: Holding components constant, change in local embeddedness of linkages is positively related to the novelty of the resulting innovation. Change in extent of linkages and external use In order to use a new innovation created by another firm, firms will need to appreciate the underlying routines associated with it and reconfigure their current routines to reflect such new understandings. However, its current routines are likely to hinder both activities. As noted by Nelson and Winter (1982), Prevailing routines define a truce, and attempts to change routines often provoke a renewal of the conflict which is destructive to the participants and to the organization as a whole (p. 134). As a result, changing a firm s routines in order to utilize novel innovations created by another will be very difficult. We noted earlier that changed linkages among knowledge components indicates a change in the conceptualizing, reasoning, coordination and working patterns of the underlying routines (Nelson and Winter, 1982; Dosi, Nelson and Winter, 2000; Henderson and Clark, 1990). However, for external firms that did not create the innovation but wish to adopt it, these changes will likely be difficult to understand, for two reasons. First, when no new knowledge components are involved in the adoption process, it will be hard to correctly detect what needs to be done to 33

43 reconfigure their underlying routines and to accurately recognize all necessary changes in this regard (Henderson and Clark, 1990; Galunic and Rodan, 1998). As noted by Teece et al. (1997), A firm s previous investments and its repertoire of routines (its history ) constrain its future behavior (p. 17). That is, a firm s current routines impair organizational members ability to perceive, understand and adopt significantly different ones (Nelson and Winter, 1982; Teece, Pisano and Shuen, 1997). As Henderson and Clark (1990) note, architectural innovation can often initially be accommodated within old frameworks The fact that they may be talking about the wrong things may only become apparent after there are significant failures or unexpected problems with the design (p ). As a result, a firm s current routines make it very difficult to comprehend the necessary changes in its routines for adopting a new innovation. When the extent of change in linkages is large, the degree of commensurate change in underlying routines will also be large, increasing the difficulty of external firms in detecting them. Second, even when the external firm recognizes the change in the underlying routines, it may not possess the ability to actually make this change in its routines. A firm s current routines are the foundation for its capabilities (Nelson and Winter, 1982; Teece, Pisano and Shuen, 1997), engrained in the psyche of organizational members. Existing routines represent persistent patterns of activities based on past experiences (Nelson and Winter, 1982), exercised over time to become a habit for the firm, programmed into its structure (Nelson and Winter, 1982; Stene, 1940; Simon, 1976; March and Simon, 1958; Cyert and March, 1963; Allison, 1971; Gioia and Poole, 1984; Carley, 1996; Carley and Lin, 1997; Levitt et al., 1999). Hence, the greater the extent of change in linkages among knowledge components, the more difficult we would expect it to be for external firms to adopt or use the innovation. 34

44 In line with this argument, we propose the following hypothesis: H3: Holding components constant, change in the extent of linkages is negatively related to external use of the resulting innovation. Change in local embeddedness of linkages and external use We noted earlier that the local embeddedness of a linkage, consisting of other interdependent and intertwined linkages in its neighborhood, influences its meaning and interpretation. The entire set of linkages defines how a given routine works within other routines to support knowledge recombination (Nelson and Winter, 1982; Teece, 1982; Kogut and Zander, 1992; Grant, 1996). In a new innovation, an increase in local embeddedness of linkages indicates that background information plays a greater role in correctly interpreting their meaning. There are two implications of this for external firms adoption of an innovation characterized by increased local embeddedness of linkages. First, the increased complexity of such a context makes it difficult for external firms to correctly perceive and identify the larger meaning that is created. Since external firms are unable to appreciate the rich detail in which may be embedded understanding of why a routine functions as it does in that context (Galunic and Rodan, 1998: 1198), their ability to accurately understand the knowledge associated with the routine is handicapped. Since these underlying routines reflect interdependent activities, the high level of simultaneity and interaction among a set of locally embedded routines increases the difficulty of other firms organizational members in adjusting their mental models to them. According to Galunic and Rodan (1998), mental models increase the tendency to perceive new 35

45 incoming information in a manner congruent within one s preconceived notions and unconsciously ignore any nonconforming information. Second, when complexity of this context is high, external firms face a great deal of hardship in implementing all the necessary changes of routines reflected in interdependent linkages among knowledge components. The intertwined linkages make the routines and capabilities that support knowledge recombination activities highly synergistic (Nelson and Winter, 1982, Teece, Pisano and Shuen, 1997). Thus, even if the requisite changes are identified, their implementation may be hampered by the existing relational structure within the organization, since a firm s capabilities cannot be separated from how it is currently organized (Kogut and Zander, 1992). In this way, external firms current routines may hamper the ability to invoke the rich knowledge flows and information exchanges associated with a new set of locally embedded linkages. Thus, we propose the following hypothesis: H4: Holding components constant, change in the local embeddedness of linkages is negatively related to external use of the resulting innovation. Change in extent of linkages and internal vs. external use Both the innovating and external firms endeavor to utilize novel innovations, including those created by a change in linkages. However, the firm that originally created the innovation is argued to have an advantage over external firms in adopting it, since it will have a better understanding of the changes in both linkages and context in which they operate. 36

46 When an innovation is created through a change in linkages, the innovating firm retains a close understanding of the motivation and rationale for why these changes are necessary since these were figured out in the process of creating the innovation. This firm has experienced how the underlying routines that may directly or indirectly support knowledge recombination work. In contrast, external firms trying to imitate the knowledge recombination process associated with the innovation may not recognize the high levels of coherence within the linkages as well as the firm that actually created it. As noted by Teece, Pisano and Shuen (1997), partial imitation or replication of a successful model may yield zero benefits (1997: 519). As a result, the innovating firm is expected to experience less difficulty than external firms with regard to the understanding of necessary changes in both routines and the context which routines root in. In utilizing the innovation it created, the focal firm is also expected to have less difficulty in accomplishing the necessary changes in routines compared to external firms. Since knowledge creation tends to path-dependent, it is likely that the firm creating the innovation drew upon and extended its present routines and capabilities, whether or not done intentionally. Due to this, it is in a good position to identify problems of adoption and figure out ways to solve them. In contrast, when external firms endeavor to copy changes in routines, they will face more challenges since they are less likely to have a connection between their existing routines and the new routines they need to employ. Unlike the innovating firm, it is improbable that external firms will already possess many or most of the necessary supportive routines and capabilities to help them to utilize the innovation created by a different firm. Without such a history, external firms face more difficulty in utilizing the innovation, since their prior routines constrict the development of new capabilities (Nelson and 37

47 Winter 1982, Kogut and Zander, 1992). Indeed, when the change in linkages is significant, the difficulty external firms face is intensified in this regard. Thus, we propose following hypothesis: H5: Holding components constant, more change in the extent of linkages gives the innovating firm a greater relative advantage over other firms in using the resulting innovation. Change in local embeddedness of linkages and internal vs. external use In the case of an innovation characterized by an increase in local embeddedness of linkages, we expect a difference in the ability to adopt the innovation between the innovating firm and other firms. Compared to the firm that created the innovation, external firms are expected to have more difficulty in accurately understanding the complete meaning and function of these interdependent linkages, for the following reasons. When the context of linkages is more complex that is, the degree of local embeddedness of these linkages is greater firms current routines serve as the foundation for comprehending these linkages (Hoetker, 2007). Since external firms have different routines for making sense of knowledge components, their interpretations may be inconsistent with the meanings originally associated with the innovation. External firms will be less likely to replicate the entire set of intertwined routines necessary for the innovation to work, reason how these routines jointly work, or appreciate the amplified meaning delivered by their joint operation. On the other hand, the firm that has experienced the creation process has already worked out the role of each routine within its complex setting. Its routines will therefore reflect the templates, skills and experiences it has accumulated for 38

48 performing its activities (Nelson and Winter, 1982). Since the innovating firm already possesses most of the necessary routines and has gained detailed knowledge on how to manage them, it gains an advantage over external firms from utilizing its own innovation. Even if external firms are able to eventually correctly diagnose the meaning afforded by these highly interdependent routines, it will be necessary for them to make significant organizational changes in order to appropriately utilize the innovation, possibly constructing the complex context from scratch. Bresman (2013) suggests that changing routines involve four distinct sets of processes: identification, translation, adoption and continuation. External firms ability will be limited for replicating these processes in order to reconstruct the context in which linkages function. Indeed, the more intricate the context, the more difficulty external firms will have in instituting the necessary changes. In contrast to external firms, the innovating firm is not starting from scratch. While there is a change in local embeddedness of their existing routines, at least some of the requisite routines are already in place due to their path-dependent nature (Nelson and Winter, 1982). Since the change that they have to institute is much lower in scope, they are more likely to be able to successfully adopt the innovation. Thus, we propose following hypothesis: H6: Holding components constant, more change in the local embeddedness of linkages gives the innovating firm greater relative advantage over other firms in using the resulting innovation. 39

49 METHODS Data and variables To test our hypotheses relating knowledge linkage features of an innovation and its outcomes, we use patent data from the pharmaceutical industry. Even though patent data are not perfect records of technological innovations (Levin et at., 1987), and are conditional upon successful patent applications, they have been utilized by many other scholars studying innovation (Guler and Nerkar, 2012). Patents in the pharmaceutical industry are thought to appropriately represent firms innovative efforts since their patent propensity is high and all relevant players in this industry patent at the United States Patent and Trademarks Office (USPTO). USPTO patents provide much information on each patented innovation, including their application and approval dates, classes, subclasses, claims, and other useful information. The USPTO uses subclass to identify the specific technologies associated with the patent, and the approximately 100,000 such subclasses allows for fine-grained classification of key pieces of knowledge underlying a patented innovation (Fleming and Sorenson, 2004). Since the whole assigning process is done by USPTO, the information gained through examining subclasses is not likely to be biased by a firm s strategic decisions (Carnabuci and Operti, 2013). As a result, subclass information is an ideal approach for studying knowledge recombination (Fleming, 2001; Fleming and Sorenson, 2004; Carnabuci and Operti, 2013). Hereafter, we refer to subclasses as knowledge components or simply components. By observing the entire patenting history of a firm, we are able to identify historical connections between components. Thus, we utilize an evolutionary approach to construct the linkages of each of the firm s innovations. 40

50 Due to the fact that the pharmaceutical industry is highly concentrated, we collect data on patents from the 33 largest firms constituting about 90 percent market share, following previous research (e.g., Guler and Nerkar, 2012). Since the goal of the research is to understand the effects of changes in knowledge linkages, we need to identify corresponding pairs of firms patents for which knowledge components remain the same but linkages vary. After examining all patents granted by USPTO from the period between 1975 (the earliest full-text record available for these patents) and 2014, we were able to identify 4,014 pairs of firm-specific patents in which the only difference is a change in linkages from the first patent to the next. These pairs of patents were identified by tracking the backward citations of a patent. Since we are interested in the change in knowledge recombination occurring within a firm, we only consider the backward citations that belong to this firm. In this way, we are able to construct pairs of successive firm-specific patents that utilize the same subclasses but differ in innovation outcomes. As can be seen in Table 3.1, among the 33 largest firms within the pharmaceutical industry, Procter and Gamble has the most pairs of patents with 1376 followed by Merck with 345 and Bayer with 263, and Chiron with the fewest at 2 pairs. The wide variation in patents reflects differences in innovativeness among firms. Interestingly, when submitting patents to USPTO, firms sometimes use different formats for their names or make spelling errors, and thus, disambiguation of firms names is a major problem when assigning a firm to a patent (Li et al., 2014). To illustrate, Table 3.2 shows the example of Hoechst, which has used 27 different formats for its name. To ensure a more accurate and complete dataset, this study uses fuzzy matches to capture all the different formats used for firm names in the USPTO database. 41

51 Dependent variables Novelty of innovation. The novelty of a patented innovation is indicated in the number of claims within the patent. A claim is a statement listed within the patent noting a new element of the patent reflecting knowledge previously unknown to the field (Beaudry and Schiffauerova, 2011). More claims associated with a patent reflect greater novelty or newness of the innovation (Duchesneau Cohn and Dutton, 1979; Hage, 1980; Ettlie, Bridges and O Keefe, 1984; Dewar and Dutton, 1986). Thus, we measure the novelty of an innovation by the number of claims within the patent. While prior research has used other methods to capture novelty, including number of independent claims and number of forward citations, we feel the above measure is a superior approach in this regard. An independent claim is a statement indicating the uniqueness of the knowledge used in the patent. However, the higher the degree of novelty associated with an independent claim, the more likely it will require more than just one independent claim to elaborate it, as well as other supporting claims to completely describe it. As a result, simply counting the number of independent claims may not fully capture the degree of novelty associated with a patent. Another measure of novelty is based on the forward citations received by a patent. A concern with this approach for measuring novelty is that the number of forward citations of a patent is affected by firms ability to use it. We thus consider it to be more appropriate for reflecting the use of a patent, and employ it for this purpose below. External & internal use of an innovation. The use of a patent is measured by its forward citations (Fleming, 2001; Fleming et al, 2007, Yayavaram and Ahuja, 2008). Since every patent is built upon prior patents, the creation of patent is a path- 42

52 dependence process utilizing prior knowledge. The more forward citations of a patent, the greater is its role in creating future patents. To measure external use of a patent, or the use of an innovation by firms other than the one that created it, we count the citations of the patent made by these other firms. To measure internal use of a patent, or the use of an innovation by the firm that created it, we count the forward citations of the patent made by this firm. This study calculates the forward citations for all patents until June 2014 in order to capture more completely the usefulness of the patent. When a patent is not externally cited by many firms, but these firms cite it a lot, it results in high external citations. To better reflect the general usefulness of the patent, we also measure external use by taking account of the number of external firms that cite the patent. We use both measures in our analyses. Independent Variables: Measures of Linkages The measurement of the independent variables of this study is based on knowledge components and their linkages. As in prior research (Carnabuci and Operti, 2013; Fleming, 2001), knowledge components are captured in this study by the subclasses co-assigned to each firm s patented innovations. The patents in our sample have a total of 41,024 different subclasses representing knowledge components in our study. We conceptualize a linkage between knowledge components in a manner similar to coupling by other scholars (e.g., Fleming and Sorensen, 2004; Yayavaram and Ahuja, 2008). A linkage reflects how two components coordinate with each other in order to work appropriately. If the linkage between these components is strong, their underlying routines are closely connected, enhancing their joint operation and 43

53 leading to an amplified meaning. In order to assess the strength of a linkage, we measure the interdependence between the components. One way to do this is suggested by Fleming and Sorenson (2004), who notes that a strong linkage between components creates high interdependence among them that will not make it easy to recombine with other components. In contrast, when interdependence is low, the addition of any new components is more easily accomplished, and allows for change in established routines. As a result, the strength of a linkage between two components is reflected in the ease of recombining the two components with other components. We construct the measure of linkage strength by using the procedure below. We measure the degree of interdependence between each set of two components by examining the number of other components being co-listed with them in all the prior patents before the focal patent. For example, to measure the linkage connecting component i and j, we count the number of other components recombined with component i and j in the entire set of the firm s prior patents. The greater the number of components that have been previously combined with components i and j, the greater will be the ease (or lower the difficulty) of recombining these two components, and the lower is the strength of the linkage between them. Next, we account for the likelihood that knowledge components vary in the number of purposes for which they can be used. Components that have more purposes will be used in more patents and thus will be employed more frequently. To address this issue, we divide the above by the number of patents that include both components i and j. 4 The formula for calculating a linkage between two components in a patent is as follows: 4 While the measure outlined here is similar to the coupling measure outlined in Fleming (2004), it also differs in a number of ways. First, while Fleming s measure calculates the coupling for an entire patent by taking the average value of the coupling for every subclass, the focus here is on 44

54 Linkage or coupling between components i and j =1/ [!"#$%!"!"#$%&!!'!!"#$%&'()*!"#$%&'(!"#$!"#$%&!!'!!!"#!!"#$%!"!"#$%&'(!"#$%#&!"!"#$!"#$%&!!'!!!"#! Finally, we calculate the strength of each linkage any two components in a patent and construct a linkage matrix for a patent that incorporates the values for every linkage within the patent. We then construct a linkage matrix for every patent possessed by a firm in which all the components are treated as nodes of a network and linkages as ties. Treating it as a weighted adjacent matrix for a network, this linkage matrix of a patent is the foundation for deriving the two salient characteristics of linkage studied here the extent and local embeddedness of linkages. The calculation of each linkage of every patent requires a very dynamic approach, since the strength of a linkage in any point of time is determined by its most recent value. Measuring linkages in this way allows us to better capture the dynamic way in which a firm develops linkages and designs the routines that recombine knowledge components. The amount of computer processing time required to calculate each and every linkage in all the patents is tremendously high, taking months of CPU time. To illustrate how changes in linkages among knowledge components influence knowledge recombination, we draw on the example of two consecutive patents owned by Pfizer (patents numbered and ). As seen in Figure 3.2, both ] the tie that exists between each set of two components within a patent. Second, Fleming s study conducts cross-industry level analyses, whereas the unit of analysis for this study is the firm level. Third, and most importantly, Fleming s measure is static, capturing a one-time snapshot of coupling. This paper calculates linkages in a more dynamic way, in which couplings between components have different values over time to account for past experience in how its knowledge is recombined each time the firm successfully files a patent. For example, since Bayer has filed 12,720 patents, its knowledge recombination will change as many times. 45

55 patents contain the same eight knowledge components pertaining to nitrogen, chalcogen, and others (shown at the nodes). The linkages among these knowledge components are seen in the lines that connect them. The darker the line, the stronger is the linkage. This example demonstrates that a successful new patented innovation can be created by changing only the linkages of a previous patent. Change in extent of linkages. We use network density to measure the extent of linkages in a patent. Density captures the degree of connectivity within a network and thus reflects the degree to which components within a network are tied to each other (Kijkuit and Van Den Ende, 2010). As noted earlier, we identified pairs of prior and subsequent patents in which knowledge components are held constant. To assess change in the extent of linkages, we take the absolute value of the difference between the extent of linkages in the subsequent and prior patents. Change in local embeddedness of linkages. We measure local embeddedness through the degree of clustering, which allows us to capture how closely tied a linkage is to others in its neighborhood (Yayavaram and Ahuja, 2008). Local embeddedness elicits continuous values, with larger values indicating greater local embeddedness. To capture change in local embeddedness of linkages, we examine the local embeddedness between the prior and subsequent patents. We measure the real change between the local embeddedness of linkages between the two. Control variables We control for a number of contextual features associated with a patent that might influence the relationships we are studying. A larger number of inventors could represent more knowledge and lead to more forward citations for the patent they 46

56 create (Singh and Fleming, 2010), while more subclasses indicate more diverse knowledge and thus may enhance innovation novelty (Fleming, 2001; Rosenkopt and Nerkar, 2001). For these reasons we control for both number of inventors and number of subclasses for each patent. Knowledge maturity may also influence the innovation outcome and its use, since mature knowledge is easier to understand and could be more easily utilized by firms (Sørensen and Stuart, 2000). Based on previous studies (Agarwal and Hoetker, 2007; Lanjouw and Schankerman, 2003), we construct technology maturity of a patent by the number of prior arts within this patent. Knowledge diversity can also influence the dependent variables (Argyres and Silverman, 2004; Jaffe et al., 2002; Singh, 2008), since the more diverse the range of knowledge, the more areas a patent touches and more potential uses it can generate. Knowledge diversity is calculated by the proportion, pi, within a patent s backward p 2 1 i citations that belong to technology class i. The formula for this variable is i, with values falling between 0 and 1. We also control for the number of prior patents that the firm has to indicate their innovation capabilities, since different capabilities may influence both innovation novelty and use (Kogut and Zander, 1992; Teece, Pisano and Shuen, 1997). This variable is measured by the number of patents that a firm has applied for in a five-year window prior to the focal patent. Thus we have 3856 pairs of patents entering the model. We have both focal firm age and patent age as control variables. However, negative binomial model analysis automatically omits patent age in the regressions due to its high relation with focal firm age. Firm age is also controlled for, since innovation capabilities develop over time. We also control for patent age since this influences the level of forward citations associated with the patent. 47

57 Methods Because our dependent variables novelty (number of claims), external and internal use of an innovation (number of external and internal forward citations) are count variables, a Poisson or negative binomial estimation method is used to conduct the analysis. Since our independent measures are change-related measures, a random effects model is appropriate (Yayavaram and Chen, 2013). We therefore use a negative binomial estimation approach with random effects to conduct the analysis. We also perform a Poisson estimation with fixed effects for robust checks. The negative binomial function is takes the following form: Γ(y + k) " k % P(Y = y X, k) = $ ' Γ(k)Γ(y +1) # k + µ & k y " µ % $ ' # k + µ & The expectation of Y is: E(Y) = µμ Applying a log link, we get: g(µμ)=log(µμ)=α + βx where X captures the independent variables including change in extent of linkages, change in local embeddedness of linkages, and control variables; y relates to the dependent variables including novelty, external forward citations and internal minus external forward citations. The equations examined take the following form (with differing dependent variables): 48

58 Log (Novelty) = α! + β! Change in extent of linkages + β! Change in non decomposability of linkages + controls+ e RESULTS AND DISCUSSION Table 3.3 displays descriptive statistics and correlation for all the variables in this paper. The control variables firm age and patent age are seen to be highly correlated, leading the negative binomial model analysis to automatically omit patent age in the regressions. As can be seen from the Table 3.3, the independent variables exhibit high correlation with each other, but the fact that they both remain significant within the analyses indicates that they are reasonably distinct constructs. The correlation coefficients for the other variables are reasonably low and thus appropriate for testing the hypotheses. 5 Table 3.4 shows the regression results for the relationship between changed linkages and innovation novelty. Model 1 of Table 3.4 is the baseline model, with results for only the control variables, while Model 2 adds in the independent variables. We discuss the findings of the control variables in this and all other tables after discussing the findings for the explanatory variables. For hypothesis 1, we had argued that a change in the extent of linkages among knowledge components from one patent to another would enhance the novelty of the subsequent patent. The rationale for this hypothesis stemmed from the notion that changed linkages among knowledge components reflect an alteration in the logic by which organizational members conceptualize and interpret this knowledge, resulting 5 Since we control for number of firm s prior five years patents for any given patent, this value is missing or incomplete for patents that come about before We therefore omit these observations from our sample, resulting in 3856 pairs of patents entering the model. 49

59 in new insights that give rise to novelty. The finding for this hypothesis is observed in Model 2 of Table 3.4. We find a positive and significant coefficient for change in extent of linkages (β=241.07, p<0.01), consistent with our expectations. This finding confirms the role of changed linkages among knowledge components in enhancing the novelty of the subsequent patent. Thus, hypothesis 1 is supported. Hypothesis 2 had argued that a change in the local embeddedness of linkages among knowledge components from one patent to another would enhance the novelty of the subsequent patent. This argument was based on the expectation that the more sophisticated and interdependent context created would lead to more thorough understanding that enhances novelty. Since when the local embeddedness of routines is changed, local context to interpret the linkage within the intertwined routines is altered. It gives linkages new meanings and thus gives rise to novelty. As can be observed in Model 2 of Table 3.4, the coefficient for change in local embeddedness is positive and significant (β=1.43, p<0.05), consistent with this argument. Hypothesis 2 is thus also supported. Table 3.5 shows the regression results for the external use of the patent created by changing the linkage. Model 1 of Table 3.5 is the baseline model, with results for only the control variables, while Model 2 adds in the independent variables. For hypothesis 3, we had argued that a change in the extent of linkages among the knowledge components of one patent to another would reduce the external use of the subsequent patent. We expect that greater change in routines would increase the difficulty of external firms in understanding the nature of these changes or the meaning of their combined effect. Thus, while more changes in the extent of linkages increase novelty, it hurts external firms ability to utilize the patent. The finding for this hypothesis is observed in Models 2 (in which the dependent variable is measured 50

60 in terms of external forward citations) and 4 (in which the dependent variable is measured by number of external citing firms) in Table 3.5. We find that either way of measuring the dependent variable elicits negative and significant coefficients for the change in extent of linkages (β= , p<0.01; β= , p<0.001;), consistent with our expectations. Hypothesis 3 is supported. According to Hypothesis 4, a change in the local embeddedness of linkages among knowledge components from one patent to another will reduce the external use of the subsequent patent. The rationale for this hypothesis centered around the idea that when the context of closely related routines becomes more complicated through increased local embeddedness of linkages, external firms will have more difficulty in understanding how the underlying routines work and thus face more challenges in changing their organizational routines. As can be observed in Models 2 and 4 of Table 3.5 based on alternative measures of the dependent variable, the coefficients for change in local embeddedness are both negative and highly significant (β=-9.68, p<0.001; β=-6.63, p<0.001), consistent with this argument. Hypothesis 4 is supported. Table 3.6 shows the regression results for the advantages of internal versus external use of an innovation created through changed linkage. Model 1 is the baseline model, with results for only the control variables, while Model 2 adds in the independent variables. For hypothesis 5, we had argued that a change in the extent of linkages among knowledge components of one patent to another would increase the advantages of internal versus external use of the subsequent patent. Even though changing routines is very difficult for both the innovating and other firms, we expect that the innovating firm will have an advantage in utilizing the innovation due to their closer understanding of the underlying routines. While it will benefit from some of the 51

61 necessary routines being already in place, other firms would need to start from scratch in this regard. The finding for this hypothesis is observed in Model 2 of Table 3.6. We find a positive and significant coefficient for the change in extent of linkages (β= , p<0.001), consistent with our expectations. Thus, hypothesis 5 is supported. Hypothesis 6 argued that a change in the local embeddedness of linkages among knowledge components from one patent to another will create an advantage for internal versus external firms use of the subsequent patent. The already established routines of the innovating firm serve as the foundation for comprehending new routines, whereas other firms have different routines in place that make interpretations inconsistent with desired meaning. In this regard, a relative advantage is afforded to the innovating firm. As can be observed in Model 2 of Table 3.6, the coefficient for change in local embeddedness is negative and significant (β=-15.71, p<0.001). These findings are inconsistent with our argument. Instead, the findings suggest that external firms have greater relative advantage over the innovating firm in utilizing patents characterized by changed local embeddedness of linkages. While surprising, one explanation is that external firms bypass the complicated routines and come up with their own routines. Since other firms lack the history of creating the innovation, they can come up with a new logic free from the path-dependent approach of the innovating firm. This makes it possible for them to create their own routines for utilizing the innovation. Another explanation for this finding may come from Galunic and Rodan (1998), who point out that very complicated innovations require a tremendous amount of learning and changing of routines. The focal firm may have over-invested in learning and changing its routines, while external firms that did not 52

62 experience the creation process are able to avoid overly complex routines and more easily incorporate the necessary routines. In any case, Hypothesis 1.6 is not supported. Discussion of control variables We controlled for a number of other factors that could potentially influence the relationships of interest. Since firms develop greater competence through experience over time, we controlled for firm age. We see in Table 3.5 that firm age has a negative relationship with novelty. Thus, consistent with other findings in the literature, older firms are seen to achieve less novelty. However, Tables 3.5 and 3.6 show that age is not significantly related to external use, or useful in enhancing advantage for internal use over external use. We had also controlled for number of patents possessed by a focal firm, since this reflects a firm s overall innovative capability. Table 3.5 indicates that number of patents is positively related to novelty, as expected. Table 3.6 shows that it is negatively related to external use, which is also expected since stronger innovative capability of the focal firm reduces the ability of external firms to understand and utilize the patent. The positive relationship between number of patents and internal versus external use, seen in Table 3.6 suggests that the firm that creates the patent is able to benefit from it more than other firms, creating advantage for this firm. We also controlled for knowledge diversity associated with the patent, since these reflect the different areas of knowledge components used for the patents. As seen in Tables 3.4, 3.5 and 3.6, knowledge diversity has a positive effect on novelty, external use as well as advantage for internal use over external use. These findings are consistent with expectations as well, since diverse knowledge has more potential to generate novelty, as well more ways in which it can be utilized by both internal and external firms. 53

63 The maturity of the knowledge associated with the patent is also controlled for since it captures the degree to which knowledge associated with a given patent is already well understood. Table 3.5 shows that knowledge maturity has a positive effect on novelty, reflecting the value of deeper understanding of the knowledge associated with the patent that allows for more novel combinations. On the other hand, Tables 6 and 7 indicate negative effects between knowledge maturity and external use as well as internal versus external use. The finding in Table 3.6 can be explained by the difficulty external firms face in utilizing patents created by changed linkages. However, knowledge maturity has a negative effect on innovating firms advantage over external firms, is unexpected. It is unclear why such firms should benefit less than external firms from knowledge maturity. It may be that, as knowledge matures, there are fewer opportunities to apply this knowledge in new ways. However, other firms may be better able to recognize more opportunities for utilizing this knowledge than the innovating firm, giving them an advantage in this regard. We also control for number of subclasses, since these influence knowledge recombination possibilities. As seen in Table 3.5, number of subclasses has a positive effect on novelty. This is consistent with our expectation that more subclasses provide additional knowledge sources for increasing novelty. Table 3.6 shows that number of subclasses positively influence external use, consistent with the interpretation that such firms are offered more possibilities for recombining and using them. Interestingly, however, Table 3.6 shows that more subclasses do not give the innovating firm an advantage in utilizing the patent. This finding most likely stems from the fact that subclasses in a patent can be clearly detected by external firms and adopted without tremendous difficulty, competing away the advantage that the 54

64 innovating firm might have had. We controlled for the number of inventors on the patent, since it represents varying intellectual complexity associated with the patent. Table 3.5 shows that number of inventors has a positive effect on novelty, as expected, since more inventors offer more intelligence that can enhance novelty. However, as shown in Tables 1.5 and 1.6, it has a negative relationship with external use as well as the difference between internal and external use. These findings suggest that greater intellectual complexity associated with a patent makes both it difficult for both internal and external firms to utilize it. However, the finding that innovating firms have more difficulty in this regard than external firms is unexpected. Finally, we see in Table 3.6 that number of external citing firms positively influences the difference between internal versus external use. The interest that external firms have for the patent reflects its greater potential, which increases the advantage for the innovating firm. In summary, the results confirm that different linkage of components, demonstrated by two salient characteristics the change of extent of linkages and the change of local embeddedness of linkages influence both the novelty and the use of an innovation. CONCLUSIONS AND IMPLICATIONS This study examined the effect of changing linkages of knowledge components on innovation novelty and usefulness using patent data from the pharmaceutical industry. To isolate the effects of changed linkages, we identified pairs of patents of a firm in which knowledge components did not change, but for which the linkages did change. We argued that these changed linkages reflect a 55

65 change in routines that alter the interpretation and meaning of the underlying knowledge components. Our findings indicate that changed linkages do indeed play an important role in the knowledge recombination process. We believe these results shed additional light on the relationship between linkages and knowledge recombination in a number of ways. First, this study helps to address a gap in our knowledge regarding an important but neglected dimension of the knowledge recombination process the linkages connecting knowledge components. It highlights the role of different characteristics of linkages in influencing the novelty and use of an innovation. We studied two such characteristics extent and local embeddedness of linkages. Extent of linkages depicts the degree to which connections exist between knowledge components within an innovation, while local embeddedness reflects the integrated nature of routines. We held components constant in order to isolate the effects of linkages. Our findings indicated that both characteristics have a systematic influence on innovation novelty. In doing so, this research sheds greater light on how innovation is created. We conceptualized linkages among knowledge components as organizational routines, since both influence how organizational members coordinate, understand, and interpret knowledge components. Consistent with routines, linkages constitute the channels by which new meanings are accorded to these components. By the same token, a complex set of intertwined routines creates amplified meaning and understandings that enhance innovation. In this way, this research extends the theoretical understanding for how organizational routines influence innovation. Our methodology allows us to capture the evolutionary basis of routines and understand the manner in which firms update relationships between knowledge components to 56

66 create their innovations. Tracking changes in these linkages over successive innovations provided us with a means to make organizational routines visible and highlight ongoing organizational efforts to bring about innovative accomplishments (Feldman, 2000; Feldman and Pentland, 2003; Becker, Lazarik, Nelson and Winter, 2005). This approach allowed us to view the firm s preservation of past routines and that manner in which they paved the way for future routines (Nelson and Winter, 1982). We also demonstrated that, even though changing routines is very difficult for firms (Nelson and Winter, 1982; Teece, 1982; Edmondson, Bohmer and Pisano, 2001; Feldman and Pentland, 2003; Bresman, 2013; Hargadon and Sutton, 1997), those that are able to successfully do so enhance their ability to create novel innovations. Since firms find it very hard to copy other firms routines (Henderson and Clark, 1990), innovations based on changed routines create advantages for the innovating firm. Routines are not particularly visible, and a change in routines is even trickier to spot by external firms. In contrast, change in components is more obvious, and therefore, easier to copy. Thus, firms that create innovations on the basis of changed routines gain important new understandings that they themselves control, giving them an innovative edge to these firms. In addition, this study uncovered a more complicated relationship between the novelty and the use of an innovation than seen in prior work. Much of the literature suggests a positive relationship between the two (Fleming, 2001; Kaplan and Vakili, 2014), due to the expectation that more novel innovations will be more used. Our findings indicate, however, that the use of an innovation also depends on how novelty is created. We show that, while an innovation created through changed routines is more novel, it is also more difficult for other firms to put into use. By accounting for 57

67 how innovation novelty is created, this research showcases a counterintuitive relationship between novelty and use. Doing so helps us to better understand the importance of the mechanisms by which novelty of an innovation influences its use. This study also suggests a number of directions for future research. In particular, the methodology utilized here paves the way for future investigations related to evolutionary theory, dynamic capabilities and the role of routines in learning. Researchers investigating routines have been hampered by the opaque nature of routines, and the difficulty of identifying them. The approach developed in this research may assist in this regard, as it accounts of the evolution of innovationrelated routines over time. We believe that future research can draw upon this approach to examine other issues relating to routines. One such issue is the influence of continuous and discontinuous change in routines on a firm s learning and performance. It may be that, while continuous routine change may lead the firm to engage in ongoing learning, such learning is more incremental than learning from discontinuous routine change. This latter type of change may lead to more significant novel insights for firms than those that have routinized routine change. In examining linkage change, this study assumed that a change in one routine is similar to a change in another routine, and gave equal weight to changes in all the linkages. However, these may have different value depending on the knowledge components that are linked. It may be that changing more highly established or routinized linkages will have a greater impact on innovation novelty than changing less established routines. It may therefore be worthwhile to compare the effects on innovation stemming from changes in more recently developed routines and those that have been in place for a longer period of time. Similarly, there may also be new theoretical insights gained from considering how sets of routines work together in 58

68 greater detail. While we found that increase in local embeddedness actually enhances novelty, under some conditions such local embeddedness may create undue complexity and actually hamper new insights. Since the characteristics of different neighborhoods of linkages may vary, it may be useful to consider how changes in complexity in these locations influence novelty of the resulting innovation. In addition, since the ability to change routines is the basis of dynamic capabilities, this research provides a basis for investigating the impact of changing routines under differing environmental conditions, including more or less uncertainty. Firms that are able to make appropriate changes in routines could be shown to have greater dynamic capabilities in comparison to those that do not demonstrate significant change despite changing environments. At the same time, researchers may consider what managerial or organizational features allow for such change, or investigate the types of routines that should undergo change while others remain the same. Demonstration of performance differences between firms that are seen to have such dynamic capabilities and those that are not would help to further this literature. This research held knowledge components constant in order to isolate the effects of architectural changes on innovation outcomes. However, prior literature has also stressed the role of changing components for innovation novelty. It is unclear which type of change leads to greater novelty. It would be interesting at this point for future research to consider the relative value for innovation from changed components versus changed linkages. According to Henderson and Clark (1990), significant change in both will lead to radical innovation. Thus, a consideration of their interactive effects is therefore an important future direction of this research stream, especially since linkages do not exist without components or vice versa. A 59

69 more detailed examination of their joint relationship will allow us to develop a closer appreciation for how recombination gives rise to innovation outcomes. It is also possible that, for firms that did not come up with the innovations on their own, their ability to use the resulting innovations depends not only on the understanding and replication of associated organizational routines, as argued here, but also on their motivation for adopting the innovation. For example, the firm s top management team, strategic direction, complementary assets, and other such factors, can all affect whether the firm chooses to adopt a particular innovation. To better understand why firms adopt innovations, future research can incorporate these other factors to gain a more complete picture of innovation use. The methodology developed here for capturing linkages has additional implications for research on routines. For example, future research could use this approach to measure organizational linkages, or that of specific parts of the organization such as a division or group. For example, Yayavaram and Ahuja (2008) use patent component data to measure the knowledge structures of specific firms. Our methodology can be extended to augment this approach and identify routine differences among firms. Since routines are characterized as the basis of firms capabilities (Nelson and Winter, 1982), such an approach will allow us to assess whether more or less complex routines are the basis of superior capabilities, or whether unique knowledge components embedded within the firm is the primary basis of such capabilities. The approach for measuring linkages developed here drew upon Fleming s (2004) measure of coupling and incorporated dynamic updating of linkages over time taking into account every new patent the firm possesses. However, there may be alternative approaches to measuring linkages that may provide other insights. For 60

70 example, rather than tracking the linkages from the entire history of the firm s patenting activity, an adjusted window of time reflecting the period of active use for a particular patent can be used instead. Doing so may highlight the most active or relevant linkages among knowledge components for a firm. 61

71 Figure 3.1 Hypothesized Relationships between Change in Architectural Characteristics and Innovation Outcomes 62

72 63

73 Table 3.1 Numbers of Paired Patents for a Sample of Pharmaceutical Firms Firm Name # Paired Patents Abbott Laboratories 237 Allergan 145 Alza 85 American Home Products 110 Bausch and Lomb 32 Bayer 263 Boehringer 80 Bristol Myers Squibb 210 Chiron 2 Ciba-Geigy 80 Eli Lilly 145 Fujisawa 25 Glaxo 49 Hoechst 111 Janssen 42 Johnson and Johnson 53 Merck 345 Pfizer 139 Pharmacia 25 Procter and Gamble 1376 Rohm and Haas 25 Schering Plough 25 Searle 76 Smith Kline Beecham 3 Roussel 17 Upjohn 32 Warner Lambert 56 Wyeth 71 Yamanouchi 4 64

74 Table 3.2 Variations in Names by which Hoechst s Patents are Filed Filed Name # Patents Hoechst Aktiengesellschaft 3996 Hoechst AG 70 Hoechst Akteingesellschaft 3 Hoechst Aktiengellschaft 2 Hoechst Aktiengsellschaft 2 Hoechst Atkiengesellschaft 2 Kalle, Niederlassung der Hoechst AG Hoechst Aktiengesellschaft 1 Farbwerke Hoechst Aktiengesellschaft 1 Farbwerke Hoechst Aktiengesellschaft 1 vormals Meister Lucius & Bruning Hoechst AG Werk Ruhrchemie 1 Hoechst AK 1 Hoechst Aketiengesellschaft 1 Hoechst Aktiegesellschaft 1 Hoechst Aktiengelselschaft 1 Hoechst Aktiengeselischaft 1 Hoechst Aktiengesell 1 Hoechst Aktiengesellachaft 1 Hoechst Aktiengesellchaft 1 Hoechst Aktiengesellcshaft 1 Hoechst Aktiengeselleschaft 1 Hoechst Aktiengesellschafat 1 Hoechst Aktiengesellschaft Knapsack 1 Hoechst Aktiengesellschat 1 Hoechst Aktiengesellshaft 1 Hoechst Aktinegesellschaft 1 Hoechst Atiengesellschaft 1 65

75 66

76 Table 3.4 Relationship between Linkage Change and Innovation Novelty (Tests for H1 & H2) Model 1 Model 2 Constant *** *** (0.0711) (0.0710) Firm age *** (0.0009) Number of prior patents *** (0.0000) Knowledge diversity *** (0.0203) Knowledge maturity *** (0.0000) Number of subclasses *** (0.0019) Number of inventors *** (0.0017) *** (0.0009) *** (0.0000) *** (0.0203) *** (0.0000) *** (0.0019) *** (0.0017) Δ Extent of linkages ** (76.211) Δ Local embeddedness of linkages * (0.6310) Observations Number of groups Log-likelihood Wald chi2(6) Wald chi2(6), Prob >chi LR test LR test, Prob>= chibar Notes: Standard errors in parentheses. * p<0.05, ** p<0.01, *** p< Year dummies are also included in all models. 67

77 Table 3.5 Relationship between Linkage Change and External Use (Tests for H3 and H4) External use based on external forward citations External use based on number of external citing firms Model 1 Model 2 Model 3 Model 4 Constant * (1.2390) ** (1.2003) (1.1205) (1.0835) Firm age (0.0282) Number of prior patents *** (0.0000) Knowledge diversity *** (0.0277) Knowledge maturity *** (0.0030) Number of subclasses *** (0.0030) Number of inventors *** (0.0037) (0.0274) *** (0.0000) *** (0.0276) *** (0.0002) *** (0.0030) *** (0.0037) Δ Extent of linkage * ( ) Δ Local embeddedness of linkages *** (0.9174) (0.0255) *** (0.0000) *** (0.0445) *** (0.0003) *** (0.0046) (0.0055) (0.0246) *** (0.0000) *** (0.0445) *** (0.0003) *** (0.0046) *** (0.0055) *** ( ) *** (1.4248) Observations Number of groups Log-likelihood Wald chi2(6) Wald chi2(6), Prob >chi LR test LR test, Prob>= chibar Notes: Standard errors in parentheses. * p<0.05, ** p<0.01, *** p<

78 Table 3.6 Relationship between Linkage Change and Internal versus External Use (Tests for H5 & H6) Model 1 Model 2 Constant (1.2179) (1.2897) Firm age (0.0278) Number of prior patents -7.31e-06 (0.0000) Knowledge diversity *** (0.0297) Knowledge maturity *** (0.0002) Number of subclasses *** (0.0037) Number of inventors *** (0.0038) Number of citing firms *** (0.0011) (0.0294) *** (0.0000) *** (0.0297) *** (0.0002) *** (0.0037) *** (0.0038) *** (0.0011) Δ Extent of linkage *** ( ) Δ Local embeddedness of linkages *** (0.9468) Observations Number of groups Log-likelihood Wald chi2(6) Wald chi2(6), Prob >chi LR test LR test, Prob>= chibar Notes: Standard errors in parentheses. * p<0.05, ** p<0.01, *** p<

79 Chapter 4: The Interactive Effects of Changed Linkages and Added Knowledge Components on Innovation Outcomes Researchers have stressed that knowledge recombination is a key way in which new innovations are created (Schumpeter, 1934; Henderson and Clark, 1990; Fleming and Sorenson, 2004). In particular, the adding of new knowledge components to those already known to the firm has been identified as a particularly important means for recombining knowledge, as these new elements can refresh the meanings and functions of the firm s existing knowledge, and introduce novel insights that lead to innovation (Fleming, 2001; Galunic and Rodan, 1996; Katila and Ahuja, 2002). At the same time, the addition of new knowledge components changes the manner in which all components are linked or connected with each other for creating the desired functions (Henderson and Clark, 1990). Since linkages affect how individual knowledge components interact to give rise to a larger meaning, they also play a critical role in the innovation process. Altered or transformed linkages reflects a change in how knowledge components work together to elicit the requisite outcome. Nonetheless, studies examining the addition of new knowledge components in recombination have not accounted for the role of changed linkages in this regard. Ignoring the effects of changed linkages associated with newly added knowledge reduces understanding of how firms accomplish the integration of 70

80 differing knowledge components associated with an innovation. The goal of this research is to address this problem. Linkages among knowledge components (hereafter, linkages) are important as they are the means by which otherwise isolated components are connected (Dosi, Nelson and Winter, 2000; Henderson and Clark, 1990). Connections between otherwise disparate knowledge components are not automatic the same knowledge components can be connected in varying ways, and give rise to alternative meanings. In order to ensure that organizational members link them in a manner that creates the desired meaning, appropriate routines must be in place. Likewise, when knowledge components are newly added, new linkages must also be created as these new components need to be connected to existing ones. These new linkages are also accomplished through the institution of suitable new routines. In this way, linkages are reflected in the routines that must be in place for them to work. Drawing on the argument above, we investigate how change in linkages interplays with the addition of knowledge components in determining the ability of firms, including the innovating firm as well as firms that did not come up with the innovation, to institute and use the innovation. Treating linkages as a reflection of requisite routines, we focus on two central features that capture both general and local contexts of routines: (a) extent of overall connectedness of routines linking knowledge components, and (b) degree of local embeddedness or clustering of routines into local subgroups. Since a change in knowledge components is relatively explicit (in a patent), and thus identifiable, we predict that both innovating and non-innovating firms are equally capable of utilizing an 71

81 innovation created from changed knowledge components. In contrast, change in knowledge linkages and their corresponding routines is much more difficult to identify or understand due to their tacit nature. The need to change already established routines is also problematic (Teece, 1982; Feldman and Pentland, 2003). Thus, change in knowledge linkages is argued to create difficulties for both innovating and non-innovating firms in utilizing the new innovation. Nonetheless, we expect that changed routines will create comparatively more problems for non-innovating firms, as they will find the diagnosis of appropriate routines particularly difficult in comparison to the innovating firm that had developed superior understanding in this regard from the process of creating the innovation. Six sets of testable hypotheses are developed in line with these predictions. The hypotheses are tested using patent data for the pharmaceutical industry over a 38-year period culminating in To isolate the effects of added knowledge components and changed linkages on innovation use, we began by identifying 6,707 firm-specific pairs of prior and subsequent patents in which knowledge components are increased in the subsequent patent. We assessed linkages associated with each patent using the method presented by Fleming and Sorenson (2004) and Yayavaram and Ahuja (2008), and measured change in linkages by subtracting the set of pairwise coupling relationships of the second patent from that of the first. We used network measures to assess changes in extent and local embeddedness of linkages for each pair of patents. The findings using these data strongly support our arguments regarding the critical role of linkages and their associated routines in successfully adopting an 72

82 innovation. In particular, we demonstrate that, while the addition of new knowledge components allows firms to successfully use the resulting innovation, change in extent and non-decomposability of routines that link these new components detract from its successful use. Even so, the negative effects of linkage change are less for the innovating firms than non-innovating firms. These findings shed important light on the important role of linkages in influencing the value of the resulting innovation for firms. In addition, by calling attention to the role played by routines in linking knowledge components, this research underscores the critical importance of organizational imperatives for successfully adopting an innovation. Finally, by highlighting differences in the use of the resulting innovation by innovating and non-innovating firms, this research stresses the relevance of context specificity for routines (Nelson and Winter, 1982). RECOMBINATION THROUGH ADDING KNOWLEDGE COMPONENTS: REVIEW OF THE LITERATURE Within the knowledge recombination literature (Schumpeter, 1934; Henderson and Clark, 1990; Fleming and Sorenson, 2004), it has been widely noted that knowledge components comprise the basic building blocks for innovation (Fleming, 2001; Rosenkopf and Nerkar, 2001). When knowledge components are recombined, new knowledge is created and novel insights afforded (Fleming, 2001; Katila and Ahuja, 2002). The adding of new and previously unused knowledge components to ones that are already known to the firm is particularly beneficial in this regard. While familiar 73

83 knowledge allows the firm to benefit from its deep understanding stemming from prior experience with its use, recombining it with new knowledge extends the firm s ability to leverage it in fresh directions. Recombination of familiar and new knowledge components yields unique insights through their interaction, resulting in significantly transformed knowledge (Simon, 1962; Galunic and Rodan, 1996). New knowledge enhances novelty of an innovation through three basic mechanisms. First, the addition of new and unfamiliar components can increase knowledge diversity, which creates unusual opportunities for recombining knowledge and potential for more novel arrangements (Fleming, 2001; Simonton, 1995; Uzzi, Mukherjee, Stringer, and Jones, 2013). Second, incorporating new knowledge components enlarges prospects for reinterpreting and rejuvenating existing knowledge (Mezias and Glynn, 1993). Although such variance may create wild and unexpected outcomes and increase hazard rates, observed in the lower tail of the whole variance distribution, it also increases the probability for highly novel outcomes, seen in the upper tail (Levinthal and March, 1981; Ahuja and Lampert, 2001). Third, new knowledge broadens the perspectives of an organization, stimulates creative thinking, and helps generate impactful outcomes (Wuyts and Dutta, 2014). Indeed, scholars have also empirically demonstrated that adding new knowledge components in knowledge recombination enhances both the novelty and impact of resultant innovations (Fleming, 2001; Rosenkopf and Nerkar, 2001; Ahuja and Lampert, 2001). For example, Rosenkopf and Nerkar (2001) emphasize the value of knowledge components from new domains, finding that most innovative technologies come from a 74

84 recombination of diverse knowledge components that span technological categories and organizational boundaries. Likewise, Schilling (2005) finds that adding new knowledge components from diverse domains results in atypical connections that allow novel ideas to emerge. Drawing on patent data from 1985 to 1996, Miller, Fern and Cardinal (2005) show that incorporation of knowledge from across divisions of a firm enhances the impact of an innovation. Consistent with these findings, Carnabuci and Operti (2013) find that adding new knowledge components stimulates radical innovations. While the adding of new knowledge is often equated with increased knowledge diversity, other scholars have argued that new knowledge from familiar domains can also enhance innovation novelty, through improved understanding and refinement (Fleming and Sorenson, 2004; Katila, 2000; Mezias and Glynn, 1993). Greater knowledge familiarity, stemming from knowledge components from old domains used recently or frequently (Fleming, 2001), can be valuable because the components selected are better understood and more likely to work together in appropriate ways, and the performance of the created innovations more predictable. For example, Kaplan and Vakili (2014) highlighted the benefits of adding knowledge components from familiar knowledge domains, showing that deep understanding of previously familiar knowledge is necessary for the creation of breakthrough innovations. According to Freeman and Soete (1997), however, drawing upon already known knowledge creates exploitative rather than explorative innovation. Whether from enhanced diversity or deeper familiarity, the literature on knowledge recombination has paid a great deal of attention to the value of adding 75

85 knowledge components for innovation. In contrast, despite acknowledgement that linkages between knowledge components also play a role in enhancing innovation outcomes (Schumpeter, 1934; Henderson and Clark, 1990), there has been little attempt to consider the effects of linkages from adding knowledge components. This is a concern since the value of knowledge components can be altered by changes in knowledge linkages. We discuss the significance of this issue below. Knowledge linkages are defined as the pattern of connections between all knowledge components that affect their role in delivering a well-defined function. The manner in which new and existing knowledge components are coordinated and combined with existing ones affect how they operate together and deliver desired outcomes (Henderson and Clark, 1990). There are two major reasons to study the role of linkages. First, they serve as connectors of knowledge components, establishing how these components are jointly conceptualized and applied to create particular outcomes. As Henderson and Clark (1990) have pointed out, even when knowledge components remain the same, variations in linkages can lead to different meanings and outcomes (Henderson and Clark, 1990). Altering the linkages between knowledge components lead to reconceptualization of their role in creating the innovation (Gentner, 1983; Huber, 1991; Kaplan and Simon, 1990). Thus, above and beyond the adding of knowledge components, change in linkages can result in novel insights. Second, since the adding of knowledge components to existing ones automatically changes linkages, due to the fact that they all need to function as an integrated system. Since linkages can differ across firms (Carnabuci and Operti, 2013; Henderson and Clark, 76

86 1990), it is clear that implementing knowledge recombination is a deliberate process and involves design of how new components are connected with existing ones such as how many linkages are constructed and how strong these linkages are to create generate novel outcomes. Thus, the co-occurrence of changed components and linkages requires their joint study. In all, the role of knowledge linkages in the recombination process should not be ignored, since they play an important part in determining the value of new components for innovation outcomes. At this time, however, little effort has been devoted in the literature to explaining the effect of linkages when recombination through the addition of new knowledge components occurs. There remains a need for a more systematic theoretical explanation for how linkages actually operate in this regard. In the section below, we argue that linkages among knowledge components are the manifestations of organizational routines. Only by changing routines can linkages be altered, which creates problems in their accomplishment. LINKING KNOWLEDGE THROUGH ORGANIZATIONAL ROUTINES Knowledge components the basis for knowledge recombination are stored within an organization as disparate knowledge pieces in potentially different locations. As isolated entities, they have little value due to their inability to offer unique insight. If integrated with other knowledge components in a creative and advantageous way, they have the potential to create amplified meaning which the organization can leverage for accomplishing its goals and tasks. How, then, does an organization accomplish the 77

87 linkage of discrete and dispersed pieces of knowledge? We argue that the active linking of knowledge components occurs through organizational routines. Organizational routines are patterned sequences of actions that accomplish organizational tasks (Cyert and March, 1963; Teece, Pisano and Shuen, 1997). They are multi-actor, in that different organizational members can accomplish different parts of a routine, or participate in different but connected routines. Most critically, the accomplishment of routines requires tacit skill and know-how to combine specialized knowledge in a way that creates desired outcomes (Nelson and Winter, 1982). The routines establish the cognitively embedded sense-making process that allows organizational members to conceptualize and interpret knowledge necessary for completing their tasks (Cohen and Bacdayan, 1994; Pentland and Reuter, 1994). In this regard, organizational routines provide imperatives to connect knowledge components that are otherwise isolated, and inculcate the way in which they should be combined. Since routines are path-dependent and develop in evolutionary ways (Nelson and Winter, 1982), they tend to be firm-specific and reflect dissimilar capabilities for linking knowledge. Thus, we argue that the underlying mechanisms of knowledge linkages are organizational routines, in the absence of which, disparate pieces of knowledge have no means by which they can be connected. These routines determine the role knowledge linkages play in knowledge recombination. The adding of new knowledge components requires changing of linkages this in turn mandates a change in underlying routines. Researchers point out, however, that changing routines is difficult to accomplish for organizations (Cohen and Bacdayan, 1994; 78

88 Cyert and March, 1963; Nelson and Winter, 1982; Nystrom and Starbuck, 1984), due to habituation. Thus, we point to a critical conundrum for using recombination while the adding new knowledge components can enhance the firm s access to an innovation, the need to change existing routines creates an obstacle for its implementation. Below we develop this argument taking into account two types of changes in routines: (a) change in the overall extent of routines, and (b) change in the local embeddedness or degree to which routines cluster into local subgroups (Nelson and Winter, 1982; Teece, 1982). We argue that, due to their context specificity and tacit nature (Cyert and March, 1963; Nelson and Winter, 1982), the identification and installation of these changed routines will pose difficulties for firms, but more so for noninnovating firms those that do not directly participate in the creation of an innovation but intend to utilize it than the innovating firm. We derive six sets of hypotheses in line with this argument. Relationship between adding new knowledge components and innovation use According to scholars, recombining new knowledge components with ones already known to or previously used by the firm can be beneficial for creating innovation (Ahuja and Lampert, 2001; Fleming, 2001; Kaplan and Vakili, 2014; Rosenkopf and Nerkar, 2001; Taylor and Greve, 2006). Adding knowledge components from new and unfamiliar knowledge domains can enhance the value of innovations by creating more possibilities for knowledge recombination and thus more novel arrangements (Fleming, 2001; Simonton, 1995). Indeed, the more new knowledge components added in this 79

89 regard, the more options that are created for intrinsic new content that can facilitate the emergence of creative insights (Levinthal and March, 1981; Ahuja and Lampert, 2001; Wuyts and Dutta, 2014). That is, combining the new components with existing ones allows for the reconceptualization and reapplication of familiar components in a manner that enhances the potential for innovation (Fleming, 2001; Fleming and Sorenson, 2004). The change in knowledge components is reasonably explicit andobservable for both innovating and non-innovating firms. For the firm that developed both the original and the subsequent innovation through the recombination of new and old knowledge components, the ability to internally implement the new innovation is likely to be straightforward. Since organizational members are already familiar with the old knowledge components, fostering an understanding of the new components and incorporating them into firm practices will be an incremental process. Thus, we expect that the innovating firm will be able to successfully implement a new innovation that involves recombination of new with old knowledge components. For non- innovating firms as well, we expect the addition of new knowledge components recombined with existing ones to be relatively easy to identify, as they can compare the patent with the prior ones on which it is built to assess how knowledge components have changed. Even though they did not create the innovation themselves, the non-innovating firms can still identify and acquire the new components necessary for utilizing the innovation. Appreciation for how the new components work can result in these firms developing their own ways of recombining them to create novel innovations (Henderson and Clark, 1990; Wuyts and Dutta, 2014). 80

90 In line with these arguments, we propose the following hypothesis: H1: For both innovating and non-innovating firms, addition of new knowledge components is positively related to use of the resulting innovation. The moderating role of changed extent of routines In knowledge recombination, newly added components have to be linked to existing ones in systematic ways for the resulting innovation to deliver its function. The addition of new components will require a reworking of linkages across all knowledge components in order for their effects to be properly realized. Thus, the value of adding components is contingent upon the changed pattern of linkages in the new innovation. In particular, we expect that integrating new components with old ones will tend to increase the extent of linkages between components, due to the need for new connections. Likewise, in implementing the innovation in the organization, we expect that this change in linkages will be reflected in changed organizational routines that facilitate the connection of these newly added components with ones already existing within the organization. As noted by Dosi, Nelson and Winter (2000), an organization s routines are the means by which an organization perceives, reasons and coordinates relationships among differing knowledge components. In this case, the underlying routines that link new and existing knowledge components are the means by which members can understand how they coordinate and work with each other ((Galuic and Rodan, 1996; Henderson and Clark, 1990). For both the innovating and non-innovating firms, it is 81

91 important to understand the nature of organizational routines that must be implemented for effective use of the innovation. In particular, it is vital for them to change their routines in a manner that ensures that the new knowledge components are appropriately used and the value of the new innovation is realized. At the same time, it must be remembered that, once established, routines are generally difficult to alter (Cohen and Bacdayan, 1994). A firm s established set of routines serves as a blueprint for how it organizes its activities (Nelson and Winter, 1982), exercised as a programmed gene or habit embedded in organizational members minds (Stene, 1940, Simon, 1976; March and Simon, 1958; Cyert and March, 1963; Gioia and Poole, 1984; Carley, 1996; Carley and Lin, 1997). As such, routines persist over time (Cohen and Bacdayan, 1994; Galunic and Rodan, 1998), and changing them requires a significant investment of time and resources (Edmondson, Bohmer and Pisano, 2001; Feldman and Pentland, 2003; Bresman, 2013). For example, Teece (1997:1335) notes that departure from routines will lead to heightened anxiety within the organization, while Nelson and Winter (1982: 134) point out that attempts to change routines often provoke a renewal of conflict. Drawing on this logic, we expect that the altered routines resulting from a change in linkages will be difficult for both innovating and noninnovating firms to accomplish. The inability to do so will impede both from fully understanding, incorporating, and using this innovation. 82

92 In line with the above, we present the following hypothesis: H2a: For both innovating and non-innovating firms, change in extent of linkages reduces the positive effect of adding knowledge components in using the resulting innovation. Even though the implementation of new routines necessary for using the innovation will be difficult for both innovating and non-innovating firms, they will not be equal in this regard. In particular, we expect the innovating firm to have an advantage over non-innovating firms in implementing the necessary routines. Since the innovating firm created the new innovation by adding new components to existing ones, it already has deep and tacit understanding of how the original components and associated routines function. The additional components and routines represent incremental rather than radical learning in this regard. Compared with the innovating firm that created the innovation, such altered routines are harder to spot for non-innovating firms, since they did not directly participate in their creation. While the innovating firm is able to go down the learning curve in this regard, non-innovating firms will find it difficult to assess the nature of routines necessary for implementing the innovation. Insufficient understanding of why it is necessary to add or change routines will deter these firms from fully appreciating the transformed meanings shaped by the new routines. The underlying mechanisms through which these nuances were created may remain unclear to them. As a result, the pathdependent nature of routine change favors the innovating firm while hampering other firms from effectively applying the resultant innovation. 83

93 This argument leads to the following hypothesis: H2b: With added knowledge components, change in extent of routines gives the innovating firm relative advantage over non-innovating firms in using the resulting innovation. The moderating effect of local embeddedness of routines While we noted above that the overall extent of linkages can be affected when new components are linked to existing ones to create an innovation, local characteristics of linkages can also undergo change. Local embeddedness of linkages among knowledge components refers to the degree to which linkages cluster together as a neighborhood, interacting with each other as a cohesive subgroup to deliver a function (Nelson and Winter, 1982; Teece, 1982). The higher the local embeddedness of linkages, the greater need exists for multiple routines to operate together in order to create the necessary meaning from components. In other words, individual routines alone are not adequate for generating the amplified meaning necessary for the resulting innovation. Such a context of routines reflects tacit organizational knowledge, since meaning is derived from a complex set of intertwined routines that are not easily extricated. Locally embedded routines creates a context in which any given routine helps to generate greater value compared to when it operates alone. For any given organization, the formation of these neighborhoods is grounded in the unique way in which it interprets and conceptualizes meanings and functions of knowledge components. Such a closely connected and coherent set of routines facilitate highly complex knowledge flow among components. 84

94 Since these routines contain highly tacit knowledge, they cannot effectively stand alone (Teece, 1982). In contrast, a routine that is not locally embedded with other routines can operate more or less independently in delivering the desired function. The addition of new knowledge components may change local embeddedness of routines required to operationalize the innovation (Nelson and Winter, 1982; Teece, 1982; Kogut and Zander, 1992; Grant, 1996). In particular, an increase in local embeddedness indicates that groupings of organizational routines play an enhanced role in correctly interpreting meanings of components and delivering their functions. This is accomplished by additional new routines that work with other routines to deliver the innovation. Increased complexity and tacitness of knowledge required for utilizing the innovation is reflected in such conditions. The rich meanings and functions of these newly added components may be embedded understanding of why a routine functions as it does in that context (Galunic and Rodan, 1998: 1198). It suggests that appreciating and implementing locally embedded tacit routines is key for accurately applying the meanings and functions associated with the resulting innovation. We would thus expect that this increased complexity and tacitness of routines will impede the ability of both the innovating and the non-innovating firms to effectively use the innovation or fully realize its value. Thus, we hypothesize the following: H3a: For both innovating and non-innovating firms, change in local embeddedness of routines reduces the positive effect of adding knowledge components in using the resulting innovation. 85

95 While local embeddedness of routines will adversely affect firms ability to implement the innovation, we do not expect this effect to be uniform across innovating and non-innovating firms. This is due to the fact that the innovating firm already has experience with some of the locally embedded routines from implementing earlier innovations, thus gaining direct experience and first-hand knowledge about the ways in which the new and existing routines work together. As a result, even though increased local embeddedness enhances the complexity and tacitness of routines in the resulting innovation and makes it difficult to fully realize the value of the newly added components, this prior experience and knowledge can help the innovating firm alleviate some of these difficulties. In contrast, the non-innovating firms that did not participate in the creation of this innovation are likely to encounter difficulties in comprehending the meanings associated with a change in local embeddedness. These firms will find it hard to diagnose the coherence of the interacting set of routines or appreciate the nature of these embedded routines, which were developed in relation to another firm s organizational context (Galunic and Rodan, 1996). Even if these other firms are able to understand the rich functioning of local neighborhoods of routines, they may not be able to change their own well established routines adequately in order to adopt and absorb new ones. The development of routines involves a path-dependent process (Nelson and Winter, 1982), which results in firm-specific routines or ones that are unique to a given firm. As a result, the innovating firm can have a relative advantage over the non-innovating firms in 86

96 implementing new locally embedded routines related to the new innovation. Following this logic, we propose the following hypothesis: H3b: With added knowledge components, change in local embeddedness of routines gives the innovating firm relative advantage over noninnovating firms in using the resulting innovation. METHODS Sample To test our hypotheses, we use patent data from the pharmaceutical industry, a research setting that has been commonly adopted in the extant technological innovation literature (Guler and Nerkar, 2012; Levin et al., 1997). It has been widely acknowledged that patents in the pharmaceutical industry appropriately represent firms innovative efforts because they are all motivated to apply for patents to protect their intellectual property. Moreover, almost all players in this industry patent at the United States Patent and Trademarks Office (USPTO) (Nerkar, 2003). The USPTO patent database provides a great deal of information on each patented innovation, including application and approval dates, classes, subclasses, claims, and other useful information. The USPTO database uses subclasses to identify the specific technologies associated with each patent; the approximately 100,000 such subclasses allow for fine-grained classification of key knowledge components involved in a patented innovation (Fleming and Sorenson, 2004). Since the whole assigning process is conducted by USPTO, the information gained 87

97 through examining subclasses tends to not be biased at the firm level (Carnabuci and Operti, 2013). As a result, subclass information is an ideal approach for studying knowledge recombination (Fleming, 2001; Fleming and Sorenson, 2004). Hereafter, we refer to subclasses as knowledge components or simply components. By observing the entire patenting history of a firm, we are able to identify historical connections between components of each patent on the basis of all prior patents. Accordingly, we use an evolutionary approach to capture the linkages of each of the firm s innovations. Because the pharmaceutical industry is highly consolidated, we follow previous research and collect data on patents from the 33 largest firms that comprise a total of 90 percent market share (c.f., Guler and Nerkar, 2012). Since our study strives to understand the effects of the addition of new knowledge components and the commensurate change in knowledge linkages, it is necessary to identify a pair of patents where a second patent is created from adding new knowledge components to an earlier patent. After examining all patents granted by USPTO between 1975 (the first year in which full-text records were available) and 2014, we were able to identify 6,707 pairs of firm-specific patents of this kind on the basis of the backward citations of each patent. Since we are interested in the change in knowledge recombination occurring within a firm, we only consider the backward citations that were patented by this firm. In this way, we are able to construct pairs of successive firm-specific patents created through adding of new subclasses and linkages. As can be seen from Table 4.1, among the 33 largest pharmaceutical firms, Procter and Gamble has the most pairs of patents with 1705 followed by Bayer with

98 and Abbott Laboratories with 480, and Yamanouchi with the fewest at 5 pairs. The prominent differences in patent pairs among firms reflect the variance in the innovativeness of these firms. Interestingly, when submitting patents to USPTO, firms sometimes use different formats for their names or make spelling errors. Therefore, the ambiguity of firm names is a major problem when assigning a firm to a patent (Li et al., 2014). To address this problem, this study adopted fuzzy matches to capture all the different formats used for firm names in the USPTO database. Dependent variable Innovation use. The use of a patented innovation is measured by its forward citations (Fleming, 2001; Fleming et al, 2007, Yayavaram and Ahuja, 2008). Since every patent is built upon prior patents, the creation of a patent is a path-dependence process utilizing prior knowledge. The more forward citations of a patent, the greater is its role in creating future patents. To measure external use of a patent, or use of an innovation by firms other than the one that created it, we count the citations of the patent made by these other firms. To measure internal use of a patent, or the use of an innovation by the firm that created it, we count the forward citations of the patent made by this firm. This study calculates the forward citations for all patents until June When a patent is not externally cited by many firms, but the ones that do cite it a lot, it still results in high external citations. To better reflect the general usefulness of the patent, we also measure external use by taking account of the number of external firms that cite the patent. We use both measures in our analyses. 89

99 Independent variables Addition of new knowledge components. As in prior research (Carnabuci and Operti, 2013; Fleming, 2001), knowledge components are captured in this study by the subclasses co-assigned to each firm s patented innovations. The patents in our sample have a total of 41,024 different subclasses representing knowledge components in our study. The addition of knowledge components is the number of subclasses that are added to the subsequent patent relative to the prior patent. Measure of linkages/routines. We conceptualize a linkage between knowledge components in a manner similar to coupling by other scholars (e.g., Fleming and Sorensen, 2004; Yayavaram and Ahuja, 2008). A linkage reflects how two components coordinate with each other in order to work appropriately. If the linkage between these components is strong, we expect their underlying routines to be closely connected, enhancing their joint operation and leading to an amplified meaning. In order to assess the strength of a linkage, we measure the interdependence between components. We draw on Fleming and Sorenson (2004) who note that a strong linkage between components creates high interdependence among them that will not make it easy to recombine with other components. In contrast, when interdependence is low, the addition of any new components is more easily accomplished, and allows for change in established routines. As a result, the strength of a linkage between two components is reflected in the ease of recombining the two components with other components. 90

100 Drawing on the above, we measure the linkage, or degree of interdependence, between each set of two components by examining the number of other components being co-listed with them in all the prior patents before the focal patent. For example, to measure the linkage connecting component i and j, we count the number of other components recombined with component i and j in the entire set of the firm s prior patents. The greater the number of components that have been previously combined with components i and j, the greater will be the ease (or lower the difficulty) of recombining these two components, and the lower is the strength of the linkage between them. Next, we account for the likelihood that knowledge components vary in the number of purposes for which they can be used. Components that have more purposes will be used in more patents and thus will be employed more frequently. To address this issue, we divide the above by the number of patents that include both components i and j.6 The formula for calculating a linkage between two components in a patent is as follows: Linkage or coupling between components i and j =1/ [!"#$%!"!"#$%&!!'!!"#$%&'()*!"#$%&'(!"#$!"#$%&!!'!!!"#!!"#$%!"!"#$%&'(!"#$%#&!"!"#$!"#$%&!!'!!!"#! ] 6 While the measure outlined here is similar to the coupling measure outlined in Fleming (2004), it also differs in a number of ways. First, while Fleming s measure calculates the coupling for an entire patent by taking the average value of the coupling for every subclass, the focus here is on the tie that exists between each set of two components within a patent. Second, Fleming s study conducts crossindustry level analyses, whereas the unit of analysis for this study is the firm level. Third, and most importantly, Fleming s measure is static, capturing a one-time snapshot of coupling. This paper calculates linkages in a more dynamic way, in which couplings between components have different values over time to account for past experience in how its knowledge is recombined each time the firm successfully files a patent. For example, since Bayer has filed 12,720 patents, its knowledge recombination will change as many times. 91

101 Finally, we calculate the strength of each linkage associated with any two components in a patent and construct a linkage matrix for a patent incorporating the values for every linkage within it. We then construct a linkage matrix for every patent possessed by a firm in which all the components are treated as nodes of a network and linkages as ties. Treating it as a weighted adjacent matrix for a network, this linkage matrix of a patent is the foundation for deriving the two salient characteristics of linkage studied here the extent and local embeddedness of linkages. The calculation of each linkage of every patent requires a very dynamic approach, since the strength of a linkage in any point of time is determined by its most recent value. Measuring linkages in this way allows us to better capture the dynamic way in which a firm develops linkages and designs the routines that recombine knowledge components. The amount of computer processing time required to calculate each and every linkage in all the patents is tremendously high, taking months of CPU time. Change in extent of linkages. We use the average degree of centrality of each knowledge components to measure the extent of linkages in a patent. Degree centrality refers to the total number of its direct connections with other nodes in the network, capturing the general connectivity of the nodes with others (Carpenter, Li, and Jiang, 2012; Scott, 2000). As noted earlier, we identified pairs of prior and subsequent patents. To assess change in the extent of linkages, we take the real value of the difference between the extent of linkages in the subsequent and prior patents. Change in local embeddedness of linkages. We measure local embeddedness through the degree of clustering, which allows us to capture how closely tied a linkage is 92

102 to others in its neighborhood (Yayavaram and Ahuja, 2008). Local embeddedness elicits continuous values, with larger values indicating greater local embeddedness. To capture change in local embeddedness of linkages, we examine the local embeddedness between the prior and subsequent patents. We measure the real change between the local embeddedness of linkages between the two. Control variables We control for a number of contextual features associated with a patent that might influence the relationships we are studying. One is the number of inventors, since a larger number of inventors could represent more knowledge and lead to more forward citations for the patent they create (Singh and Fleming, 2010). A second is the number of subclasses, since more subclasses indicate more diverse knowledge and thus may enhance innovation novelty (Fleming, 2001; Rosenkopt and Nerkar, 2001). In addition, knowledge maturity may also influence the innovation outcome and its use, since mature knowledge is easier to understand and could be more easily utilized by firms (Sørensen and Stuart, 2000). Based on previous studies (Agarwal and Hoetker, 2007; Lanjouw and Schankerman, 2003), we assess technology maturity of a patent by the number of prior arts within this patent. Knowledge diversity can also influence the dependent variables (Argyres and Silverman, 2004; Jaffe et al., 2002; Singh, 2008), since the more diverse the range of knowledge, the more areas a patent touches and more potential uses it can generate. Knowledge diversity is calculated by the proportion, pi, within a patent s backward citations that belong to technology class i. The formula for this variable is 93

103 p 1 i i 2, with values falling between 0 and 1. We also control for the number of prior patents that the firm has to indicate their innovation capabilities, since different capabilities may influence both innovation novelty and use (Kogut and Zander, 1992; Teece, Pisano and Shuen, 1997). This variable is measured by the number of patents that a firm has applied for in a five-year window prior to the focal patent. Thus we have 3856 pairs of patents entering the model. Firm age is also controlled for, since innovation capabilities develop over time. We also control for patent age since this influences the level of forward citations associated with the patent. Note, however, that the negative binomial model analysis automatically omits patent age in the regressions due to its high relation with firm age. In addition, we controlled for the number of claims of a patent. The novelty of a patented innovation is indicated in the number of claims within the patent. A claim is a statement listed within the patent noting a new element of the patent reflecting knowledge previously unknown to the field (Beaudry and Schiffauerova, 2011). More claims associated with a patent reflect greater novelty or newness of the innovation (Duchesneau Cohn and Dutton, 1979; Hage, 1980; Ettlie, Bridges and O Keefe, 1984; Dewar and Dutton, 1986). Methods Because our dependent variables novelty (number of claims), external and internal use of an innovation (number of external and internal forward citations) are count variables, a Poisson or negative binomial estimation method is used to conduct the 94

104 analysis. Since our independent measures are change-related measures, a random effects model is appropriate (Yayavaram and Chen, 2013). We therefore use the negative binomial estimation method with random effects to conduct the analysis. We also perform a Poisson estimation with fixed effects for robust checks. The negative binomial function is takes the following form: Γ(y + k) " k % P(Y = y X, k) = $ ' Γ(k)Γ(y +1) # k + µ & The expectation of Y is: E(Y) = µμ Applying a log link, we get: g(µμ)=log(µμ)=α + βx where X captures the independent variables including change in extent of linkages, change in local embeddedness of linkages, and control variables; y relates to the dependent variables including novelty, external forward citations and internal minus external forward citations. The equations examined take the following form (with differing dependent variables): Use of resulting innovation= α! + β! added knowledge components + β! added knowledge components increased extent of linkages + β! added knowledge components change in local embeddedness of linkages + controls+ e k y " µ % $ ' # k + µ & 95

105 RESULTS AND DISCUSSION Table 4.2 shows descriptive statistics and correlation for all the variables in this research. Due to the relatively high correlation coefficients across major independent variables, we examined the uncentered VIFs in our models. Results show that the average VIF value was 3.75 (the maximum VIF was 5.62), indicating that multicollinearity is unlikely to be of concern. Furthermore, albeit their relatively high correlations, these independent variables still remain significant in the analyses, thus indicating that they are reasonably distinct constructs. Table 4.3 presents the results of the random-effect negative binomial estimation for innovation use, including three models to test different hypotheses. Each model in Table 4.3 includes two panels reporting the results for the internal and external use of a resulting innovation respectively. Model 1 reports the baseline models including only control variables. Model 2 reports the results of the main effect models, which include the three major predictors, i.e., the addition of new components, the change in the extent of linkages, and the change in local embeddedness. Model 3 presents the two interactions between the addition of new components and the changes of knowledge linkages. Table 4.4 reports the results for the difference between internal and external use of a resulting patent. Models 1 through 3 of Table 4.4 respectively present the baseline model, main effect model, and the interaction model regarding the results for the difference of a resulting innovation s internal and external use. For Hypothesis 1, we had argued that the addition of new knowledge components in the resulting innovation would encourage both the innovating firm and the non- 96

106 innovative firms to use this innovation. The rationale for this hypothesis stemmed from the expectation that both types of firms could easily identify the change in components and thus incorporate their use when implementing the innovation. Indeed, the results in Model 2 of Table 4.3 show that the addition of new knowledge components in a resulting patent has a positive and strongly significant relationship to its use by the innovating firm (β=19.83, p<0.001) as well as the non-innovating firms (β=6.05, p<0.001). Thus, hypothesis 1 is strongly supported. Hypothesis 2a predicted that increased extent of linkages, reflecting greater need for new routines associated with the innovation, would weaken the positive effects of the addition of knowledge components on use by innovating and non-innovating firms. Our argument for this relationship rested on the notion that the diagnosis and implementation of relevant routines would be difficult for both types of firms due to their tacit nature. Results in Model 3 of Table 4.3 show that the interaction between the addition of knowledge components and the increase in the extent of linkages in a resulting patent has a highly significantly negative relationship to use of this patent by both the innovating firm (β=-1.77, p<0.001) and other firms (β=-13.19, p<0.001). Hypothesis 2a is thus also strongly supported. Hypothesis 2b predicted that the influence of increased extent of routines is stronger for non-innovating firms than for the innovating firm. We expected that the innovating firm already possessed some of the relevant routines, and that the implementation of new routines would be relatively easier for them than for other firms that had no routines already in place. The results for this hypothesis are in Model 3 of 97

107 Table 4.4, and indicate that the interaction between added knowledge components and increased extent of linkages is positively related to the difference between internal and external use (β=1.70, p<0.10). Hypothesis 2b is thus marginally supported. Hypothesis 3a predicted that increased local embeddedness of routines would attenuate the positive effects of adding new components for both innovating and noninnovating firm use of this innovation. This relationship was predicated on the difficulty of understanding and implementing such nested sets of routines. Results in Model 3 of Table 4.3 show that the interaction between added new components and increase in local embeddedness is highly significant and negatively related to both innovating firm (β= , p<0.001) and non-innovating firm use (β=-25.86, p<0.001) of an innovation. Hypothesis 3a is supported. According to hypothesis 3b, increased local embeddedness of routines when adding knowledge components will give the innovating firm an advantage over noninnovating firms in implementing the innovation. The rationale for this relationship rested on the increased knowledge of the locally embedded routines on the part of the innovating firm relative to the non-innovating firms. The results for this hypothesis are seen in Model 3 of Table 4.4. We find the interaction term to be positively related to the difference between innovating and non-innovating firm use (β=2.25, p<0.05), as predicted. Hypothesis 3b is supported. With respect to our significant control variables, Model 1 of Table 4.3 shows that the age of an innovation is positively related to both internal (β=44.85, p<0.001) and external use (β=69.81, p<0.001), as expected, while the number of prior patents of this 98

108 innovation is negatively related to internal (β=-24.63, p<0.001) and external use (β= , p<0.001). The number of inventors involved in the innovation is negatively related to its internal use (β =-7.82, p<0.001), but can encourage external use (β=8.35, p<0.001). Furthermore, the number of claim of an innovation, which reflects its novelty (Beaudry and Schiffauerova, 2011), can encourage both the innovating firm (β=15.42, p<0.001) and non-innovating firms (β=47.50, p<0.001) to use this innovation. An interesting finding is that the increased knowledge diversity of a resulting innovation can significantly discourage non-innovating firms from using it (β=-8.19, p<0.001), but not the innovating firm (β=-0.54, n.s.). While prior literature has focused solely on the value of new knowledge components for innovation, this research highlights the vital importance of accounting for the change in knowledge linkages that occurs at the same time, with all hypotheses supported. Consistent with the literature, we find that new knowledge components enhance the use of an ensuing innovation by both the innovating and non-innovating firms. However, we also demonstrate that linkage changes that accompany the change in components cause problems in implementing the innovation. We argue that these problems stem from the inability of firms to institute the necessary new routines. In particular, both change in extent and non-decomposability of routines play a consistent role in this regard. At the same time, our findings show that, unlike incorporation of new components, problems in implementing the necessary routines are not equivalent for both innovating and non-innovating firms. Rather, the innovating firm benefits from a priori knowledge relating to the previous innovation and the fact that some of the necessary 99

109 routines are already in place. Since non-innovating firms do not have prior experience with these knowledge components and their implementation, they are at a relative disadvantage in this regard. These findings support our argument that linkages are instituted as routines within the organization. Consistent with the tacit nature of routines (Nelson and Winter, 1982), we find that linkages are difficult to institute and can impair the ability to the firm to benefit from an innovation. CONCLUSIONS AND IMPLICATIONS The purpose of this study was to examine the joint effects of newly added knowledge components and changed routines on the ability of firms to utilize the resulting innovation. We pointed out that the adding of knowledge components, a key focus of recombination and innovation researchers, co-occurs with changing of linkages between knowledge components. However, the joint effects of linkages have been largely ignored in the literature. Since the meaning of knowledge components can be altered through the manner in which they are linked to each other, linkages play a critical role in recombination outcomes. We argued that knowledge linkages are enacted through organizational routines. Thus, any change in knowledge linkages requires a corresponding change in organizational routines, allowing for different conceptualization, interpretation and reasoning of knowledge components. Using data on patents from the pharmaceutical industry, our findings confirm the positive effect of adding knowledge components to the use of the innovation. More importantly, our findings indicate that knowledge linkages play a significant negative role 100

110 in the relationship between the addition of knowledge components and innovation use. That is, appropriate linkages are difficult to institute and impair the use of the innovation. In addition, we find that the negative effects of knowledge linkages diminish for the innovating firm relative to non-innovating firms, reflecting the value of the innovating firm s prior experience. This research contributes to the literature on knowledge recombination in a number of ways. One is the additional light shed on the relationship between added knowledge components and innovation. In highlighting the crucial but neglected role of knowledge linkages for knowledge recombination, we demonstrate that the addition of new knowledge components is not the sole basis for innovation value. The manner in which they are linked is also demonstrated to be vitally important in this regard, since they can detrimentally affect the ability of firms to use the innovation. In this way, this research stresses the need to consider not only knowledge components but also the knowledge linkages that play a key role in creating the novel insights associated with the innovation. Another contribution is the theoretical connection pointed to between knowledge linkages and organizational routines. Although other researchers have also noted the embeddednesss of linkages in organizational communication channels (e.g., Henderson and Clark, 1990), we go further in this regard by showing the intrinsic connection between the two concepts. We highlight how, similar to linkages, routines serve to connect disparate and otherwise unconnected knowledge attributes, and play an essential 101

111 role in the sense-making process (Nelson and Winter, 1982). In doing so, this research provides a theoretical mechanism for the role of linkages in knowledge recombination. The differential effects of knowledge linkages on creating relative advantage for innovating firms over non-innovating firms are also a new insight in this literature. We argued that the context specificity of organizational routines would have different implications for these two types of firms. Because the innovation firms came up with the innovation, it already possesses some of the necessary organizational routines for utilizing the resulting innovation. Through the creation process, the innovating firm understands how these organizational routines function and thus finds is relatively easy to change its routines for the innovation. In contrast, the non-innovating firms did not experience the creation process, which make it difficult to reason how the routines function together to deliver the desired outcome. In particular, when the underlying routines are highly intertwined and complicated, non-innovating firms have a great deal of difficulty in figuring out how these highly interdependent routines work hand in hand in applying the resulting innovation, thus giving relative advantage to the innovation firm. In this way, this study extends the literature of organizational routines and innovation by highlighting different effects with respect to innovating and non-innovating firms. This study also suggests directions for future research. In examining linkage change, this study assumed that a change in one routine is similar to a change in another routine, and gave equal weight to changes in all the linkages. However, these may have different value depending on the knowledge components that are linked. It may be that changing more highly established or routinized linkages will have a greater impact on 102

112 ongoing organizational routines than changing less established routines. It may therefore be worthwhile to compare the effects on innovation stemming from changes in more recently developed routines and those that have been in place for a longer period of time. In addition, the methodology utilized here paves the way for future investigations related to organizational routines. Researchers investigating routines have been hampered by the opaque nature of routines, and the difficulty of identifying them. The approach developed in this research may assist in this regard, as it captures the ongoing routines and their changing patterns using the linkages across different knowledge components. We believe that future research can benefit from this approach by generalizing it to examine other issues related to routines. In this study we mainly focused on two central features of routines extent and local embeddedness and showed that they function similarly in moderating the benefits of new knowledge components. However, given their different characteristics, these two features may have other distinct implications for the knowledge recombination process. For example, we found different main effects of these characteristics of routines on internal and external use of a resulting innovation (see Model 2 in Tables 4 and 5). Future research can benefit from further exploring the difference between these two structural features of knowledge linkages. Similarly, another possible extension to this study is to explore more structural features of knowledge linkages. Lastly, since the ability to change routines is the basis of dynamic capabilities, this research provides a basis for investigating the impact of changing routines under differing environmental conditions, including more or less uncertainty. Firms that are 103

113 able to make appropriate changes in routines could be shown to have greater dynamic capabilities in comparison to those that do not demonstrate significant change despite changing environments. At the same time, researchers may consider what managerial or organizational features allow for such change, or investigate the types of routines that should undergo change while others remain the same. Demonstration of performance differences between firms that are seen to have such dynamic capabilities and those that are not would help to further this literature. 104

114 Table 4.1 Numbers of Paired Patents for Firms in the Sample Firm Name # Paired Patents Abbott Laboratories 480 Allergan 199 Alza 321 American Home Products 226 Bausch and Lomb 32 Bayer 646 Boehringer 123 Bristol Myers Squibb 189 Chiron 21 Ciba-Geigy 332 Eli Lilly 378 Fujisawa 80 Glaxo 80 Hoechst 233 Janssen 48 Johnson and Johnson 67 Merck 419 Pfizer 271 Pharmacia 40 Procter and Gamble 1705 Rhone Poulenc Rorer 22 Rohm and Haas 113 Sandoz 24 Sanofi 65 Schering Plough 59 Searle 110 Smith Kline Beecham 6 Syntex 56 Roussel 29 Upjohn 71 Warner Lambert 186 Wyeth 71 Yamanouchi 5 105

115 106

116 Table 4.3 Relationships among Added Components, Changed routines and Innovation Use 107

117 Table 4.4 The Difference between Internal and External Use 108

118 Chapter 5: The Effects of Changed Linkages in Subtracted Knowledge Components on Innovation Outcomes In the knowledge recombination literature, the value of adding new knowledge components to existing ones has been widely acknowledged (Fleming, 2001; Katila and Ahuja, 2002; Carnabuci and Operti, 2013). Due to their ability to bring in fresh knowledge, the adding of new components to those already known to the firm can enhance originality of concepts, ideas and design. New knowledge components can also boost knowledge diversity that gives rise to novel relationships (Rosenkopf and Nerkar, 2008; Schilling and Green, 2011; Carnabuci and Operti, 2013). While the adding of knowledge components is one way to achieve knowledge recombination, other ways of recombining knowledge should also have the potential to elicit innovative outcomes (Schumpeter, 1939; Henderson and Clark, 1990). In particular, recombining a reduced set of knowledge components, accomplished by subtracting components from a previous innovation and recombining the ones that remain, can, theoretically speaking, also give rise to a novel innovation. Nonetheless, little research has investigated how subtracting knowledge components can influence innovation outcomes. Subtracting knowledge components is not the opposite of adding components, in which insights are reduced due to fewer components. Subtracting knowledge components may allow organizational members focus more keenly on relationships among the 109

119 abridged set, allowing them to gain deeper insights. Subtraction of knowledge components necessarily changes how the remaining components are connected to each other, thus affecting their joint meaning. The removal of knowledge linkages, or the structural pattern of how knowledge components are connected, changes the functioning of the reduced set of knowledge components. The manner in which knowledge components are recombined represents a framework for conceptualizing their underlying relationships. Thus, subtraction of some knowledge components indicates that the logic by which remaining knowledge components work together is altered. Correspondingly, changed knowledge linkages reflect this altered logic, resulting in new meanings and functions (Henderson and Clark, 1990). We thus argue that innovation novelty stemming from knowledge subtraction is a result primarily from changed knowledge linkages among the reduced set of components. As a result, the rationale of how subtraction of existing knowledge components gives rise to innovation novelty is distinct from that used for adding knowledge components in recombination. The unique features of knowledge subtraction suggest the need for its further study in order to gain a more comprehensive understanding of the ways in which knowledge recombination works. The goal of this research is to highlight the mechanisms and implications of subtracting knowledge components in knowledge recombination. In order to examine the effects of subtracting knowledge components on innovation, we account for changes in two key features of knowledge linkages extent and local embeddedness. While extent of linkages represents the overall connectedness of all knowledge components associated with an innovation; local embeddedness reflects the clustering of linkages into local neighborhoods or subgroups. We argue that, when 110

120 subtracting components, change in both extent and local embeddedness of linkages increases novelty of the newly created innovation. Since the way through which knowledge components are connected represents how organizational members conceptualize, interpret, make sense and reason the relationships among components (Dosi, Nelson and Winter, 2000), and involves a path-dependent process (Nelson and Winter), we expect that change in linkages will be hard to identify and accomplish, resulting in difficulties in using the resulting innovation. However, the understanding and experience accumulated in the creation process benefits the innovating firm in this regard. As a result, we expect the innovating firm to have a relative advantage in using the innovation compared to other firms that did not directly participate in the innovation creation but which intend to use it. Four sets of hypotheses are developed in relation to these arguments. Drawing upon patent data from 33 of the largest firms in the pharmaceutical industry between 1976 and 2014, we find strong support in our argument. These findings contribute to the existing knowledge recombination and innovation literatures in a number of ways. First, this study identifies a type of knowledge recombination subtraction of components that has not been previously much explored in the literature. In doing so, this research provides insight into additional ways in which recombination can give rise to innovation. Second, we highlight a differing rationale for how knowledge subtraction gives rise to innovation novelty as compared to knowledge addition. We stress that for knowledge subtraction, novelty originates from changed linkages and an altered framework for conceptualizing relationships among remaining components. Third, we differentiate the implication of knowledge subtraction for 111

121 innovating and non-innovating firms. By distinguishing between these two types of firms, we demonstrate differences in the ability to use the resulting innovation. We show that knowledge subtraction creates a relative advantage to the innovating firm due to its better understanding of underlying linkages. RECOMBINATION BY SUBTRACTING KNOWLEDGE COMPONENTS: CONCEPTUAL UNDERPINNINGS It has been widely acknowledged that knowledge components construct the basic blocks for an innovation (Fleming, 2001; Rosenkopf and Nerkar, 2001). Fleming (2001) argues that the choices of knowledge components come from social construction or previous association (p.118), while Nelson and Winter note that the creation of any sort of novelty in art, science or practical life consists to a substantial exert of a recombination of conceptual and physical materials that were previous in existence (p.130). It stands to reason, then, that an innovation can be achieved through change in knowledge components (Fleming, 2001; Katila and Ahuja, 2002; Carnabuci and Operti, 2013), as well as by linking them in different ways (Henderson and Clark, 1990). In particular, the arguments above imply that knowledge recombination can be accomplished in multiple ways, including adding new knowledge components to existing ones, as well as subtracting components and using a subset of previously used components. Nonetheless, a review of the literature suggests that scholars have heavily emphasized recombination through adding new components, but have not yet considered 112

122 the role of subtracting existing components in this regard. In the sections below, we discuss the nature of prior work on recombination. Prior Emphasis on Adding Knowledge Components for Recombination Innovation scholars have long recognized the importance of adding new knowledge components in knowledge recombination. New components are important in that they embody previously unused ideas and provide fresh insights for the recombination (Fleming, 2001; Katila and Ahuja, 2002). By combining new components with those that are already familiar to the firm, channels for enhanced knowledge flow and interaction are created, leading to new understandings that can be leveraged for transformed outcomes (Simon, 1962; Galunic and Rodan, 1996). Indeed, the empirical evidence reflects a consensus over the value of adding new knowledge components for knowledge recombination (Fleming, 2001; Rosenkopf and Nerkar, 2001; Ahuja and Lampert, 2001; Uzzi, Mukherjee, Stringer, and Jones, 2013). For example, Schilling (2005) has shown that the addition of diverse knowledge components from different knowledge domains allows novel ideas to emerge. Likewise, Rosenkopf and Nerkar (2001) demonstrate that most innovative technologies come from a recombination of diverse knowledge components that span technological categories and organizational boundaries, thus highlighting the value of knowledge components from new domains. Miller, Fern and Cardinal (2005) find that interdivisional knowledge enhances the impact of an innovation, while Carnabuci and Operti (2013) confirm that adding diverse knowledge components stimulates radical innovations. In addition, regarding the benefits of adding knowledge components from familiar knowledge 113

123 domains, Kaplan and Vakili (2014) highlight the value of adding new knowledge components from familiar domains, demonstrating that deep understanding of such knowledge can also create breakthrough innovations. It is understandable why researchers have concentrated on the effects of adding new knowledge components for an innovation, as their role in infusing new meaning is straightforward. However, this is not the only way in which recombination of knowledge components can occur As Schumpeter (1939) argues, innovation comes from recognizing how underlying parts of a system operate and how they can be reorganized to create superior or novel outcomes. Thus, theoretically speaking, recombination should be possible through reshuffling of existing components (Henderson and Clark, 1990) or by only subtracting some components. For the latter in particular, little consideration has been given in the literature. Subtraction of components may result in elimination of previously important components, greater emphasis may on previously peripheral functions, or other forms of reworked relationships among the remaining components. As we discuss below, the subtraction of previously used knowledge components and use of only a subset of them may also be an important means for creating innovation. Rationale for considering subtraction of knowledge components There are several reasons why subtracting knowledge components can give rise to novel innovation. One stems from the notion that reducing the number of knowledge components allows the firm to focus better on the functioning of a subset of knowledge components with which it has greater familiarity. A firm s accumulated experiences and capabilities in managing these existing components may give them an appreciation for 114

124 new ways in which they can be utilized or reorganized. This is in accordance with Kogut and Zander (1992), who point out that innovation is an outcome of a firm s combinative capabilities to generate new applications from existing knowledge (Kogut and Zander, 1992: 391). It is also consistent with March s (1991) organizational learning view that a firm s ability to innovate is influenced by learning from existing knowledge. While some have suggested that focusing on a subset of knowledge components can only allow for the refinement and extension of an existing innovation through exploitative learning (March, 1991; Benner and Tushman, 2002; He and Wong, 2004), it is possible for such a focus to result in highly novel or radical innovation as well. Deep understanding of existing knowledge components allow organizational members to push the boundaries of their application with greater likelihood of success (Sternberg and O Hara, 2000; Taylor and Greve, 2006). Kaplan and Vakili (2014) also argue that novelty is generated from focused knowledge and that incisive insight comes from honed appreciation of one s own knowledge. In Taylor and Greve s (2006) work, novel combinations are seen to come from extensive experience associated with existing knowledge components. Following this logic, subtracting knowledge components can help a firm to focus on interesting new combinations of existing components, thus allowing for the creation of novel insights. At the same time, the alteration of knowledge components through subtraction implies a new framework for conceptualizing their underlying relationships. Each combination of knowledge components signifies the existence of some logic for how they work with each other. Accordingly, when components are recombined, there will also be an underlying rationale for their joint operation. The recombined set of knowledge 115

125 components is unlikely to be random. Thus, the fundamental essence of subtracting knowledge components comes from the implementation of the new framework for integrating them. As Henderson and Clark (1990) also point out, the meaning of a knowledge component is realized in its linkages, or the way in which it is connected with others. Subtracting knowledge components not only changes the composition of an innovation, but, more importantly, it changes the ways of how the remaining components are connected with each other. Removal of some knowledge components requires rearrangement of remaining components in order for them to yield novel insights. Note that subtracting knowledge components is not simply the opposite of adding knowledge components in terms of knowledge recombination it has its own unique features. A key distinction between the addition of new components and the subtraction of existing components is their different sources of novelty. While the rationale for adding new knowledge components is relatively more obvious, due to the potential for novel insights from new knowledge, the benefits derived from subtracting knowledge components is not as easy to figure out for those that did not come up with the innovation. In the case of adding components, novelty is created through the fresh insights gained from new knowledge. Enlarging the pool of knowledge pieces rejuvenates the functioning of the entire set of components. In the case of subtracting knowledge components, novelty is created by a very different means through alteration in the way in which the remaining knowledge components are coordinated with each other. That is, rather than incorporating external knowledge to create a new innovation, subtraction requires a reworked set of connections among the rest of the components, In this way, as noted earlier, the value of subtracting 116

126 knowledge components lies in a new logic for connecting the reduced set of components. Thus, a hidden meaning underlies the subset of knowledge components. In contrast to the adding of new knowledge components, in which the new components play as important a role as the new linkages that integrate them with existing components, for knowledge subtraction the role of changed linkages plays the critical role in creating novelty. That is, once components are subtracted, changed linkages among remaining components are the key source of novelty of the resulting innovation. The new meanings afforded by the changed linkages among established components are vital in bringing about innovation. While the literature has paid keen attention to the adding of new knowledge components, the discussion above suggests that subtracting knowledge components can also result in valuable recombination. In particular, we argue that altered linkages among the reduced set of knowledge components can give rise to new meanings, and thus, novel innovation. Recognizing the actual meaning of an abridged set of knowledge components therefore requires consideration of their associated linkages. To date, however, no work has considered the effects of subtracting knowledge components on innovation outcomes. There is a need for further research to systematically investigate the role of knowledge subtraction in this regard. HYPOTHESES DEVELOPMENT We noted above that subtracting some knowledge components requires rearrangement of remaining components in order for them to yield novel insights, based on a new rationale for how they work with each other. In this section, we develop a systematic logic for how subtracted knowledge components and their accompanying 117

127 linkage changes influence novelty of the resulting innovation. Drawing on the notion that linkages among knowledge components reflect the organizational routines necessary for their implementation, we also account for the ability of innovating and non-innovating firms to use the innovation derived from component subtraction. We argue that innovations based on subtraction will favor the innovating firm over other firms in this regard. In line with these arguments, four sets of testable hypotheses are developed. Relationship between subtracting knowledge components and innovation outcomes As noted earlier, the subtraction of some knowledge components from an original set of components represents one of the ways in which knowledge recombination can be accomplished, Reducing components leads to a revised logic by which knowledge components are related to and work with each other, and this in turn enhances the potential for innovation novelty. Indeed, the more knowledge components subtracted in the process of recombination, the greater potential will be created for novelty of the resulting innovation. The rationale for this stems from the extent of change that is introduced into the arrangement of the resulting set of components. The dropping of well-established components will go hand-in-hand with significant rethinking of how the remaining components will work. Thus, greater subtraction will enhance the likelihood of new insights generated from the combination of the abbreviated set of components. By the same token, the fewer the knowledge components that are subtracted from the original set of components, the more the remaining components can work in a manner that remains consistent with their previous functioning. Thus, the dropping of fewer 118

128 components will tend to elicit results that are similar to the original innovation. Such incremental change will result in more incremental innovation. Following this logic, we hypothesize as follows: H1a: Subtracting knowledge components is positively related to the novelty of the resulting innovation. In order to benefit from an innovation created by subtracting knowledge components, firms must be able to use it. The ability to use it, however, will be different for innovating and non-innovating firms, or those that do not directly participate in the creation of an innovation. In order to use the newly created innovation, it is imperative that firms understand how knowledge components within the innovation work with each other. Since the innovating firm is the one that actually created the innovation, and participated closely in the innovation process, we expect it to have a good understanding of the rationale behind subtracting knowledge components. Such understanding facilitates its ability to use the innovation. When fewer knowledge components are subtracted to create the novel innovation, less change or effort is needed in order to use the resulting innovation. However, when more knowledge components are subtracted, the rationale of how the remaining knowledge components work with each other increasingly diverges from that associated with the original innovation. The greater the number of knowledge components subtracted from the original set, the more important it will be for firms to understand the rationale underlying the reduced set of knowledge components in order to use the resulting innovation. We expect that the innovating firm will have the ability to use the 119

129 resulting innovation due to its understanding of the mechanism underlying the remaining knowledge components. Accordingly, we propose the hypothesis: H1b: For the innovating firm, subtracting knowledge components is positively related to use of the resulting innovation. Non-innovating firms also need to understand the underlying rationale of the reduced set of components in order to utilize the newly created innovation. However, they may not have the correct understanding of the new rationale regarding how the reduced set of knowledge components work with each other and may assume that the remaining components work in a way similar to the previously non-abridged set of components. Indeed, when fewer knowledge components are subtracted to create the innovation, non-innovating firms will have less difficulty in using the innovation, since the remaining components may work in a way that is mostly consistent with the original set. However, as more components are subtracted, the underlying mechanisms and relationships become increasingly divergent from the original one. Due to their lack of familiarity with the process by which the innovation was created, we expect that it will be difficult for such firms to diagnose the correct rationale underlying the components. This poor understanding may cause a great difficult for non-innovating firms to use the resulting innovation. Therefore, we hypothesize as follows: H1c: For non-innovating firms, subtracting knowledge components is negatively related to use of the resulting innovation. 120

130 The effects of change in extent of linkages When subtracting knowledge components to create an innovation, the linkages among the reduced set of components play a very crucial role in influencing the meaning of knowledge components and creating novel functions. We argue knowledge linkages among components reflect organizational routines, sharing key features. Organizational routines provide the imperative to link dispersed pieces of knowledge together (Nelson and Winter, 1982; Cohen and Bacdayan, 1994). Knowledge linkages perform the same function, connecting otherwise isolated knowledge components together in order for them as a whole system to deliver its function. These knowledge linkages can connect knowledge components in a number of ways. Each way represents how organizational members interpret, reason, make sense and coordinate the relationships among knowledge components. This is also the key function of organizational routines (Dosi, Nelson and Winter, 2000). In addition, knowledge linkages and organizational routines exhibit context specificity. They are not formed randomly, but involved organizational deliberate thoughts. Every organization may have its unique linkages and underlying routines. Therefore, due to above reasons, we argue knowledge linkages connecting components reflect associated organizational routines. When knowledge components are subtracted, linkages among components are changed too. It indicates the underlying routines are also changed. These altered routines indicate the metal logic to reason and coordinate the remaining knowledge components is different. New insights can in turn be generated from the revised logic. How knowledge components and their underlying routines are connected or to what degree they are linked 121

131 indicates how organizational members perceive the relationship among them. Extent of linkages captures the overall connectedness of knowledge components and associated routines. Change in extent of linkages represents the mechanism through which the remaining knowledge components are recombined and reworked with each other are altered. New functions can be derived from the new mechanism behind the reduced set of knowledge components. Thus, the more change there is in the extent of linkages, the more difference there will be in the new logic connecting knowledge components compared to the original one, giving rise to a more novel outcome. In this way, the meaning of knowledge components is determined by the linkages between them (Henderson and Clark, 1990). Changed extent of linkages also alters the meaning of knowledge components that are connected within the innovation. Although this reduced set of knowledge components are all part of a pre-existing set, different linkages connecting them give them different meaning that allows for new functions to emerge (Galunic and Rodan, 1998). More change in the extent of linkages leads to greater departure of the new meaning from the original one, making the resulting innovation more novel. Following this logic, we propose the following hypothesis: H2a: Change in extent of linkages due to subtracting knowledge components is positively related to the novelty of the resulting innovation. In the previous section, we argued that knowledge linkages reflect organizational routines. A critical feature of organizational routines is their organizational context 122

132 specificity. Organizational routines are tacit and specifically embedded in the unique contexts within each organization (Cyert and March, 1963; Nelson and Winter, 1982). To utilize the resulting innovation, both innovating and non-innovating firms need to institute appropriate new routines. This may mean uprooting existing routines. However, organizational routines, once are established, are hard to change (Nelson and Winter, 1982; Teece, 1982; Edmondson, Bohmer and Pisano, 2001; Feldman and Pentland, 2003; Bresman, 2013; Hargadon and Sutton, 1997). In light of this, we argue that change in knowledge linkages is difficult for both innovating and the non-innovating firms in using the resulting innovation. Change in extent of linkages indicates that the overall connectedness of underlying routines has changed. For the innovating firm, its prior routines exercised over time, became a habit for the firm, and programmed into its structure (Nelson and Winter, 1982; Stene, 1940; Simon, 1976; March and Simon, 1958; Cyert and March, 1963; Allison, 1971; Gioia and Poole, 1984; Carley, 1996; Carley and Lin, 1997; Levitt et al., 1999). Changing existing routines to new ones to connect disparate pieces of knowledge in new ways can be difficult, due to inertia of these routines (Nelson and Winter, 1982; Feldman and Pentland, 2003). Change in routines can also cause and heighten anxiety within the organization (Teece, Pisano and Shuen, 1997). When a firm changes its routines, it can often provoke conflict (Nelson and Winter, 1982). Therefore, it is difficult for the innovating firm to change its routines in order to use the innovation. Indeed, we expect that the more changes in the underlying routines there are, the more difficult it will be for the innovating firm to use the innovation. 123

133 Non-innovating firms not only experience similar difficulty as the innovating firm in changing existing routines, but they also encounter other difficulties in instituting new routines. Recognizing and appreciating the necessary changes in routines is particularly difficult for them. The tacit nature of linkages and their underlying routines makes it very difficult for non-innovating firms to recognize or spot the change. They may assume that the reduced set of knowledge components still work as before and are consistent with the previous framework of routines (Henderson and Clark, 1990). In addition, even though non-innovating firms may recognize the associated changes in the underlying routines, they may not appreciate why these changes are necessary due to the lack of participation in the innovation creation process. Without adequate understanding of the reason and rationale for how the remaining components newly work and interact, they will face a great deal of difficulty in using the innovation. Based on above argument, we hypothesize as follows: H2b: For both innovating and non-innovating firms, change in extent of linkages due to subtracting knowledge components is negatively related to use of the resulting innovation. The effects of change in local embeddedness of linkages Local embeddedness of a linkage forms a local context that is consisting of other closely related and intertwined linkages in its neighborhood. The meaning of a linkage is influenced by this neighborhood or context within which it functions (Nelson and Winter, 1982; Teece, 1982). This local neighborhood creates a context that enhances the function of linkages within it. This coherent set of underlying routines generate a larger meaning 124

134 of for linkages within it then simply aggregating them together, because these linkages need to co-operate and coordinate together to derive the desired outcome (Teece, 1982; Galunic and Rodan, 1998). In this way, closely intertwined knowledge linkages and associated routines create a great amount of tacit knowledge and enhance the deep meaning of associated components. Because of the interdependence of knowledge linkages within the local neighborhood, a particular linkage has to coordinate with others in order to function coherently (Simon, 1962; Kogut and Zander, 1992; Galunic and Rodan, 1998). When there is a change in the local embeddedness of linkages, the context within which a particular linkage functions has changed. Its meaning also changes, because it may work or coordinate with other closely related linkages in a different way. As a result, the altered context can generate different functions for linkages within it. The more change in the local embeddedness of linkages there is, the more different meanings and functions that can be created, which in turn give rise to novelty. Therefore, according to this logic, we propose the following hypothesis: H3a: Change in local embeddedness of routines due to subtracting knowledge components is positively related to the novelty of the resulting innovation. In order to utilize the innovation, the innovating firm has to change a set of closely intertwined routines in order for linkages to function coherently. Nonetheless, performing such changes is very difficult, because the innovating firm has to change a set of coherently connected routines. Change in these routines requires a great deal of change 125

135 in other routines (Nelson and Winter, 1982; Teece, 1982). In addition, these closely local embedded routines consist of tacit knowledge that carries information flow and knowledge exchange (Henderson and Clark, 1990; Galunic and Rodan, 1998). Thus change the entire set of linkages and underlying routines in a desired way is very difficult, since tacit knowledge is difficult to figure out or manipulate. The more change there is in the local embeddedness of linkages, the more difficulty the innovating firm will likely experience in using the innovation. Non-innovating firms face greater difficulty in understanding any change in locally embedded linkages or in the local neighborhood formed by a set of closely interdependent related linkages and underlying routines. To begin, it is unlikely that noninnovating firms possess the same routines as the innovating firm. Therefore, the ability of non-innovating firms to perceive the meanings of linkages may not be consistent with the desired one. Lack of accurate understanding of an enlarged meaning created by local embeddedness of linkages impedes non-innovating firms ability to use the resulting innovation. The more change in the local embeddedness of linkages, the greater change in the set of routines necessary for ensuring appropriate knowledge integration. This will enhance the difficulty of non-innovating firms in creating the necessary changes for using the innovation Accordingly, we hypothesize in the bellows: H3b: For both innovating and non-innovating firms, change in local embeddedness of routines due to subtracting knowledge components is negatively related to use of the resulting innovation. 126

136 The relative effects of linkage change on innovating versus non-innovating firms Although both innovating firm and non-innovating firms have a great deal of difficulty in instituting the necessary changes in routines, we argue that change in linkages and underlying routines favors the innovating firm over non-innovating firms in using the innovation. In general, innovating firm has a better understanding of why such change is necessary. This advanced understanding gives the innovating firm a relative advantage in this regard. With regard to change in extent of linkages, the innovating firm knows well the motivation and rationale for why change in underlying routines is necessary for delivering the desired function. It helps the innovating firm to perform the required change in routines for using the innovation. In contrast, non-innovating firms may not have this high level of understanding. Partial or incorrect understanding may have no benefit for them in this regard (Teece, Pisano and Shuen, 1997). In addition, the innovating firm may already have some or necessary supportive routines in place due to its participation in innovation creation process. Non-innovating firms have to start from scratch to perform all necessary changes in extent of linkages and underlying routines, increasing the likelihood of instituting suboptimal changes. Taken together, we propose the following hypothesis: H4a: Change in extent of routines due to subtracting knowledge components gives the innovating firm a relative advantage over non-innovating firms in using the resulting innovation. 127

137 Regarding change in local embeddedness of linkages and underlying routines, the local set of closely intertwined routines make it difficult for both innovating and noninnovating firms to change their routines. For the innovating firm, good understanding is retained of how the changed local context generates altered meaning for linkages within it. Therefore, the innovating firm can appreciate the enlarged meaning created by the entire local set of embedded linkages and apply corresponding change in its routines in using the innovation. On the other hand, non-innovating firms will find it very difficult to recognize the high level of coherence created by the local neighborhood of highly interdependent routines (Teece, Pisano and Shuen, 1997). Therefore, their interpretation of the meaning of linkages may not be consistent with the desired ones. This inconsistency hampers their ability to use the innovation. Even though non-innovating can understand and appreciate the change in the local embedded linkages and underlying routines, they may not be able to replicate the entire set of routines necessary for using the innovation. Partial replication of routines may cause problematic outcomes that diverge from the desired function (Galunice and Rodan, 1998). In addition, the way in which these closely coherent routines work and coordinate with each other may be formed over time (Nelson and Winter, 1982). Without such a history, non-innovating firms may experience significant difficulty in using the innovation, giving the innovating firm a relative advantage in this regard. Following this logic, we propose the following hypothesis: H4b: Change in local embeddedness of routines due to subtracting knowledge components gives the innovating firm a relative 128

138 advantage over non-innovating firms in using the resulting innovation. METHODS Sample We follow prior innovation studies and use patent data from the pharmaceutical industry to test our hypotheses. It has been widely acknowledged that the pharmaceutical industry provides an ideal research setting for technological innovation research for several reasons (Guler and Nerkar, 2012; Levin et al., 1997). First, it has been confirmed that patents in the pharmaceutical industry appropriately represent firms innovative efforts, because pharmaceutical firms are generally motivated to apply for patents so as to protect their intellectual property. Second, almost all players in this industry patent at the United States Patent and Trademarks Office (USPTO), thus allowing USPTO patent database to provide detailed information about each patented innovation, such as their application and approval dates, classes, subclasses, claims, and so on. More importantly, the USPTO database uses subclasses to identify the specific technologies associated with each patent, and the approximately 100,000 such subclasses allow for fine-grained classification of key knowledge components involved in a patented innovation (Fleming and Sorenson, 2004). Since the whole assigning process is conducted by USPTO, the information gained through examining subclasses tends not to be biased by a firm s strategic decisions (Carnabuci and Operti, 2013). As a result, subclass information provides an ideal approach for studying knowledge recombination (Fleming, 2001; Fleming and Sorenson, 2004). Hereafter, we use subclasses to represent knowledge 129

139 components. By observing the entire patenting history of a firm, we are able to identify historical connections between components of each patent on the basis of all prior patents. Accordingly, we use an evolutionary approach to capture the linkages of each of the firm s innovations. Follow previous research, we collected data on patents from the 33 largest firms that comprise a total of 90 percent market share (c.f., Guler and Nerkar, 2012). We did so because the pharmaceutical industry is highly consolidated, with the majority of innovations created by the biggest players in this industry. Since our study focuses on highlighting the effects of the subtraction of existing knowledge components and the commensurate change in knowledge linkages, it is necessary to identify each corresponding pair of focal and resulting patent, where the resulting patent was originated from the subtraction of existing knowledge components from the focal patent. After examining all patents granted by USPTO between 1975 (the first year in which full-text records were available) and 2014, we were able to identify 7,476 pairs of firm-specific patents on the basis of the backward citations of each patent. Since we are interested in the change in knowledge recombination occurring within a firm, we only consider the backward citations that were patented by this firm. In this way, we are able to construct pairs of successive firm-specific patents that were created through subtracting existing subclasses and linkages. Table 5.1 lists the distribution of our sample patent pairs. As can be seen, Procter and Gamble has the most pairs of patents (1544 pairs) among the 33 largest pharmaceutical firms, followed by Bayer (595 pairs) and Merck (573 pairs), and Rhone Poulenc Rorer with the fewest at 15 pairs. This prominent different in patent pairs reflects 130

140 the variance in the innovativeness among these firms. Interestingly, when submitting patents to USPTO, firms sometimes use different formats for their names or make spelling errors. Therefore, the ambiguity of firm names is a major problem when assigning a firm to a patent (Li et al., 2014). To address this problem, this study adopted fuzzy matches to capture all the different formats used for firm names in the USPTO database. Dependent variable Innovation use. The use of a patented innovation is measured by its forward citations (Fleming, 2001; Fleming et al, 2007, Yayavaram and Ahuja, 2008). Since every patent is built upon prior patents, the creation of a patent is a path-dependence process utilizing prior knowledge. The more forward citations of a patent, the greater is its role in creating future patents. To measure external use of a patent, or use of an innovation by firms other than the one that created it, we count the citations of the patent made by these other firms. To measure internal use of a patent, or the use of an innovation by the firm that created it, we count the forward citations of the patent made by this firm. This study calculates the forward citations for all patents until June When a patent is not externally cited by many firms, but the ones that do cite it a lot, it results in high external citations. To better reflect the general usefulness of the patent, we also measure external use by taking account of the number of external firms that cite the patent. We use both measures in our analyses. 131

141 Independent variables Subtraction of new knowledge components. As in prior research (Carnabuci and Operti, 2013; Fleming, 2001), knowledge components are captured in this study by the subclasses co-assigned to each firm s patented innovations. The patents in our sample have a total of 41,024 different subclasses representing knowledge components in our study. The subtraction of knowledge components is the number of subclasses that are subtracted from the prior patent relative to the successive patent. Measure of linkages/routines. We conceptualize a linkage between knowledge components in a manner similar to coupling by other scholars (e.g., Fleming and Sorensen, 2004; Yayavaram and Ahuja, 2008). A linkage reflects how two components coordinate with each other in order to work appropriately. If the linkage between these components is strong, we expect their underlying routines to be closely connected, enhancing their joint operation and leading to an amplified meaning. In order to assess the strength of a linkage, we measure the interdependence between components. We draw on Fleming and Sorenson (2004) who note that a strong linkage between components creates high interdependence among them that will not make it easy to recombine with other components. In contrast, when interdependence is low, the addition of any new components is more easily accomplished, and allows for change in established routines. As a result, the strength of a linkage between two components is reflected in the ease of recombining the two components with other components. Drawing on the above, we measure the linkage, or degree of interdependence, between each set of two components by examining the number of other components 132

142 being co-listed with them in all the prior patents before the focal patent. For example, to measure the linkage connecting component i and j, we count the number of other components recombined with component i and j in the entire set of the firm s prior patents. The greater the number of components that have been previously combined with components i and j, the greater will be the ease (or lower the difficulty) of recombining these two components, and the lower is the strength of the linkage between them. Next, we account for the likelihood that knowledge components vary in the number of purposes for which they can be used. Components that have more purposes will be used in more patents and thus will be employed more frequently. To address this issue, we divide the above by the number of patents that include both components i and j. The formula for calculating a linkage between two components in a patent is as follows: Linkage or coupling between components i and j =1/ [!"#$%!"!"#$%&!!'!!"#$%&'()*!"#$%&'(!"#$!"!"#$%%&%!!"#!!"#$%!"!"#$%&'(!"#$%#&!"!"#$!"#$%&!!'!!!"#! Finally, we calculate the strength of each linkage associated with any two components in a patent and construct a linkage matrix for a patent incorporating the values for every linkage within it. We then construct a linkage matrix for every patent possessed by a firm in which all the components are treated as nodes of a network and linkages as ties. Treating it as a weighted adjacent matrix for a network, this linkage matrix of a patent is the foundation for deriving the two salient characteristics of linkage studied here the extent and local embeddedness of linkages. The calculation of each linkage of every patent requires a very dynamic approach, since the strength of a linkage in any point of time is determined by its most recent value. Measuring linkages in this way allows us to better capture the dynamic way in which a 133 ]

143 firm develops linkages and designs the routines that recombine knowledge components. The amount of computer processing time required to calculate each and every linkage in all the patents is tremendously high, taking months of CPU time. Change in extent of linkages. We use the average degree centrality of each knowledge components to measure the extent of linkages in a patent. Degree centrality refers to the total number of its direct connections with other nodes in the network, capturing the general connectivity of the nodes with others (Carpenter, Li, and Jiang, 2012; Scott, 2000). As noted earlier, we identified pairs of prior and subsequent patents. To assess change in the extent of linkages, we take the real value of the difference between the extent of linkages in the subsequent and prior patents. Change in local embeddedness of linkages. We measure local embeddedness through the degree of clustering, which allows us to capture how closely tied a linkage is to others in its neighborhood (Yayavaram and Ahuja, 2008). Local embeddedness elicits continuous values, with larger values indicating greater local embeddedness. To capture change in local embeddedness of linkages, we examine the local embeddedness between the prior and subsequent patents. We measure the real change between the local embeddedness of linkages between the two. Control variables We control for a number of contextual features associated with a patent that might influence the relationships we are studying. First, a larger number of inventors could represent more knowledge and lead to more forward citations for the patent they create (Singh and Fleming, 2010), while more 134

144 subclasses indicate more diverse knowledge and thus may enhance innovation novelty (Fleming, 2001; Rosenkopt and Nerkar, 2001). For these reasons we control for both number of inventors and number of subclasses for each patent. Knowledge maturity may also influence the innovation outcome and its use, since mature knowledge is easier to understand and could be more easily utilized by firms (Sørensen and Stuart, 2000). Based on previous studies (Agarwal and Hoetker, 2007; Lanjouw and Schankerman, 2003), we construct technology maturity of a patent by the number of prior arts within this patent. Knowledge diversity can also influence the dependent variables (Argyres and Silverman, 2004; Jaffe et al., 2002; Singh, 2008), since the more diverse the range of knowledge, the more areas a patent touches and more potential uses it can generate. Knowledge diversity is calculated by the proportion, pi, within a patent s backward citations that belong to technology class i. The formula for this variable is p 1 i i, with values falling between 0 and 1. We also control for the number of prior patents that the firm has to indicate their innovation capabilities, since different capabilities may influence both innovation novelty and use (Kogut and Zander, 1992; Teece, Pisano and Shuen, 1997). This variable is measured by the number of patents that a firm has applied for in a five-year window prior to the focal patent. Thus we have 3856 pairs of patents entering the model. We have both focal firm age and patent age as control variables. However, negative binomial model analysis automatically omits patent age in the regressions due to its high relation with focal firm age. Firm age is also controlled for, since innovation capabilities develop over time. We also control for patent age since this influences the level of forward citations associated with the patent. In addition, we controlled for the number of claims of a patent

145 The novelty of a patented innovation is indicated in the number of claims within the patent. A claim is a statement listed within the patent noting a new element of the patent reflecting knowledge previously unknown to the field (Beaudry and Schiffauerova, 2011). More claims associated with a patent reflect greater novelty or newness of the innovation (Duchesneau Cohn and Dutton, 1979; Hage, 1980; Ettlie, Bridges and O Keefe, 1984; Dewar and Dutton, 1986). Methods Because our dependent variables novelty (number of claims), external and internal use of an innovation (number of external and internal forward citations) are count variables, a Poisson or negative binomial estimation method is used to conduct the analysis. Since our independent measures are change-related measures, a random effects model is appropriate (Yayavaram and Chen, 2013). We therefore use the negative binomial estimation method with random effects to conduct the analysis. We also perform a Poisson estimation with fixed effects for robust checks. The negative binomial function is takes the following form: Γ(y + k) " k % P(Y = y X, k) = $ ' Γ(k)Γ(y +1) # k + µ & The expectation of Y is: E(Y) = µμ Applying a log link, we get: g(µμ)=log(µμ)=α + βx k y " µ % $ ' # k + µ & 136

146 where X captures the independent variables including change in extent of linkages, change in local embeddedness of linkages, and control variables; y relates to the dependent variables including novelty, external forward citations and internal minus external forward citations. The equations examined take the following form (with differing dependent variables): Use of resulting innovation= α! + β! subtracted knowledge components + β! change in extent of linkages + β! change in local embeddedness of linkages + controls+ e RESULTS AND DISCUSSION Table 5.2 shows the descriptive statistics and correlation coefficients for all the variables in our analyses. Due to the relatively high correlation across our major predictors, we examined uncentered VIFs in our models. Results show that the average VIF value is 3.82 (the maximum VIF was 5.79), indicating multicollinearity is unlikely to be of concern. Moreover, despite their relatively high correlations, these independent variables still remain significant in the analyses, indicating they are reasonably distinct constructs. Table 5.3 presents the results of the random-effect negative binomial estimation for innovation use, including four models to test different hypotheses. Each model in Table 5.3 includes two panels of results for different predicted effects. Model 1 reports the baseline models only including the effects of control variables on the novelty and overall citation of a resulting innovation. Model 2 reports the main effects of three major predictors the subtraction of existing components, the change in the extent of linkages, 137

147 and the change in local embeddedness on the novelty and overall citation of a resulting innovation. Commensurately, Models 3 and 4 respectively present the control model and the main effects of the three major predictors on the internal and external use of the resulting innovation. Table 5.4 reports the results for the difference between internal and external use of a resulting patent. Models 1 and 2 of Table 5.4 respectively present the baseline model and main effect model regarding the results for the difference of a resulting innovation s internal and external use. Hypothesis 1a predicts that the subtraction of existing knowledge components in the resulting innovation would enhance the novelty of this innovation by helping deepen the understanding of the remaining familiar components and altering the ways in which these components interplay and coordinate with each other. The results in Model 2 of Table 5.3 support this hypothesis, such that the subtraction of existing knowledge components in a resulting patent has a positive and strongly significant effect on the novelty of this patent (β=4.95, p<0.001). H1a is supported. Hypotheses 1b and 1c predicted the opposite effects of the subtraction of existing knowledge components on the internal and external use of the resulting innovation. That is, subtracting components will encourage the innovating firm that has created the resulting innovation to use this innovation, but impede non-innovating firms from applying this innovation. Both hypotheses are supported by the results in Model 4 of Table 5.3. That is, the subtraction of existing components is positively related to the internal use of the resulting innovation (β=12.84, p<0.001) and negatively related to its external use (β=-3.75, p<0.001) H1b and H1c are supported. 138

148 Hypothesis 2a predicted that in knowledge component subtraction, the changed extent of routines, reflecting greater need for new routines associated with the innovation, would enhance the novelty of the resulting innovation. Results in Model 2 of Table 5.3 provide strong supports to this hypothesis. That is, the change in the extent of routines due to component subtraction is positively related to the novelty of the resulting innovation (β=9.00, p<0.001). H2a is supported. For Hypothesis 2b, we had argued that the change in the extent of routines in the resulting innovation would discourage both the innovating firm and the non-innovative firms from using this innovation. The results in Model 4 of Table 5.3 show that the changed extent of routines in a resulting patent has significantly negative effects on its internal use by the innovating firm (β=-10.90, p<0.001), but is positively related to its external use by non-innovating firms (β=2.87, p<0.01). In such a manner, Hypothesis 2b is only partially supported. Hypothesis 3a predicted the effect of the change in another linkage dimension, i.e., the local embeddedness, expecting that in knowledge component subtraction, the changed local embeddedness would enhance the novelty of the resulting innovation. Results in Model 2 of Table 5.3 strongly support to this hypothesis. That is, the change in the local embeddedness due to component subtraction is positively related to the novelty of the resulting innovation (β=5.09, p<0.001). H3a is supported. For Hypothesis 3b, we had argued that the change in the local embeddedness in the resulting innovation would impede both the internal and the external use of this innovation. The results in Model 4 of Table 5.3 provide strong support for this hypothesis. That is, the changed local embeddedness in a resulting patent has significantly negative 139

149 effects on its use by both the innovating firm (β=-11.65, p<0.001) and the non-innovative firms (β=-23.26, p<0.001). H4b is supported. Hypotheses 4a and 4b predicted that in knowledge component subtraction, both the changed extent of routines and the changed local embeddedness will give the innovating firm an advantage over non-innovating firms in implementing the innovation. The results for these two hypotheses can be seen in Model 2 of Table 5.4. We find that the change in the extent of linkages is negatively related to the difference between internal and external use of the resulting innovation (β=-4.78, p<0.001), which goes against the prediction of Hypothesis 4a. In contrast, as predicted by Hypothesis 4b, the change in the local embeddedness is positively related to the difference between internal and external use of the resulting innovation (β=2.16, p<0.05). H4a is not supported. H4b is supported. With respect to our significant control variables, Models 1 and 3 of Table 5.3 show that in knowledge component subtraction, the age of an innovating firm is negatively related to the novelty of the innovation (β=-20.57, p<0.001) and to both internal (β=-77.53, p<0.001) and external use (β=-93.21, p<0.001). In contrast, the number of prior patents of this innovation is positively related to both the novelty of the innovation (β=2.91, p<0.01) and its internal use (β=17.56, p<0.001). Likewise, the maturity of the innovation is positively related to both the novelty of the innovation (β=22.97, p<0.001) and its internal use (β=2.77, p<0.01), but is negatively related to the external use of this innovation (β=-21.96, p<0.001). Furthermore, the number of inventors involved in the innovation is positively related to its novelty (β=9.63, p<0.001) and external use (β=2.03, p<0.05), but negatively related to its internal use (β=-18.64, 140

150 p<0.001). An interesting finding is that in knowledge component subtraction, the increased knowledge diversity of a resulting innovation can reduce the novelty of this innovation (β=-10.06, p<0.001), but significantly encourage both the innovating firm (β=9.52, p<0.001) and non-innovating firms (β=2.14, p<0.05) from using it. CONCLUSION AND DISCUSSION Innovation has consistently been one of the most important topics in management and organization research. Knowledge recombination perspective provides a very useful tool in studying innovation (Schumpeter, 1939; Fleming, 2001; Fleming and Sorenson, 2004). So far, the extent knowledge recombination literature has mainly focused on the value of addition of new knowledge components to create novel innovations (Fleming, 2001; Katila and Ahuja, 2002; Carnabuci and Operti, 2013). Adding knowledge components that are not previously applied in an innovation can not only introduce fresh insights and expand the knowledge set in the resulting innovation, but also rejuvenates the function of existing knowledge components, thus giving rise to novelty. The motivation of addition of knowledge components is to draw upon external knowledge pieces beyond the original innovation to bring in fresh insights. Prior studies have devoted considerable research efforts to highlighting the mechanisms and effects of adding new knowledge components in knowledge recombination. However, albeit the well understanding about the story on adding new knowledge components, little attention has been paid to the other side of the story how to create novel innovations without introducing external knowledge components beyond the original innovation. Put 141

151 differently, how could firms reconceptualize and revitalize existing innovations to create novel ones without adding new knowledge components? A typical example of such endogenous innovations can be achieved through the subtraction of knowledge components from existing innovations. Nonetheless, subtraction of knowledge components is not the opposite side of addition of components. They have different fundamental mechanisms associated with their knowledge components. That is, adding new components may not necessarily change the way in which existing components are associated and integrated with each other. In contrast, since subtracting existing knowledge components cannot bring any fresh insights or materials into the existing components, the essential way to create novelty by doing so is to alter and reconstruct the way in which the remaining components connect with each other and function together. Accordingly, the essence of subtracting knowledge components lies in a new logic for connecting the reduced set of components, which are reconceptualized and recombined to realize an altered meaning and deliver a differed joint function from the original innovation. In such a manner, regarding knowledge subtraction, the changed linkages across the remaining set of knowledge components play a critical role in creating novelty. That is, changed linkages among remaining components are the key source of novelty of the resulting innovation due to subtraction of knowledge components. Thus the new meanings afforded by the changed linkages among established components are vital in bringing about innovation. Drawing on the above insight, we strive to highlight the effect of subtraction of knowledge component on the novelty and internal and external use of the resulting innovation based on patent data from the pharmaceutical industry. To effectively identify 142

152 the subtraction of existing knowledge components within an innovation, we constructed pairs of patents of a firm in which the resulting patent had added no new components into the original patent but only removed existing ones. Our findings show that subtraction of knowledge components does play an important role in the knowledge recombination process. We find that subtraction of knowledge components can create novel innovations. We believe these results shed additional light for knowledge recombination literature in a number of ways. First, this study helps to address a gap in our knowledge regarding an important but neglected form of knowledge recombination the subtraction of knowledge components to create innovations. As discussed above, understanding the mechanisms and effects of subtraction of knowledge components can complement the extant research on knowledge recombination, which mainly focuses on the addition of new components. We expect that subtracting knowledge components can be an effective source of creating novel innovations as a form of knowledge recombination. Reducing the number of knowledge components allows the firm to focus on the subset of knowledge components with which it has greater familiarity. This enhanced familiarity can help a firm to accumulate experiences and to develop capabilities in managing these existing components. Therefore it can identify new ways in which the remaining knowledge components can be utilized or reorganized and appreciate the new insights embedded in the recombination. Our findings confirmed our prediction that subtracting existing knowledge components can indeed contribute to the novelty of resulting innovations. However, this benefit of subtraction of knowledge components may only help the innovating firm to use this innovation, but impede non-innovating firms in using the innovation. 143

153 Meanwhile, we further argue that the subtraction of existing knowledge components will intrinsically alter the linkages across the remaining components. Specifically, subtracting knowledge components not only changes the composition of an innovation, but, more importantly, alters the ways of how the remaining components are connected with each other, because the subset of existing components will need to be recombined and reworked with each other in order to achieve the goal of creating novelty and delivering desired functions. To serve this purpose, the recombined set of knowledge components is unlikely to be random, but represents organizational mental logic. We argue knowledge linkages across components reflect organizational routines, because both of them share key similar features as illustrated earlier in this paper. We focused on two central features of knowledge linkages the extent of linkages and the local embeddedness of linkages. Our findings indicated that both characteristics have a positive influence on the novelty of the resulting innovation. However, these two linkage features will have different implications for the internal and external use of the resulting innovation. We expect that changed linkages of these two features features impede the use of the innovation, since getting rid of routines may be difficult for both innovating and non-innovating firms. Our hypotheses basically support our expectation. Change in the local embeddedness of a resulting innovation has significantly negative effects on its use by both the innovating firm and the non-innovative firms. However, we have a surprising finding that the change in extent of linkages hamper the use of the resulting innovation by the innovating firm, but facilitate the use of this innovation by noninnovating firms. Such finding may tell us that non-innovating firms may bypass and drop the overall linkages and avoid the difficulty to change the underlying routines. 144

154 However, the embeddedness of linkages and associated routines are still difficult for them to work with. Moreover, the different effects of knowledge linkages on a relative advantage for innovating firms over non-innovating firms in using the innovation resulting from subtraction of knowledge component also provide a new insight to the knowledge recombination literature. We expect that change in both extent and local embeddedness of linkages gives the innovating firm an advantage in using the resulting innovation compared to non-innovating firms. The finding regarding the effect of local embeddedness linkages confirms our predication. But we also find that the change in the extent of linkages does not give the innovating firm the relative advantage in using the innovation. Rather, non-innovating firms have a relative benefit in using it. The reason is similar to the previous surprising finding that non-innovating firms can bypass and drop knowledge linkages and avoid the difficulty of changing associated routines. Taken together, these findings show the distinct implications of subtraction of knowledge component and change in linkages for generating novelty and using the newly created innovation. Subtraction of knowledge components alone and changed linkages can both create novelty, but through different mechanisms. Subtraction of knowledge components, as a recombination form, represents a new logic to conceptualize the relationships among the remaining ones. Changed linkages indicate the underlying routines to interpret and make sense the meaning of the connected components are also altered. Thus changed components and changed linkages have different influence on the use of resulting innovation. The benefit of reduced knowledge components can be appreciated by the innovating firm, but not by the non-innovating firm. In general 145

155 dropping linkages and associated routines is difficult for both innovating and noninnovating firms, since routines are hard and resistant to change. This study also suggests other potential theoretical and empirical directions for future research. First, subtraction of certain knowledge has been considered as a useful tool for gaining efficiency. However, its value of creating novelty has been overlooked. This study highlights the value of subtraction in management literature. Future research can also study subtraction of certain resources or firm-level knowledge base to advance our understanding of the merit of subtraction. This study suggests the role of linkages is the essential part in creating novelty and argue knowledge linkages reflect organizational routines. However, researchers investigating routines has been hampered by the opaque nature of routines and the difficulty of identifying them. The approach developed in this research may assist in this regard, as it captures the ongoing routines and their changing patterns using the linkages across different knowledge components. We believe that future research can benefit from this approach by generalizing it to examine other important topics related to routines. Furthermore, in this study we mainly focused on two central features of routines extent and local embeddedness. Given their different characteristics, our findings indicate these two features may have distinct implications for the knowledge recombination process. For example, we found different main effects of these characteristics of routines on internal and external use of the resulting innovation. Future research can benefit from further exploring the difference between these two structural features of knowledge linkages. In addition, future study can study other important features of routines to further our understanding on routines. 146

156 TABLE 5.1 Numbers of Paired Patents for Firms in the Sample Firm Name # Paired Patents Abbott Laboratories 493 Allergan 245 Alza 365 American Home Products 431 Bausch and Lomb 37 Bayer 595 Boehringer 212 Bristol Myers Squibb 296 Chiron 17 Ciba-Geigy 397 Eli Lilly 381 Fujisawa 88 Glaxo 123 Hoechst 201 Janssen 116 Johnson and Johnson 57 Merck 573 Pfizer 342 Pharmacia 20 Procter and Gamble 1544 Rhone Poulenc Rorer 15 Rohm and Haas 89 Sandoz 37 Sanofi 77 Schering Plough 71 Searle 90 Smith Kline Beecham 19 Syntex 75 Roussel 30 Upjohn 97 Warner Lambert 198 Wyeth 129 Yamanouchi

157 148

158 149