Master Thesis Two-Faced Knowledge Proximity: The Effect of Knowledge Proximity on Innovation Success from a Transaction Cost Perspective

Master Thesis

Two-Faced Knowledge Proximity: The Effect of Knowledge Proximity on Innovation Success from a Transaction Cost Perspective

THESIS INFO: ABSTRACT

Keywords: Alliance, Knowledge Proximity, Joint Steering Committee, Alliance Scope, Partner Similarity,

Technological Overlap, Innovation Success

Joran Kuperus s2558114

J.kuperus.1@student.rug.nl Supervisor: Mr. Marvin Hanisch Co-assessor: Dr. J. D. (Hans) van der Bij

Groningen, July 14, 2020 Academic year 2019-2020

Word count (excluding references and appendices): 14550 Word count (including references and appendices): 16981

Acknowledgements ... 3

1. Introduction ... 4

2. Theoretical Background ... 7

2.1. Alliances in the Pharmaceutical market ... 7

2.2. Knowledge Proximity in Alliances, an Absorptive Capacity Explanation ... 8

2.3. Willingness to collaborate: The Role of Transaction Costs Economics in Alliances 10 3. Hypotheses Development ... 12

3.1. The role of knowledge proximity in alliances ... 12

3.2. The moderating effects of Joint Steering Committees ... 16

3.3. The moderating effects of the Alliance Scope ... 18

4. Methodology ... 22 4.1. Empirical Setting ... 22 4.2. Data Collection ... 22 4.3. Measurements ... 23 4.3.1. Dependent variable ... 23 4.3.2. Independent variable ... 24 4.3.3. Moderators ... 25 4.3.4. Control variables ... 25 4.4. Analysis ... 29 5. Results ... 30

5.1. Descriptive Statistics and Correlations ... 30

5.2. Regression results ... 33

5.3. Marginal effect analysis ... 36

5.4. Robustness checks ... 37

5.5. Post-hoc analysis ... 37

6. Discussion and Conclusions ... 39

6.1. Theoretical Implications ... 39

6.2. Managerial implications ... 41

6.3. Limitations and future research ... 42

6.4. Additional Insights ... 44

6.5. Conclusion ... 44

References ... 46

Acknowledgements

1. Introduction

One out of every 5000 compounds. That is the answer to the question of how many compounds on average, after discovery, make it to the pharmacists shelf (Schilling, 2013). Research found that only one third of those successful compounds create high enough revenues to recoup the R&D investments from the entire R&D process (Schilling, 2013). Furthermore, DiMasi, Grabowski and Hansen (2016) indicated that the average cost of bringing a compound from discovery to market amounts to a total of $2558 million in capitalized cost. Adding the post-approval R&D costs to that would further increase the total costs to $2870 million (DiMasi et al., 2016).

To maintain these streams of innovations, firms need to access sufficient resources and capabilities to amend their knowledge bases (Devarakonda & Reuer, 2018; Rosenkopf & Almeida, 2003). It is for this reason that the rate of inter-organizational cooperation in the bio-pharmaceutical industry is amongst the highest in all industries. Interfirm collaboration can increase the competitive advantage in the marketplace (Dacin, Oliver, & Roy, 2006; Prashant Kale, Singh, & Perlmutter, 2000), increase profitability and performance (Ritala & Hurmelinna-Laukkanen, 2013) and allow for cost and risk sharing (Dacin et al., 2006). Alliances further pose a solution for the lack of further internal combinative potential by allowing bio-pharmaceuticals to access new external knowledge (Gnyawali & Park, 2011; Prashant Kale et al., 2000; Kessler & Chakrabarti, 1996). Despite these positive knowledge gathering effects, alliances are plagued with high failure rates (Ireland, Hitt, & Vaidyanath, 2002; Prashant Kale et al., 2000).

Some of the factors that contribute to this failure rate is a disbalance between the ability and willingness to successfully collaborate between partners (Devarakonda & Reuer, 2018; Prashant Kale et al., 2000; Sampson, 2007; Zhang, Jiang, Wu, & Li, 2019). Whereas the ability is the capability of partners to effectively share knowledge and collaborate in alliances. The willingness is the extent to which partners are willing to use the ability to effectively collaborate. One of the alliance characteristics that influences both these factors, is the extent of similarity in terms of resources between the partners which is termed in this research as knowledge proximity. Knowledge proximity has largely been explored only as a measure of absorptive capacity and can substantially increase the mutual absorptive capacity between the partners (ability) (Ahuja & Katila, 2001; Lane & Lubatkin, 1998).

the willingness and the ability to collaborate are pivotal for alliance success (Sampson, 2007). The existence of this gap stems from the absorptive capacity and resource-based view, in which exploration of high-level proximity seems limited by the assumption of low-level resource diversity, which means that it would not contribute to the overall firm performance (Nooteboom, Van Haverbeke, Duysters, Gilsing, & van den Oord, 2007; Sampson, 2007). However, research suggests an increase in coopetitive alliances for which knowledge proximity in alliances is high (Bouncken & Kraus, 2013; Estrada, Faems, & de Faria, 2016). The current literature could therefore benefit from a comparative analysis that puts the concept of knowledge proximity in a transaction cost perspective as well as an absorptive capacity perspective.

According to the Transaction Cost Economics (TCE), firms need to reduce the appropriation concerns and manage uncertainties by exercising control and coordination (Dacin et al., 2006; Devarakonda & Reuer, 2018; Kale et al., 2000; Sumo, van der Valk, van Weele, & Duysters, 2016). Knowledge proximity can inflate the potential transaction costs by facilitating easy knowledge spillovers and fast internalization of tacit knowledge between partners (Kavusan, Noorderhaven, & Duysters, 2016; Makri, Hitt, & Lane, 2010). This holds very different predictions for control and success than for the absorptive capacity perspective. The following research question is formulated:

“How does Knowledge Proximity between the alliance partners influence the Innovation Success of the alliance outcomes, and how do the presence of a Joint Steering Committee and the Alliance Scope breadth moderate this relationship?”.

Finally, the third hypothesis suggests that broad alliance scope weakens, or flattens, the inverted U-shaped relationship between knowledge proximity and innovation success. For alliances with high knowledge proximity, a broad scope would only increase the ‘touch points’ and knowledge leakage between the partners (Khanna, 1998).

To test these hypotheses and answer the research question, I conducted a Probit regression analysis. I use data of 242 biopharmaceutical alliance contracts from 2005 to 2008. The regression results did not show a statistically significant relationship between knowledge proximity and innovation success. Furthermore, no results are observed for incorporating a joint steering committee of broadening the alliance scope.

This research contributes to the existing alliance by addressing the differing effects of knowledge proximity on innovation success from both a transaction cost economics perspective and an absorptive capacity perspective. By incorporating a transaction cost perspective for high levels of knowledge proximity this literature challenges the assumption that increasing levels of absorptive capacity are beneficial for alliances, by indicating opportunism (Sampson, 2004, 2007). Thus, this research reestablishes knowledge proximity not only as a measure for absorptive capacity but also as a facilitator for certain forms of opportunism.

2. Theoretical Background

Innovations are usually categorized along two main lines, that of incremental (improving or updating existing products) innovation and that of radical (creating new and previously nonexciting concepts and products) innovation (Dong & McCarthy, 2019). For the pharmaceutical market, this is translated to updating or improving existing medicine (incremental) and creating new ‘first-in-class’ drugs (radical) (Dong & McCarthy, 2019). Pharmaceutical firms need to develop new first-in-class innovations, as this allows them to perform better and extend their survival (Hill & Rothaermel, 2003). In addition, increased competitive pressures and shortening development cycles force firms to increase their combinative potential, for which the resources more often lie outside their own internal boundaries (Tihanyi, Graffin, & George, 2014), which is especially the case in high velocity industries, such as the pharmaceutical industry (Baum, Calabrese, & Silverman, 2000; Hess & Rothaermel, 2011; Powell, Koput, & Smith-Doerr, 1996).

One way of coping with these issues for firms is by entering in an alliance (Dong & McCarthy, 2019; Rothaermel & Deeds, 2004). However, the performance of alliances is often influenced by partner similarity and dissimilarity, which is directly related to both the ability and willingness to successfully collaborate in an alliance (Gulati, 1995; Sampson, 2007; Saxton, 1997). The ability of partners to collaborate in alliances is expressed in how well partners can collaborate, understand each other’s information and each other’s way of working. This is often explained from an absorptive capacity perspective (Ahuja & Katila, 2001). This line of research focusses on how well partners understand each other’s mechanisms, technological background and internal routines, measured in knowledge proximity (Kavusan et al., 2016; Lane & Lubatkin, 1998; Sampson, 2007). Higher levels of knowledge proximity allows for better ‘absorption’ of partners knowledge and is linked to higher levels of innovation performance (Sampson, 2007).

primarily expressed in an ideal governance structure that facilitates optimal collaboration and reduces opportunism concerns (Devarakonda & Reuer, 2018; Oxley, 1997; Tsang, 2000; Williamson, 1998). However, knowledge proximity remains largely unexplored from a transaction cost perspective. Knowledge proximity is mainly linked to positive effects, while I will also argue that it can substantially hurt collaboration if too much is present.

Therefore, in the next section I will delve deeper in the definition of knowledge proximity from an absorptive capacity perspective. Thereafter, I will delve deeper in transaction cost economics and how it influences and predicts governance structures of alliances. Moreover, I will lay the first hand on how knowledge proximity might induce opportunism and how a joint steering committee and an alliance scope can moderate this effect.

2.2. Knowledge Proximity in Alliances, an Absorptive Capacity Explanation

One stream of alliance research focusses on strategic fit, or strategic similarities between alliance partners. In this stream internal processes, resources and capabilities are the predictors for alliance type, governance and success (Rothaermel & Deeds, 2004; Saxton, 1997). This stream primarily focusses on the ability of firms to access important resources and capabilities resulting in an increased innovation success (Hagedoorn, Lokshin, & Zobel, 2018; Hurmelinna-Laukkanen, 2011; Saxton, 1997). The underlying mechanisms is the successful knowledge sharing and acquisition that takes place within the alliance facilitated by mutual absorptive capacity (Grant & Baden-Fuller, 2004; Hagedoorn et al., 2018; Kavusan et al., 2016). In this context one of the partner similarities that is often discussed, is the similarity in terms of technological resources. From an absorptive capacity perspective some similarities between partners are necessary to facilitate sharing of new knowledge (Cohen & Levinthal, 1990). One much used variable to express the mutual absorptive capacity between partners is knowledge proximity (Devarakonda & Reuer, 2018; Sampson, 2004, 2007). Therefore, the following section seeks to define knowledge proximity as it is originally intended: as a proxy of absorptive capacity (Jaffe, 1986).

Defining Knowledge Proximity

this research, I will define knowledge proximity as the extent to which partnering firms possess the same technological resources and capabilities prior to the alliance (Lane & Lubatkin, 1998; Sampson, 2007; Zahra & George, 2002). If partners possess high levels of similar prior technological knowledge and organizational routines, it is likely that mutual absorptive capacity exists (Cohen & Levinthal, 1990; Sampson, 2007; Zahra & George, 2002). Let us dive deeper into what this mutual absorptive capacity means for collaboration in alliances.

Mutual Absorptive Capacity

Absorptive capacity is an important factor in collaborative alliances as it allows alliance partners search, find, assimilate, transform and apply knowledge for commercial ends (Zahra & George, 2002). Cohen and Levinthal (1990) argue that absoptive capacity is the neccesary condition for a firm’s succesfull exploitation of technical capibillities and external knowledge. Therefore, firms need to possess considerable in-house expertise that complements the knowledge of the alliance partner to be able to succesfully ‘absorb’ partners’ information (Mowery, Oxley, & Silverman, 1996). Research shows that absorptive capacity emerges after firms conduct and heavily invest in their own research and development (R&D) activity for a prolonged time (Cohen & Levinthal, 1990; Jaffe, 1986).

Absorptive capacity research identified two conditions that need to be fulfilled for successfully assimilating external knowledge (Cohen & Levinthal, 1990). The first condition is that for the effective and successful understanding of external knowledge, at least some part of the internal knowledge base should be similar, or proximate, to that of the partner (Kavusan et al., 2016). Knowledge proximity between alliance partners provides this greater understanding and internalization of each other’s knowhow, because partners share the common knowledge and have aligned absorptive capacity (Lane & Lubatkin, 1998; Rosenkopf & Almeida, 2003). The second condition is that at least some of the knowledge form those external sources should be new and unfamiliar to the partners (Kavusan et al., 2016; Zahra & George, 2002). After all, one should prevent that the novelty of the knowledge flowing within an alliance gets too low, as this would render the activity of collaboration obsolete for novel innovations (Nooteboom et al., 2007).

cooperate. It still leaves the question of how knowledge proximity influences the willingness to cooperate. As part of answering this question, transaction cost economics provides arguments that lie at the start of this explanation (Oxley, 1997). Correspondingly, in the next section I will shortly introduce transaction cost economics, knowledge proximity in light of transaction cost economics and the moderators for this research.

2.3. Willingness to collaborate: The Role of Transaction Costs Economics in Alliances

One theory that is heavily used to explain the behavior of partners is transaction cost economics (TCE). TCE formulates three attributes for describing transactions: (1) the frequency of the transaction, (2) the uncertainty surrounding these transactions, and (3) the degree asset specificity (Williamson, 1998). TCE and its view on governance of alliances assumes that even the most complex contracts are not complete (Williamson, 1998).

The incompleteness of contracts results in what Williamson (1998) calls “maladaptation hazards, by reason of opportunism” (p.36). Thus, the incompleteness of contracts incentivizes the partners to conduct in opportunistic behavior (Oxley, 1997), which Williamson (1985) defines as the “self-seeking interest with guile” (p. 47) (Stone, 1986). Non-equity alliances lack the necessary mechanisms to regulate access to resources, respond to contingencies and control partners behavior and therefore adaption challenges between the partners arise. These adaptation challenges might be particularly severe for where there is extensive overlap between the technological domains (Katila, Rosenberger, & Eisenhardt, 2008). It is thus important to analyze knowledge proximity from a transaction cost perspective.

Knowledge Proximity in Alliances, a first transaction cost introduction

levels of knowledge proximity, competitive dynamics might play a role because there is an increased likelihood that they build on the same external resources, connections and assets (Oxley & Sampson, 2004).

It is for this reason that knowledge proximity might induce opportunism in alliances and can create detrimental effects on innovation success (Judge & Dooley, 2006). The opportunistic behaviour can take several forms and include appropriating partners’ resources, communicating wrong information, hidden agendas, and delivering inadequate products and services (Das & Teng, 1999). To address these issues TCE introduces several formal and informal governance mechanisms depending on the amount of uncertainty and opportunism present. Some examples of formal governance mechanisms are equity control (Oxley, 1997) and contractual safeguards or contractual incentives. Some examples of informal governance mechanism are specific-working documents, hierarchal control and integrative conflict management (Kale, Singh, & Perlmutter, 2000).

Joint Steering Committees and the Alliance Scope

A less researched, hybrid governance mechanism is that of formal oversight committees (joint steering committees) that oversee and monitor the collaboration between partners (Reuer & Devarakonda, 2016; Williamson et al., 1991). Joint steering committees (JSC) are contractually created oversight committees to formalize the joint oversight control of the non-equity alliances by both partners (Devarakonda & Reuer, 2018). To some extent, a JSC resembles the oversight board in joint ventures and mimic’s the same coordinative actions and responsibilities (Devarakonda & Reuer, 2018; Oxley & Wada, 2009). A JSC can substantially reduce opportunistic behavior and uncertainties in alliances. For example, a steering committee enhances the communication within alliances by establishing new lines of communication, oversees activities and allows for internal conflict management (Devarakonda & Reuer, 2018; Sampson, 2004).

The increased contact and face to face collaboration across the value chain, leads to the increased uncertainty of the tasks and increases the costs of monitoring the partner’s behavior (Pisano, 1989).

To test these assumptions within alliances, several hypotheses were constructed. The first hypothesis focuses on the effect of knowledge proximity from both the absorptive capacity and transaction cost perspective and how that impacts innovation success. The second and third hypothesis focusses on two governance mechanisms that can decrease and increase the opportunism evoking effects of knowledge proximity: the presence of a joint steering committee and the breadth of the alliance scope.

3. Hypotheses Development

Alliances are excellent mechanisms through which firms can transform and share knowledge with their partners. Knowledge sharing is heavily influenced by a partners’ resource fit or strategic fit (Saxton, 1997). One construct that expresses that similarity is knowledge proximity. Knowledge proximity is the similarity in the amount of technical resources and capabilities that collaborating firms have in common (Jaffe, 1986). This similarity can have performance enhancing effects as it lays the foundation of absorptive capacity between the partners. These beneficial effects are however limited by the ongoing tension of opportunism between partners in alliances. In line with Haans, Pieters, and He (2016), I will start by explaining the effect of knowledge proximity that will consist of two latent functions that will result in a turning point and will show an inverted U-shaped relationship on innovation success. It is the simultaneous occurrence of two effects that jointly make up the inverted U-shaped relationship between knowledge proximity and Innovation success (Haans et al., 2016).

The rational for curvilinearity: Absorptive Capacity and Transaction Cost Economics

The first latent mechanism is that of absorptive capacity, which positively influences innovation success.

(Lee et al., 2017; Sampson, 2004a). Joint collaboration then becomes incredibly complex due to the mismatch in organizing routines and resources (Hagedoorn & Wang, 2012; Lee et al., 2017). Thus, when knowledge proximity is low, firms lack the ability to sufficiently collaborate and share knowledge, reducing the chances of innovation success.

If knowledge proximity rises, then so does the level of mutual absorptive capacity that exists (Ahuja, 2000; Lane & Lubatkin, 1998; Oxley & Sampson, 2004; Zahra & George, 2002). Greater similarity between the partners knowledge bases points towards prior experience working in technology areas closely related to each other (Zander & Kogut, 1995), and increases the familiarity with the structure and content of the knowledge and the potential inflows that the partner experiences. Cohen and Levinthal (1990) argue that these higher levels of mutual absorptive capacity allow for better understanding and mutual adjustments, which allows for better collaboration (Zahra & George, 2002). Furthermore, firms with similar technological resources are likely to rely on similar scientific principles to develop competencies and capabilities (Makri et al., 2010), which enhances the partner’s ability to identify and recreate the tacit elements underlying each other’s knowhow (Kavusan et al., 2016). It is thus likely that this has corresponding positive effects on the joint collaboration needed for successful clinical trials (Innovation Success).

Increasing levels knowledge proximity can also increase the successful exploitation of the acquired knowledge. Alliance partners with considerable overlap in technological resources can easily and quickly acquire each other’s knowledge (Ritala & Hurmelinna-Laukkanen, 2013; Williamson, 1998). As firms gain more technological similarities, it is likely to find a better match between their dominant logics (Lane & Lubatkin, 1998). That in turn allows firms exploit the internalized knowledge better (Schildt et al., 2012).

However, these beneficial effects of knowledge proximity can reach a level at which the proximity loses it potential. After all, it is assumed that innovations rise out of new combinations of existing capabilities. There exists a minimum threshold after which the combinative potential is diminished because new combinations of knowledge are exhausted (Mowery et al., 1996; Sampson, 2007). Thus, in line with Sampson (2007), it is likely that firms most benefit from R&D alliances when partner capabilities are similar, but not so similar that innovative combinative potential is exhausted (Ahuja & Katila, 2001). Therefore, I expect that the latent causal mechanisms of absorptive capacity increase positively and linearly up until it flattens for very high levels of knowledge proximity (see graph 1, figure 3.1).

Although, knowledge proximity has many positive effects from an absorptive capacity perspective on the ability to collaborate (Kavusan et al., 2016), theory suggest that it can also hurt the willingness to collaborate from a transaction cost perspective (Oxley & Sampson, 2004; Sampson, 2007). As the transaction costs increase, due to increasing needs for control, the positive relationship between knowledge proximity and innovation success shifts downwards for higher levels of knowledge proximity, which amounts to an inverted U-shaped relationship. It is thus that opportunism, is the second latent causal mechanism for the following reasons:

First, despite the fact that informational uncertainty is low, behavioral uncertainty increases (Sampson, 2007; Simonin, 1999). Collaborations are inherently risky and are plagued with a ‘base-level’ of opportunism that is present in R&D alliances because the future value of the innovation at hand is hard to predict. Consequently, high levels of knowledge proximity allow the alliance partner to easily combine unintended spillovers with their existing internal knowledge (Devarakonda & Reuer, 2018; Zahra & George, 2002; Zander & Kogut, 1995). Hence, the possibilities and risks of opportunism increase for higher levels of knowledge proximity.

Second, higher levels of knowledge proximity increase the likelihood that there is upstream and downstream competition between the alliance partners. With extensive overlap, firms are likely to draw on the same external pools for resources, complementary assets and technological knowledge (Oxley & Sampson, 2004). When this is the case, partners might perceive each other as competitors in both the resource market as well as the downstream product market (Estrada et al., 2016; Oxley & Sampson, 2004). Thus, for high levels of knowledge proximity it is likely that competitive dynamics play a role. This further increases the appropriation concerns because both firms have a high ability and incentive to absorb valuable knowledge from each other further triggering the risk of knowledge spillovers (Hamel, 1991; Lane & Lubatkin, 1998; Ritala & Hurmelinna-Laukkanen, 2009).

Third, for knowledge proximate collaboration, it is likely that the core of the collaborating firms is more exposed to their counterpart. This is because the similarity between the partners can be so high that the only thing that is setting them apart is their core knowledge and simultaneously their only differential competitive advantage (Estrada et al., 2016). This realization might increase concerns for knowledge spillovers that will be used for private gains and substantially reduce the willingness to collaborate (Laursen & Salter, 2014; Santoro & Mcgill, 2005; Tsang, 2000).

knowledge increase as well. This increases the need for coordinating, policing and enforcing partners behavior due to the increasing levels of opportunistic behavior. The positive effects of knowledge proximity will therefore diminish as the level of knowledge proximity becomes higher. I thus expect that this latent causal mechanism shows a concave increasing relationship between knowledge proximity and opportunism (see graph 2, Figure 3.1).

Figure 3.1.

The additive effect of the two latent mechanisms

Note. The blue lines indicate the normal model and additive effect of the two proposed causal mechanisms.

However, these figures are not plotted by a statistical program and only serve the purpose of illustrating the interaction effects of the proposed arguments.

Thus, to further clarify I expect that these two latent causal mechanisms appear. A positive concave relationship of knowledge proximity on absorptive capacity, that is positively related to collaboration and knowledge sharing. On the other hand, in graph 2 we see a concave relationship between knowledge proximity and opportunism, or increasing transaction cost, with negative effects on innovation success. If we subtract these two latent mechanisms it is likely that we find an inverted U-shape relationship with innovation success. Thus:

3.2. The moderating effects of Joint Steering Committees

“Moderation in U-shaped relationships can be separated in two types: the curve can shift left or right and it’s shape can flatten or steepen” (Haans et al., 2016, p. 1178). I suggest that the presence of a Joint Steering Committee (JSC) has a moderating effect on the concave transaction cost mechanism. It weakens and controls for the opportunism arising from high levels of knowledge proximity. Thus, from a transaction cost perspective, I expect the presence of a JSC to steepen the inverted U-shaped relationship (thus, the tipping point moves upwards) between knowledge proximity and innovation success for the following reasons.

First, joint steering committees create a platform between the partners, which facilitates fast information exchange and information processing between partners (Reuer & Devarakonda, 2016). These platforms consist of members from both firms, intensifying the cross-collaboration by key decision makers. Next to fast information exchange, these platforms serve as a joint decision-making platform (Devarakonda & Reuer, 2018). This consensus-driven decision-making process allows each party to re-evaluate and renegotiate their standpoint when decisions have to be made and allows key decisions makers to jointly decide on the progress and adjustments (Ness, 2009). These mutual adjustments between the partners then serve as a safeguard for each parties interests (Reuer & Devarakonda, 2016). Hence, joint steering committees reduce the incentive for partners to act opportunistically in the first place because partners might see consenting as more fruitful in the long run than contending in the short run (Ness, 2009). Thus, the latent causal mechanisms of increasing opportunism for increasing knowledge proximity will be weakened.

Devarakonda, 2016). This reduces the occurrence of opportunism and the incentive to act opportunistically. Similar to the previous point, as the opportunism for both high and low levels of knowledge proximity are weakened, graphically the causal mechanism of opportunism is flattened.

Third, a JSC can formulate rules of engagements and control the interaction between the partners’ employees (Liebeskind, 1997). Especially where there is mutual understanding of basic principles and absorptive capacity between the partners the threat of knowledge leakage between individuals and it being used for private gains increases (Katila et al., 2008; Reuer & Devarakonda, 2016). By formulating rules, a JSC serves as an extra administrative layer over the already existing project level management (Devarakonda & Reuer, 2018). This allows the firms to manage the amount of face-to-face interaction between the partners’ employees in order to limit the understanding and transparency of knowledge, especially that of tacit knowledge (Heiman & Nickerson, 2004). By managing the face-to-face interaction, a JSC can directly reduce the leakage and access to the relevant know-how (Liebeskind, 1997). Hence, it is found that where there is large technological overlap, JSC’s reduce the extent to which partners build on the partner’s knowledge, reducing misappropriation and opportunism (Devarakonda & Reuer, 2018).

Thus, the employment of a steering committee mitigates the opportunism for very low and very high values of knowledge proximity. By increasing coordination and reducing opportunism it is likely that opportunism will not arise as quickly as without a steering committee. Combined, this results in a weakening effect of the curvilinear latent mechanism of opportunism (See graph 2, figure 3.2). By weakening the latent relationship, the beneficial function of absorptive capacity is likely to take the upper hand. This means that the curvature of the line itself does not change, yet, the overall inverted U-shaped relationship increases because the negative concave transaction cost or opportunism function is weakened. Together, this amounts to a steepening effect on the U-shaped relationship. Thus, by weakening convex cost curve on both sides a steeping effect on the inverted U-shaped relationship should occur. Thus:

Figure 3.2

The strengthening effect of a Joint Steering Committee on the inverted U-shaped relationship

Note. The blue lines indicate the normal model and additive effect of the two proposed causal mechanisms. The

red line indicates the new or changed line. However, these figures are not plotted by a statistical program and only serve the purpose of illustrating the interaction effects of the proposed arguments.

3.3. The moderating effects of the Alliance Scope

I suggest that a broad alliance scope has a moderating effect on the concave opportunism or transaction cost mechanism. It strengthens and exacerbates the opportunism arising from high levels of knowledge proximity. Therefore, a strengthening effect on the curvilinear mechanism of opportunism will be expected. Thus, from a transaction cost perspective I expect that a broad alliance to weakens the inverted U-shaped relationship between knowledge proximity and innovation success for the following reasons.

that partners have to choose to either limit the scope, potentially hurting the alliance outcome, or, increase the monitoring capabilities consequently increasing the monitoring and policing costs. Hence, an increase in scope further increases the transaction costs for partners, which pushes the curvature upwards of the latent transaction cost mechanism (figure 3.3 graph 2).

Second, for high levels of knowledge proximity, the tacit knowledge that resides in ‘star’ scientists is often more in danger (Hess & Rothaermel, 2011). The reason is that for an increase in vertical scope, the extent of knowledge sharing inevitably increases, which is necessary for operations to proceed efficiently (Li, Eden, Hitt, Ireland, & Garrett, 2012; Oxley & Sampson, 2004). Knowledge that partners gain through collaboration in the different areas of the value chain are then likely to be linked, by “inseparability of operational routines” (Li et al., 2008, p. 322). This means that what is learned elsewhere, might flow to other areas and individuals of the alliance operation (Reuer et al., 2002). This in turn increases the likelihood that boundaries are ill-defined and increases the points of contact between the partners (Khanna, 1998; Oxley & Sampson, 2004). This makes effective collaboration – without leaking tacit knowledge that is deeply rooted in partners’ core – almost impossible (Reuer et al., 2002). If this is combined with high levels of knowledge proximity between partners, the partners not only possess the ability to misappropriate, but also have access to a vast amount of knowledge to appropriate (Khanna, 1998).

For the low levels of knowledge proximity, the monitoring and coordination costs are already high, due to the possible information asymmetry and mismatch between partners organizational routines (Hagedoorn & Wang, 2012; Lee et al., 2017). When the alliance scope increases, monitoring knowledge flows and collaboration activities becomes increasingly difficult, only adding up to the already difficult situation of understanding the partners unfamiliar resources (Li, Eden, Hitt, Ireland, & Garrett, 2012; Reuer et al., 2002). Thus, also for low levels of knowledge proximity, an increase in upward curvature for the latent causal mechanism of opportunism is expected (see graph 2, figure 3.3).

for the entire inverted U-shaped relationship. This means that the relationship between knowledge proximity and innovation success is weakened and lower levels of performance will be observed for similar levels of knowledge proximity. Thus:

Hypothesis 3: The breadth of the Alliance Scope moderates the relationship between Knowledge Proximity of the alliance partners and the Innovation Success, such that increasing values of Alliance Scope has a flattening effect on that relationship.

Figure 3.3.

The weakening effect of a broad Alliance Scope on the inverted U-shaped relationship

Note. The blue lines indicate the normal model and additive effect of the two proposed causal mechanisms. The

Figure 3.4.

The conceptual model is depicted below. It illustrates the hypothesized relationships.

4. Methodology

The data consists of dyadic alliances within the biopharmaceutical market. There are several reasons why the biopharmaceutical market is an interesting and an appropriate market to conduct alliance research in (Devarakonda & Reuer, 2018). First, the pharmaceutical industry is a typical example of a high velocity industry in which the source of innovation often lays outside the firm (Baum et al., 2000; Li, Qiu, & Wang, 2019; Powell et al., 1996). Correspondingly, the biopharmaceutical is amongst the highest industries to engage in inter-firm collaboration by means of alliances (DiMasi et al., 2016; Hess & Rothaermel, 2011). Second, the biopharmaceutical firms are among the highest in employing protective measures, such as patenting (Lanjouw & Schankerman, 2004) and equity measures such as mergers and acquisitions (Schweizer, 2005) to secure new and relevant knowledge. This might resemble a high potential in responses to opportunism that I am trying to uncover. Third, the biopharmaceutical market has a strong innovation cycles, primarily driven by high levels of competitive pressures to be the first one to develop a compound (Ball, Shah, & Wowak, 2018) and the aforementioned patenting behavior. Knowledge leakage and unintended spillover are therefore likely to be a major issue in this industry. Together this creates the perfect dynamism to research knowledge proximity and it potential to fuel opportunistic behavior between the alliance partners from a Transaction-Cost perspective.

4.2. Data Collection

To test the hypotheses, I relied on multiple data types and sources to generate the final sample that was used. As mentioned, the sample consists of non-equity alliances from the biopharmaceutical market. In addition, the level of analysis is dyadic and on the alliance level.

First, Bioscience Advisors was used as the database to collect alliance contracts between 2005 and 2008. Bioscience Advisors is a consulting and database firm focused on biopharma alliances and is therefore a direct source to these contracts. This led to an initial sample of 684 possible alliance contracts. After further analysis I removed duplicates which decreased the sample to 537 possible alliances. After that, I checked for the M&A contracts and whether or not the contracts have traceable compound and molecule numbers. This reduced the sample further to a final sample of 242 coded contracts.

outside the U.S.; Susarla, 2012). These files are used to determine the R&D investments, revenue, net profit and number of employees one year prior to the alliance. However, if one of the alliance partners is headquartered outside of the U.S. financial data was sometimes unavailable through the EDGAR database. To fill that gap the ‘Wayback Machine’ from the Internet Archive was used, which contains snapshots of 20+ years of website data accumulating to approximately 330 billion web pages. This allowed me to access and look at a snapshot of websites from the alliance partners a year prior to the alliance, retrieving the missing data.

Third, the publicly available website ClinicalTrials.gov. was used to obtain the relevant NCT numbers (NCT: National Clinical Trial). The NCT number is the unique identifier linked to the clinical trials in which the compound or molecule is tested. The corresponding clinical trials were checked for the correct compound name, the correct partners and the correct area of disease. If this checked out, the matched NCT number was added to the database sample (Hoang & Rothaermel, 2005).

Fourth, the United States Patent and Trademark Office (USTPO) is used for the collection of the USTPO patent data. Collected from these patents is the three digit patent class, which is necessary for the calculation of technological overlap between alliance partners (Oxley & Sampson, 2004; Reuer & Devarakonda, 2016; Sampson, 2007).

4.3. Measurements

4.3.1. Dependent variable Innovation Success

was FDA approved and marketed. After all, the successful transition from one phase to the next captures both the research and development capabilities of the alliance partners (Dimasi et al., 2010; Hoang & Rothaermel, 2005). Thus, innovation success was coded ‘1’ if a transition has taken place and ‘0’ if no transition has taken place. This led to a total of 48 successful innovations and 194 unsuccessful innovations.

4.3.2. Independent variable Knowledge Proximity

The independent variable is defined in the theoretical section as ‘knowledge proximity’. The goal of this measure is to predict the similarity or diversity between partners’ technical resources and capabilities (Reuer & Devarakonda, 2016). It is important to look at the knowledge proximity of the partners prior to alliance as this allows us to predict the behavior during the alliance based on the unchanged patent distributions. After all, patenting patterns after or during the alliance are likely to converge, indicating an ex ante and ex post collaborative effect (Sampson, 2007). To measure this variable I use the angular measure knowledge proximity that was introduced by Jaffe (1986) and is frequently used in different streams under different names as technological overlap or technological diversity (Oxley & Sampson, 2004; Sampson, 2007). This measure uses the pre-alliance distribution of patents across the various patent classes (Mowery et al., 1996; Oxley & Sampson, 2004; Reuer & Devarakonda, 2016). The 328 patent classes that are available through the UTSPO (note: there are many more subclasses) are then used to determine the actual proximity. The proximity between firms i and j (Pij) is calculated using the angular separation calculation of the vectors Fi and Fj as depicted

below.

Formulae 4.1: Technological overlap (Knowledge Proximity) formula

KNOWLEDGE PROXIMITY= 𝐹𝐹𝑖𝑖𝐹𝐹𝑗𝑗′ �(𝐹𝐹𝑖𝑖𝐹𝐹𝑖𝑖′) (𝐹𝐹𝑗𝑗𝐹𝐹𝑗𝑗′)

4.3.3. Moderators Joint Steering Committee

Joint steering committee refers to a contractual established oversight committee tasked with the responsibility of overseeing the alliance activities (Devarakonda & Reuer, 2018; Reuer & Devarakonda, 2016). However, the terminology to express the presence of such a committee within a contract is not unilateral. Contracts use different terminology such as ‘team’, ‘committee’, or ‘taskforce with oversight capabilities’. The presence of any of these terminologies is then coded binary to indicate whether such an oversight committee is present within the contract. In this research I only focus on whether such an organ is present or not, which is in in line with the hypothesized argument that the presence of a ‘steering committee’ and the overall coordinating role has implications (Reuer & Devarakonda, 2016). Thus, steering committee is a binary variable indicating ‘1’ if a steering committee is present, indicating ‘0’ if a steering committee is not employed (Reuer & Devarakonda, 2016) .

Alliance Scope

Alliance scope is defined as “the extent to which partners combine multiple and sequential value chain activities, such as R&D, manufacturing and or marketing” (Oxley & Sampson, 2004, p. 726). Alliance scope will be coded as a continuous variable with values ranging from 1 up to 5. The alliance contracts indicate who controls an activity, only if that activity is part of the alliance. This allows me to use the delegating articles for control as a proxy for the extent of activities included in the contract. The control of activities was coded along five main areas: research, clinical trials, regulatory affairs, manufacturing and commercialization. If the activity was covered in the contract, the applicable variable was coded according to who has the responsibility, or control. Because the variable was categorical, I turned all slots containing a value into ‘1’, then taking the sum of all of the control indications per contract. This led to the variable Alliance Scope which ranges from ‘1’, a narrow scope, to ‘5’, a broad scope (Khanna, 1998; Oxley & Sampson, 2004).

4.3.4. Control variables

same partner, it is likely that it can have a broadening effect on the alliance scope, while reducing opportunistic behavior (Gulati, 1995; Oxley & Sampson, 2004). Therefore, a dummy binary variable, Prior Relationship, will be included. The variable is coded ‘1’ if prior to the focal alliance, the same set of partners have collaborated before.

Contract Complexity. I further control for the substituting effect of complex contracts. There is evidence that contract complexity can substitute for ownership and that contracts are often more complex for early stage development. This is in line with the commonly held assumption that future contingent factors are harder to predict in research intensive environments. In addition, contract complexity might act as a substituting factor for oversight committees and actual ownership structures (Robinson & Stuart, 2007; Schilke & Lumineau, 2018). Thus, complex contracts might be employed by partners to protect their core knowledge (Reuer & Devarakonda, 2016). Due to the disparity in formats and quality of the digitized contracts, byte size as suggested by Robinson & Stuart, (2007), seems to be unsuitable as a proxy. Throughout the data collection all of the words that each contract contains are counted and totaled per alliance. Thus, I will rely on the amount of words per contract as a proxy for contract complexity. The subsequent variable will be coded as continuous variable. However, I use a logarithmic transformation to prevent any skewness, extreme low or high values might create.

Equity Investment. Research suggest that equity investments in the partner, or higher equity investments within an alliance, might substitute for ownership (Robinson & Stuart, 2007). This again is especially salient in uncertain environments such as the alliances I have observed (Sampson, 2007). Therefore, I control for these possible effects of mutual equity investments. Equity investment was coded ‘1’ if one of the parties has used or has the option to do an equity investment in the other partner.

as neoplasms. Therefore, ‘Immunotherapy’ and ‘Chemotherapy’ are coded as well. Focal Technology is then turned from a categorical variable into dummy variables per category.

Therapeutic area. Prior research has shown that success rates for new drugs differ depending on the therapeutic area (DiMasi, 1995, 2001). Therefore, I have created a control variable based on the typology of Dimasi et al., (2010). Dimasi et al., (2010), uses seven typologies and one miscellaneous: antineoplastic/immunologic, cardiovascular, Central Nervous System (CNS), Gastrointestinal GI/metabolism, musculoskeletal, respiratory, systemic anti-infective and miscellaneous. I have coded the alliance compounds and focus areas according to the typologies. Thus, I can simply add therapeutic area as a dummy control variable, in which miscellaneous will not be included and serves a baseline model. This allows me to use Stata to automatically compute the dummy variables.

Same firm type. Despite the fact that the variable knowledge proximity is a measure for technological similarity, it is likely that this measure does not capture all of the similarity effects between partners. After all, knowledge proximity is a measure of technological similarity, thus, similarity in terms of capabilities (Jaffe, 1986). Therefore, we also control for alliances that consist of partners that are both pharmaceuticals, biotech or any other same firm types (Devarakonda & Reuer, 2018). These alliances might have different dynamics because it is possible that these are directly competing firms as they can be from the same sub-industry. I coded the partners along an eight-scale typology, ranging from 1 to 8. I then subtracted both typologies of the partners. If the subtraction resulted in ‘0’ it meant no difference was present and the alliance is coded same firm type ‘1’.

R&D Alliance. In addition, I control for the alliance type. The sample exists of multiple alliance types, however, the likelihood performance decreasing effects due to technological uncertainty of early phase development are likely to be especially present in R&D alliances (Devarakonda & Reuer, 2018; Oxley & Sampson, 2004). Therefore, I have created the binary variable R&D Alliance which is coded as ‘1’ if the alliance was an R&D alliance.

Deal size. There is some evidence, although limited, that deal size has a positive effect on the progression of the compound through the phases (Kim, 2011). To control for this effect I measure deal size as the total sum of the possible payments that can be earned throughout the entirety contract (Devarakonda & Reuer, 2018; Reuer & Devarakonda, 2016; Robinson & Stuart, 2007). I then us a logarithmic transformation to correct for any positive skewness this measure might result in (Reuer & Devarakonda, 2016).

International alliance. There is evidence that the aforementioned effects of trust after a prior relationship and governance mechanisms differ according to whether it is a domestic or international alliance (Gulati, 1995). Therefore, I control for possible diverging effects by creating a binary variable that is coded ‘1’ if it was an international alliance and ‘0’ if it was a domestic alliance

Phase fixed effects. As we have seen before, both the technology and therapeutic type might have significant effects on a shift between phases (Dimasi et al., 2010). In addition, the success rates differ per phase across these different technologies and therapeutic areas (Dimasi et al., 2010). Moreover, the coordination and uncertainty might also differ depending on the stage of development the alliance is currently in (Rothaermel & Boeker, 2008). To control for these differences in success rates per phase I add the variable phase fixed effects. This variable has values ranging from 1 for the discovery phase to 8 for the marketing and distribution of the drugs.

4.4. Analysis

For the analysis the program STATA will be used. The descriptive statistics are analyzed first, after which a correlation analysis will test for a relationship among the variables. The main dependent variable is innovation success which only measures a success or no success. This means it is a binary variable which can only assume values of ‘0’ and ‘1’. Linear regression is a statistical method commonly used method by economic sciences. This method, however, requires a continuous variable. Therefore, a nonlinear model, the Probit regression, seems most appropriate. The Probit model is a probability model with two categories in the dependent variable (Liao, 1994). The values of the binary variable, y, are mutually exclusive and exhaustive. The regression formula is as follows where i is alliances.

Formulae 4.2: Probit regression formula

Innovation Successi = ß0 + ß1 Knowledge Proximityi + ß2 Knowledge Proximityi 2i + ß3 Joint Steering Committeei + ß4 Knowledge Proximityi X Joint Steering Committeei + ß5 Knowledge Proximity 2

i X Joint Steering Committeei +ß6 Alliance Scope + ß7 Knowledge proximityi X Alliance Scopei + ß8 Knowledge proximity 2i X Alliance Scopei +ß9 Control Variables + ε

To prevent and check for multicollinearity within the dataset a Variance Inflection Factor analysis is performed together with manually checking the correlation matrix.

Further, I use the Z-standardization for all continuous independent variables. This allows me to compare a unit of increase for all variables under normal distribution.

5. Results

In this section, I will discuss the results of the analyses that have been done in this study. First, I will discuss the descriptive statistics and correlation matrix. Second, I will present the results of the Probit regression and the marginal effects. Third, I will present a post hoc analysis on innovation speed as an additional regression analysis that I have done to depict further results. Fourth, I will present the results of additional analysis with additional measures to test the strength of the model.

5.1. Descriptive Statistics and Correlations

Table 1 shows the descriptive statistics and the correlation matrix of this study. The sample consists of 242 alliances covering 8 different levels of diseases and 14 different technological areas. From the descriptive statistics in table 1 it can be observed that approximately 20% of the alliances had an Innovation Success. In addition, the mean of Knowledge Proximity is in the lower end (mean = 0.38) with a standard deviation of .35 this might indicate that the values close to ‘1’ were less observed. Furthermore, 50 % of the alliances used a Joint Steering Committee, or a similar committee with oversight responsibility. Alliance Scope is relatively broad (mean = 4.38) as the lowest value is 0 and the highest value is 5.

5.2. Regression results

Table 3 shows the Probit regression analysis and the results to test the three hypotheses. First of all, the log likelihood (LL) improved from model 1 (LL = -82.68) to model 6 (LL = -81.06) this means that the full model is likely to be the best. In addition, we can successfully reject the null-hypothesis because all of the models are significant at the P<0.05 level (Model 6 = Prob > Chi2 = 0.0000). This would lead me to conclude that at least one of the regression coefficients is unequal to zero.

Model 1 is the main starting or baseline model. This model only contains the control variables derived from literature and measures the effect on Innovation Success. Geographical Distance shows a negative and significant value (ß = -0.338, P<0.05) and Contract Complexity shows a positive and significant (yet above the critical significance level of P<0.05) result (ß = 0.275, P<0.1).

In Model 2 the main independent variable, Knowledge Proximity, as well as Knowledge Proximity Squared are added. Hypothesis 1 predicts that there is an inverted U-shaped relationship between the Innovation Success and Knowledge Proximity between the partners. Surprisingly, the baseline relationship Knowledge Proximity is non-significant (ß = -0.337, P>0.1) as well as Knowledge Proximity Squared (ß = 0.343, P>0.1). The relationship coefficient remains positive and non-significant in the further models for the Squared term, yet increases in Model 4 and Model 6 (Model 4: ß = 1.141, P>0.1; Model 6: ß = 1.057, P>0.1). Haans, Pieters, and He (2016) propose a three-step procedure to test for the existence of an (inverted) U-shaped relationship. The first step absolutely necessitates a significant ß2 of a

negative sign to indicate the hypothesized relationship. Therefore, hypothesis 1 finds no support in this research. This means that no significant relationship between Knowledge Proximity and Innovation Success in this research. Geographical Distance remains negative and significant (ß = -0.341, P<0.05).

Model 4 includes the main independent variables, moderators and the first interaction for Joint Steering Committee and Knowledge Proximity and Knowledge Proximity Squared to test the second hypothesis. Hypothesis 2 predicts a ‘positive’ steepening effect of the inverted U-shaped relationship because of the strengthening effect of the Joint Steering Committee. The interaction term of Joint Steering Committee remains non-significant for both Knowledge Proximity (ß = 0.817, P>0.1), as well as the Squared term (ß = -1.002, P>0.1). Thus, hypothesis 2 can be rejected. This means that no significant steepening effect of Joint Steering Committee can be found in this study.

In Model 5, I followed the same procedure as in Model 4. I included the independent variables; the moderators and this time included the interaction term of Alliance Scope to test hypothesis 3. Hypothesis 3 predicts that Alliance Scope has a flattening effect on the inverted U-shaped relationship. The interaction effects of Alliance Scope are non-significant for both the baseline variable Knowledge Proximity (ß = 0.416, P>0.1) and the Squared (ß = -0,542 P>0.1). Therefore, hypothesis 3 is not supported and is rejected. This means that no significant weakening effect of Alliance Scope can be found in this study.

5.3. Marginal effect analysis

Despite no significant relationship between Knowledge Proximity Squared and Innovation Success is found, a marginal effect analysis can further exclude significant values at extremes. This study proposes an inverted U-shaped relationship between knowledge proximity and Innovation Success. Thus, one would expect the marginal effects to first increase and then decrease if such a relationship is present. The margins plot depicts no such relationship and shows no difference in marginal effects. Further, the confidence interval increases as knowledge proximity increases confirming non-significant results from the regression. Further calculation of the actual marginal effects indicated positive but non-insignificant marginal effects (dy/dx = 0.0021851, P>0.1).

The marginal probabilities for the interaction of Joint Steering Committee are increasing and significant (P<0.01) (Appendix B). However, no statistically significant difference in marginal effects was found for the ‘presence’ or ‘absence’ of a Joint Steering Committee (P > 0.1). In addition, the margin plot depicts almost identical lines confirming the non-significant findings for the presence of a Joint Steering Committee.

plotted lines have almost completely overlapping confidence intervals, confirming the non-significant findings. Therefore, despite these increasing marginal effects, we cannot assume a significant positive relationship. Hypotheses 1, 2 and 3 have rightfully been rejected based on these post-hoc tests. An overview of the plotted marginal effects is presented in Appendix B.

5.4. Robustness checks

To substantiate my findings, I computed additional tests with different variables and amounts of data. First, I deleted all the observations from my dataset that did not contain NCT numbers. As mentioned, NCT numbers allowed me to observe the shift form one phase to another. The absence of such an NCT number led to an immediate ‘0’ or ‘No’ for Innovation Success. Therefore, it is interesting to see if a regression, using only observations with NCT numbers, will get a different result. After all, no NCT number might have different reasons, I will further discuss this as an limitation in the discussion section. Using the same variables and dummies I re-performed the regression with a total number of N=194 observations (Appendix C). The results confirm the robustness of the model. The coefficients and standard deviations show no large deviations from the original model. Further, Knowledge Proximity (ß =1.116, P > 0.1) and Knowledge Proximity Squared (ß = 1.128, P > 0.1) remain non-significant and the interaction effects remain non-significant as well. Hence, I can assume that no real effect was present by including data lines with missing NCT numbers.

Furthermore, the data contained knowledge proximity values of ‘0’ if no overlap was found or indicated. This might have created some skewness and bias because the knowledge proximity values are not equally represented (N= 84 for ‘0’, N= 0 for ‘1’). Thus, I re-performed an additional analysis without all the Knowledge Proximity values at ‘0’ at N=159 across all the models (Appendix D). Knowledge Proximity and Knowledge Proximity Squared remain non-significant. I can rule out that the non-significant findings were fueled by the presence of low proximity values across the entire range of success.

5.5. Post-hoc analysis

To test the existence of this relationship I ran a Poisson regression because Innovation Speed is a non-integer positive count variable (Appendix E).

The Poisson regression is significant across all models (Prob>Chi2 = 0.0000), the loglikelihood is steadily increasing access all models (Model 1: ll = 3150; Model 6: ll = -3104). Knowledge Proximity shows a strong, positive and significant relationship with Innovation Speed (ß = 0.094, P<0.001). Even more, Knowledge Proximity Squared shows a negative relationship (ß = -0.081, P<0.001). The negative coefficient points to the possible existence of an inverted U-shaped relationship (Haans et al., 2016). Further, the interaction term of Joint Steering Committee for Squared is statistically significant in model 4 (ß = 0.090 P<0.1) but above the critical level of (p < 0.05) an loses all significance in Model 6 (ß = 0.065, P>0.1). The interaction term of alliance scope with Squared is positive and significant in model 5 (ß = 0.069, P<0.05) but loses significance until beyond the critical point in model 6 (ß = 0.054, P<0.1).

6. Discussion and Conclusions

The results presented in the previous chapter hold several theoretical contributions. First, research is among the first to examine a relationship between knowledge proximity and alliance success. The results of this study show no significant relationship between knowledge proximity and innovation success (H1). This suggests that according to this research, the overlap between partners in terms of technological resources and capabilities has no innovation increasing or decreasing effects. More specifically, it does not seem to influence the ability of pharmaceutical firms to increase the successful completion of clinical trials.

Nevertheless, this is somewhat contrary to other literature that finds relationships between knowledge proximity and innovation performance (Sampson, 2007). Although, that research established a significant positive impact on innovation, their measure on innovation is based on patent output (Ahuja & Katila, 2001; Devarakonda & Reuer, 2018; Lane & Lubatkin, 1998; Patel & Pavitt, 1997; Rosenkopf & Almeida, 2003). However, patents do not always capture the quality and product-level success of the actual invention and tell an incomplete story (Sampson, 2007). For example, large firms often have many unexploited patents, some of which purely serve the purpose of blocking out competitors (Giuri et al., 2007). The measurement of innovation success used in this research is an alliance level predictor of the actual product-level success of the innovation (Hoang & Rothaermel, 2005). This measurement is more appropriate if the question is related to the actual innovation success in terms of market success and quality.

The non-significant results for hypothesis 1 might therefore be explained for exactly that reason. It is likely that the success that results from knowledge proximity is better reflected in knowledge outputs that directly resemble a successful flow of information and recombination, such as patents (Kavusan et al., 2016). Therefore, the research productivity of the alliances in terms of patents might be higher but not the quality and quantity of the innovative output and therefore challenges (Lanjouw & Schankerman, 2004), but builds on and substantiate the results of Makri et al. (2010).

explained as an increase informational uncertainty and coordinative complexity (Sampson, 2007). Knowledge proximity as a proxy for opportunistic behavior from a transaction cost perspective is currently under developed (Devarakonda & Reuer, 2018; Reuer & Devarakonda, 2016). With this research, I strengthened the theoretical indications that knowledge proximity also has opportunism implications, that is, misappropriation for high levels of knowledge proximity and information asymmetry for low levels of knowledge proximity (Kim, 2011; Lee et al., 2017). I thus shed new light on how knowledge proximity both influences the ability and willingness to share knowledge.

Third, this research builds further on the literature of Devarakonda and Reuer (2018), that found that steering committees mitigate increased knowledge spillovers when overlap between firms is high. Contrary to my expectations about governance mechanisms – specifically joint steering committees – I do not find the significant moderating effect of a steering committee on the relationship between knowledge proximity and innovation success, as suggested in hypothesis 2. I tried to show how a joint steering committee facilitates the adaptation and coordination of partners’ behavior (Reuer & Devarakonda, 2016; Schilke & Lumineau, 2018) and affects the relationship between knowledge proximity and innovation success. This result challenges the research of Devarakonda and Reuer (2018) and Reuer and Devarakonda (2016), that predict overall positive effects. A possible explanation and implication is that (1) a steering committee might not reduce the opportunism as much as expected, or (2) if it does, it might not be as positively related to increases in innovation success as expected.

the lower the probability for a broad scope (Oxley & Sampson, 2004). This research extends that finding by also finding no significant weakening effect on the relationship between knowledge proximity and innovation success. This might indicate that despite the assumptions in the literature a broad alliance scope might not have the detrimental effects as predicted.

Lastly, this research add to a stream of more realistic measures for innovation success (Hoang & Rothaermel, 2005, 2010). As mentioned, research largely relies on patent measures, R&D spending (Lanjouw & Schankerman, 2004) and increased sales (Dyer, Powell, Sakakibara, & Wang, n.d.) as proxies for innovative capabilities and performance. This research tried to uncover the effects of knowledge proximity on actual product-level success, reflected in a shift from one clinical phase to the subsequent phase. After all, the successful transition from one phase to the next captures the research and development capabilities of both the alliance partners (Dimasi et al., 2010; Hoang & Rothaermel, 2005). By using the measure of Hoang and Rothaermel (2005), I find no apparent relatiosnhip between knowledge proximity and innovation succes, this raises the question whether ‘innovation performance’ in earlier research is also reflected in actual product-level succes.

6.2. Managerial implications

This research holds several managerial implications. First, according to this research, knowledge proximity of alliances does not directly impact the ability of partners to would allow a compound to shift from one phase to the next phase more often. Nevertheless, it theoretically does impact the ability to build on each other’s knowledge (Cohen & Levinthal, 1990; Devarakonda & Reuer, 2018; Sampson, 2007). Thus, while it does not seem to impact product level success, it is well established that increasing levels of knowledge proximity allow for higher levels of absorptive capacity, which in turn increases the likelihood of mutual understanding (Ahuja & Katila, 2001; Cohen & Levinthal, 1990; Sampson, 2004). Managers should therefore primarily be motivated by the intention of the alliance instead of the absorptive capacity argument to determine partner-fit (Sampson, 2007). After all, the apparent knowledge proximity between partners does not seem to impact product-level innovation success.

therefore not be answered on the basis of this research. This also means that instituting a joint steering committee might not outweigh the costs, if, it is not reflected in corresponding increases in innovation success. Ultimate positive effects might be contingent on factors beyond this research. Managers should therefore take more than just partner similarity into account when deciding on the institution of a joint steering committee.

Third, according to my study, the breadth of the alliance scope does not seem to increase or decrease the effects of knowledge proximity. That means that when choosing the breadth of the alliance, managers should primarily focus on which value chain activities are necessary for the alliance success. Despite earlier research that predicts complications in terms of coordination, monitoring and unintended knowledge spillovers (Oxley & Sampson, 2004; Reuer et al., 2002), the negative effects do not seem to be reflected in negative moderating effects on knowledge proximity and innovation success.

6.3. Limitations and future research

This research holds several limitations that highlight interesting future areas of research. First, the industry of analysis is the U.S. bio – pharmaceutical industry, which has certain specific characteristics that might influence the results. One of these characteristics is the external innovation behavior of large pharmaceutical companies, that form alliances with small new entrants (Dong & McCarthy, 2019). In itself this might limit the opportunistic behavior due to the disbalance of power between the partners, power asymmetry (Hill & Rothaermel, 2003). Therefore, power asymmetry might balance the effects of information asymmetry (Kim, 2011), that is, smaller firms might not be willing to use their informational advantage over the larger firm because of the dependence on the large firm to provide money and resources (Katila et al., 2008). Future research might therefore try to uncover whether those dynamics have an effect on opportunism in alliances.

Second, the results might not be generalizable because of cultural and industry differences worldwide. The used sample only consists of alliances within the biopharmaceutical industry. Nevertheless, there are different industries that also form alliances with different levels of knowledge proximity, surrounded by different industry characteristics that might imply different effects. Future research should therefore dive into, and compare, the effects of knowledge proximity in different industries.