Reconfiguring the Firm’s Knowledge Base: The Effects of Knowledge Breadth and Knowledge Depth on the Formation Frequency of R&D Alliances.

(1)

Reconfiguring the Firm’s Knowledge Base: The Effects of Knowledge

Breadth and Knowledge Depth on the Formation Frequency of R&D

Alliances.

Master Thesis University of Groningen Faculty of Economics and Business MSc BA Strategic Innovation Management

Supervisor: dr. I. Estrada Vaquero Co-Assessor: dr. J.D. van der Bij

Niels Willem Bruggeman

E-Mail: N.W.Bruggeman@student.rug.nl Student number: S2959585

(2)

1

ABSTRACT

Firms increasingly rely on external knowledge sources to remain competitive in high-technology industries. This study investigates the role of the internal knowledge base on formation of R&D alliances by taking into account the structuration of knowledge, distinguishing between knowledge breadth and knowledge depth. In addition, it is tested whether the effects of the knowledge base on R&D alliance formation frequency are moderated by firm size. By testing these effects this study aims to explain the linkages between the firm’s internal knowledge base and its external search activities, especially in R&D alliances. The firm’s internal knowledge base can influence the firm’s ability to leverage value from external knowledge sources, explaining why some firms engage in more R&D alliances than other firms. The current study uses a panel dataset of 161 pharmaceutical and biotechnology firms in the United States in the period from 2005-2012. The findings of a Poisson regression indicate a positive linear relationship between knowledge breadth and the formation frequency of R&D alliances, whereas an inverted u-shape relationship is found for the effect of knowledge depth on R&D alliance formation frequency. Regarding firm size, no significant moderation effects were found. This study increases the understanding on the role of firm characteristics in external knowledge sourcing and provides mangers with knowledge to manage the tensions between internal and external knowledge sourcing activities in order to leverage value.

Key words: Internal knowledge base, knowledge depth, knowledge breadth, firm size, R&D alliance

(4)

3

1. INTRODUCTION

During the last decades a transition took place from firms being internally focused to develop valuable resources towards acquiring resources from external partners (Chesbrough, 2003a). Accessing information from external actors to boost innovation has become an essential part of managerial strategy, as the innovation process is becoming more open, disturbed and democratic (Laursen & Salter, 2014). Competitive tensions among firms are increasingly about whether they can create and commercialize knowledge in a timely and cost-efficient manner, especially in technology intensive industries. Interfirm R&D collaborations are frequently used to expand the firm’s knowledge base with complementary knowledge and capabilities. These collaborations allow for reaping economies of scale in R&D, shorten development time, and spread the risks and costs of R&D (Sampson, 2007). Especially in the pharmaceutical and biotechnology industry, incumbent firms have built their technological knowledge base by collaborating with R&D partners, therefore being the focal industry in this study (Rothaermel and Boeker, 2008; Whittaker and Bower, 1994; Zhang and Baden-Fuller, 2010).

This research is positioned in the knowledge-based view (Grant, 1996), which argues that heterogeneity in firm knowledge is crucial in explaining performance differences. Hence, the firm’s internal knowledge creation processes are critical in grasping strategic opportunities and enhancing revenue streams. Real search processes take place in specific historical contexts, and how firms search for problem solutions depends on these contexts (Nelson & Winter, 1982). These contexts refer to the firm’s past internal knowledge creation strategies. The firm’s prior engagement in internal knowledge creation serves as a source of experience which can strengthen the firm’s capability to leverage the value of extramural knowledge sources, through building higher-level routines (Lewin, Massini & Peeters, 2011; Wuyts & Dutta, 2014). The experience in internal knowledge creation and the increased ability to develop higher-level routines to create value in external knowledge acquisition might therefore influence the firm’s R&D alliance formation frequency. In previous research it was theorized that internal knowledge creation in terms of R&D and external knowledge acquisition are complementary to each other, such that high internal R&D has a positive effect on external knowledge leverage (Cassiman & Veugelers, 2006). Thus, indicating that it is relevant to consider the firm’s internal knowledge base as can affect the value created through R&D alliances. Besides, past research provided evidence that the linkages between internal knowledge creation and external knowledge search are influenced by firm characteristics, such as the reliance on more basic R&D (Cassiman & Veugelers, 2006) and the role of R&D organizational structure (Zhang & Baden-Fuller 2010).

(5)

4

knowledge breadth. Knowledge breadth refers to the extent to which the firm’s knowledge repository contains distinct and multiple domains, whereas knowledge depth refers to the level of sophistication and complexity of knowledge in key fields (Bierly & Chakrabarti, 1996). Second, the sparse previous research is inconclusive about the specific effects of knowledge breadth and depth on external knowledge sourcing. According to Zhang & Baden-Fuller (2010) knowledge breadth positively influences the likelihood of forming new research alliances whereas knowledge depth has a negative effect. In contrast, Zhou & Li (2012) state that in case of knowledge depth firms tend to engage more in external knowledge acquisition, whereas in case of knowledge breadth internal knowledge sharing is preferred. Third, there is still a limited understanding of how firm characteristics are possible drivers in external knowledge creation, and which firm characteristics are most important. This research complements this emerging stream of research by clarifying how firm size affects the linkage between internal knowledge creation and external knowledge sourcing. Organizational size is included as it affects the firm’s flexibility and adaptability (Bierly, 2004) which can affect the firm’s propensity of external knowledge search through R&D alliances.

Addressing the gaps in current literature, by integrating the knowledge-based view (Grant, 1996), is important in explaining how knowledge breadth and depth influence the firm’s ability to create value through R&D alliances and hence its R&D alliance formation frequency. In addition, including firm size as a moderator adds to the current understanding about firm characteristics influencing the linkage between internal and external knowledge. The differences in the firm’s ability to leverage value from R&D alliances, caused by heterogeneity in firm knowledge, enriches the understanding about why some firms have a higher R&D alliance formation frequency than other firms. Besides, it provides managers with valuable knowledge about how their internal knowledge stock relates to interfirm R&D collaboration, and what factors stimulate or prevent the firm from creating value through R&D alliances. This research aims to answer the following research question:

How do the breadth and depth of the firm’s internal knowledge base affect the formation frequency of R&D alliances, and is there a moderation effect of firm size?

This study uses panel-data in the period 2005-2012 collected from the SDC database, USPTO PatentsView and S&P’s Compustat from 250 US-based firms in the pharmaceutical and biotechnology industries. By conducting a Poisson regression, this study finds that the firm’s internal knowledge base influences its decisions to form R&D alliances. Specifically, it is found that knowledge breadth increases the frequency of R&D alliances and knowledge depth has an inverted u-shape effect on the R&D alliance formation frequency. This study does not find a significant moderation effect of firm size on the relationship between the knowledge base and the frequency of R&D alliances.

(6)

5

explaining the positive effect of knowledge breadth and the inverted u-shape effect of knowledge depth, it takes into account the firm’s knowledge structuration and acknowledges that the knowledge base contains distinctive elements. Where previous literature found a negative relationship between knowledge depth and R&D alliance frequency (Zhang & Baden-Fuller, 2010), this study enhances understanding by providing an inverted u-shape relationship. This study clarifies the firm’s ability to create value through R&D partnerships, by synthesizing the relevant dimensions that shape this ability. Furthermore, the findings contribute in enhancing the managers understanding of the firm’s knowledge breadth and depth and its effect on external knowledge sourcing through R&D alliances. With this knowledge managers can improve their ability in coordinating internal and external knowledge sources in order to leverage the benefits from both activities in the firm’s innovation strategies.

The paper is structured as follows, it starts with the conceptual background and hypotheses development. Thereafter, the methodology from this study is outlined, followed by the results section. The next section consists of a discussion, including key findings, implications, limitations and future research recommendations. Finally, the paper will end with a conclusion.

2. CONCEPTUAL BACKGROUND AND HYPOTHESES

2.1 Conceptual Background: Firm Knowledge Base and R&D alliances

Incumbent firms face major challenges caused by competence-destroying technological change (Grigoriou & Rothaermel, 2017). This new technological environment requires firms to invest in developing its knowledge domain. Besides developing knowledge internally, firms can engage in strategic alliances, acquisitions or combinations of those (Grigoriou & Rothaermel, 2017). New knowledge can be accessed via various parties such as competitors, suppliers, customers, universities and research organizations (Belderbos, Carree & Lokshin, 2004). This paper focuses on the formation of strategic alliances to expand the firm knowledge base. Strategic alliances can be defined as ‘’voluntary arrangements among independent firms, that involve the exchange, sharing, or joint development or provision of technologies, products or services’’ (Gulati, 1998: p. 293).

(7)

6

exactly with the product domain. Therefore Grant & Baden-Fuller (2004) argue that firms should focus on building core competencies in their knowledge domain while relying on imported knowledge from alliance partners in more distant knowledge areas.

Nelson & Winter (1982) argue that real search processes take place in specific historical contexts, and how firms search for problem solutions depends on these contexts (Nelson & Winter, 1982). These contexts are the firm’s past knowledge creation strategies that operate a source of experience and shape future search behaviour (Wuyts & Dutta, 2014). Therefore, the firm’s processes to access external knowledge depend on how the firm has developed its internal stock of knowledge over time. Following from this, technological knowledge search can be described as being path-dependent and the existing knowledge stock is a starting point for external knowledge search (Teece, Pisano & Shuen, 1997). External knowledge might be extremely valuable for organizations to stimulate explorative and exploitative behaviours. However, external knowledge can also be difficult to integrate and leverage into the organization’s operations, suggesting that the internal relevance of the knowledge cannot be taken for granted. Organizational members may access diverse sources of knowledge, however they may lack the awareness of what type of knowledge is needed (Tortoriello & McEvily, 2014).

The current research focuses specifically on R&D alliances, which firms increasingly use to complement their knowledge with in order to foster innovation and new product development. Especially in the pharmaceutical and biotechnology industries, R&D alliances are undertaken to build and expand the technological knowledge base. This is due to the fast-changing environment requiring firms to access new technical knowledge in order to stay competitive (Rothaermel and Boeker, 2008; Whittaker and Bower, 1994; Zhang and Baden-Fuller, 2010). Exploring new knowledge through R&D alliances and internal knowledge competencies are interrelated by providing incentives to access new knowledge and shape the scope and direction of future exploration (Katila & Ahuja, 2002). Thus, the internal knowledge base of the firm is critical in shaping the firm’s external knowledge sourcing strategies for R&D.

In this study, the knowledge base can be described in terms of its knowledge breadth and knowledge depth. Knowledge breadth can be described as the extent to which the firm’s knowledge repository contains distinct and multiple domains, as in the number of technological niches in which the firm operates (Bierly & Chakrabarti, 1996; George, Kotha & Zeng, 2008). Whereas, knowledge depth refers to the level of sophistication and complexity of knowledge in key fields, as in the firm’s expertise within a technological niche (Bierly & Chakrabarti, 1996; George et al., 2008).

(8)

7

four dimensions were identified through previous research, and in this paper they are synthesized and combined in order to explain the firm’s R&D alliance formation frequency. How these dimensions are influenced by knowledge breadth and knowledge depth and how it in turn affects the R&D alliance formation frequency will be explained in the hypotheses development.

Table 1: Ability to Create Value through R&D Alliances: Dimensions Dimensions Ability to create value

through R&D alliances:

Influenced by:

Dimension 1 The firm’s ability to scan,

recognize and evaluate opportunities in technological fields, distinguishing between: a) Within technological fields b) Across technological fields

The level of expertise with the relevant knowledge elements

(George et al., 2008; Zhang & Baden-Fuller, 2010)

Dimension 2 The firm’s absorptive

capacity to link recognized opportunities with existing knowledge elements a) Through a broad range of existing knowledge elements b) Through repeated use of existing knowledge elements

The ability to make novel associations between knowledge elements, due to enhanced understanding of existing knowledge, increasing recombination potential

(Cohen & Levinthal, 2001; George et al., 2008; Wu & Shanley, 2009)

Dimension 3 The firm’s flexibility and

adaptability in exploring new knowledge paths through R&D alliances

The inertia in exploring external knowledge due to institutionalized routines for knowledge sourcing

(Leonard-Barton, 1992; Tripsas & Gavetti, 2000; Volberda,1996)

Dimension 4 The firm’s exposure to risks

of opportunistic behaviour by partners when forming R&D alliances

The knowledge spill-overs and learning races

between partners

(Williamson, 1985; Zhang & Baden-Fuller, 2010)

(9)

8

dimensions in table 1. The next section will develop the hypotheses related to the effects of knowledge breadth and depth on R&D alliance frequency as well as how firm size moderates this effect.

2.2 Hypotheses Development

This section aims to provide an overview of the hypotheses development in this paper. The hypotheses development is illustrated in figure 1. It is argued that the firm’s knowledge base, in terms of its knowledge breadth and knowledge depth, influences the firm’s R&D alliance formation frequency. Furthermore, it is argued that firm size moderates the relationships between the knowledge base and R&D alliance formation frequency.

Figure 1: Conceptual Model

As this study aims to reconfigure the firm’s knowledge base in its knowledge breadth and knowledge depth, it is important to distinguish between how they will affect the firm’s ability to create value through R&D alliances separately. Table 2 provides an overview of the main differences as will be explained further in the following sections that develop the hypotheses.

Table 2: Comparison Knowledge Breadth and Knowledge Depth

Knowledge Breadth Knowledge Depth

Dimension 1 Increased abilities to scan across

technological fields for R&D alliances

Increased abilities to scan within technological fields for R&D alliances

Dimension 2 Enhancement of absorptive capacity

through a broad range of knowledge

elements increasing recombination of

internal knowledge with external knowledge from R&D alliances

Enhancement of absorptive capacity through repeated use of existing

knowledge elements increasing

recombination of internal knowledge with external knowledge from R&D alliances

Dimension 3 Increases flexibility and adaptability in

establishing R&D partnerships and exploring new knowledge paths

Emergence of core rigidities and

dominant mindset, leading to inward focus

and hampers ability for external R&D collaboration

Dimension 4 Limited risks regarding partner’s opportunistic behaviour because of

tacit nature of linkages and interaction between knowledge elements

Substantial risks of opportunistic

behaviour caused by learning races among

(10)

9

2.3 Knowledge Breadth and the Formation of R&D Alliances

Technological knowledge breadth can be considered as one of the most important dimensions of the internal knowledge base, therefore being widely studied (e.g. Leiponen and Helfat, 2010; Prencipe, 2000; Zhang, Baden-Fuller & Mangematin., 2007). Having a broad knowledge base increases the firm’s capability to combine knowledge across different fields in a more complex and creative way (Bierly & Chakrabarti 1996; Kogut & Zander 1992; Reed & DeFillipi 1990). In order to recognize and develop new business opportunities, firms should scan and search across multiple technological niches. Especially knowledge that is distant to the firm and unrelated to the existing knowledge base provides relevant exploration paths (Teece, 2007). Considering how knowledge breadth affects the firm’s ability to create value through R&D alliances and hence the formation frequency of R&D alliances, it is argued that of the dimensions in table 1, dimension 1a and dimension 2a are primarily associated with knowledge breadth. Dimensions 3 and 4 are less important but still need to be considered in order to compare the effects with knowledge depth.

To start, knowledge breadth enhances the firm’s ability to create value through R&D alliances by allowing the firm to track changes and developments across multiple technological field. Since it enhances the firm’s ability to scan, recognize and evaluate valuable opportunities for exploration, as outlined in dimension 1a (George et al., 2008). These abilities are enhanced due the expertise and the experience that the firm has developed across technological niches. Firms with limited knowledge breadth are hampered when acquiring knowledge from outside of their own technological niches, due to lacking expertise (Cohen and Levinthal 1990; Rosenberg 1982; Zahra and George 2002). As a result, the firm might enter into R&D collaborations that firms with a broad knowledge base would avoid. Therefore, when having a broad knowledge base, firms are likely to enter into more alliances because of increased screening, recognition and evaluation capabilities.

Second, the firm’s ability to create value through R&D alliances is influenced by knowledge breadth through dimension 2, which refers to the notion of absorptive capacity. Cohen & Levinthal (1990; p. 128) define absorptive capacity as the ‘’ability to recognize the value of new information,

assimilate it, and apply it to commercial ends’’. This ability is largely a function of the firm’s stock of

(11)

10

Third, firms with high knowledge breadth are expected to create value through R&D partnerships without largely being exposed to core rigidities (Leonard-Barton, 1995). A broad knowledge base allows for more flexibility and adaptability in dealing with fast changing market preferences and technological opportunities, as outlined in dimension 3 (Volberda, 1996). Due to the increased flexibility and adaptability, firms with high knowledge breadth have more options in establishing novel linkages between knowledge elements and are less likely to be hampered by core rigidities in their formation of R&D alliances.

Finally, firms with a broad knowledge base might be at risk when entering into R&D alliances, due to knowledge spill-overs and opportunistic behaviour from partners, as specified in dimension 4. However, firms with a broad knowledge base have knowledge that is difficult to replicate because the knowledge is formed by distinctive components and their system of interactions reside in informal communication channels shared by firm R&D groups (Henderson and Clark, 1990 ;Lave and Wenger, 1991). Therefore, it is difficult to entangle how the knowledge linkages of the firm are built up, hampering the partner’s ability to replicate the knowledge. Therefore, firms with a broad knowledge base are not largely affected by risks of opportunism when entering into R&D alliances (Zhang, 2007). In sum, firms with a broad knowledge base are expected create more value through R&D alliances hence resulting in higher R&D alliance formation frequency. The value creation potential is enhanced through four dimensions as it increases the ability to scan, recognize and evaluate opportunities across technological fields, and leads to a higher absorptive capacity to make novel associations between distant and existing knowledge elements. Due to knowledge breadth and its accompanied flexibility and adaptability, firms are less hampered by core rigidities. Besides, they face ‘’lower’’ risks of opportunistic behaviour by partners because of the tacit nature of the linkages between knowledge elements. Thus, the effect of knowledge breadth on R&D alliance formation frequency can be hypothesized as:

H1: The breath of a firm’s knowledge base has a positive effect on the frequency to which that firm forms R&D alliances.

2.4 Knowledge Depth and the Formation of R&D Alliances

(12)

11

First, as with knowledge breath firms with knowledge depth have an increased ability to scan, recognize and evaluate opportunities for R&D collaboration. However, for knowledge depth it applies to within technological field collaboration, as outlined in dimension 1b. Since the firm already has expertise in the technological field, it will be better able to evaluate the value of knowledge across the organization’s boundaries. In case of low depth, firms are more likely to experience technological lockout (Cohen & Levinthal, 1989). Given their lack of knowledge depth and hence expertise, firms with low depth are less able to scan the environment for R&D partnerships opportunities. As a result, they select R&D partners that firm with high depth would not collaborate with (Prahbu et al., 2005).

Second, knowledge depth increases the firm’s ability to create value through R&D alliances by strengthening absorptive capacity, as described in dimension 2b. Enabling firms to understand new information gathered from external sources better (Cohen and Levinthal, 1990; Zahra and George, 2002). Knowledge depth enables a firm to recognize causal linkages within particular technological fields (March, 1991), allowing the firm to select the right knowledge components to recombine. Besides, the recombination process is more efficient since firms understand the limitations of existing knowledge components through repeated use (George et al., 2008). Thus, in the case of knowledge depth, absorptive capacity relates to a refined understanding of knowledge and of its limitations through repeated use. The increased absorptive capacity allows for recombination and value creation through R&D alliances hence leading to a higher R&D alliance formation frequency.

Third, knowledge depth influences the firm’s adaptability and flexibility in exploring new knowledge paths and hence the value created through R&D alliances, as outlined in dimension 3. Expanding the firm’s knowledge base in terms of knowledge depth is considered difficult, since technological uncertainties create barriers that firms must overcome (Mitchell & Singh, 1992). In case of high knowledge depth firms seek solutions within their existing knowledge by using best practices and organizational routines. However, this may result in inertia and core rigidities when trying to explore new knowledge paths (Leonard-Barton, 1992; Tripsas & Gavetti, 2000). When a firm has a deep knowledge base, a dominant mindset will emerge within the organization (Bettis & Wong, 2003; Prahalad & Bettis, 1986). As a result, the firm might avoid partners that deviate from the dominant mindset. When the potential partner does not match with the firm’s organizational routines and mindset it is less likely that R&D alliances will be established. Thus, when the firm’s knowledge depth increases they become less open to external knowledge sources, caused by organizational inertia and internal resistance, and will form less R&D alliances.

(13)

12

knowledge is an important competitive asset of firms in R&D intensive industries and sharing knowledge may threaten the firm’s competitiveness (Conner & Prahalad, 1996; Inkpen & Tsang, 20007). Therefore, when the knowledge base is deep, the firm is less willing to form R&D alliances because of knowledge appropriation by partners.

In sum, knowledge depth influences the firm’s ability to create value through R&D alliances in such a way that it increases the firm’s ability to scan, recognize and evaluate opportunities for R&D collaboration. Besides, it enhances absorptive capacity allowing for more efficient recombination. However, a turning point exist where the firm’s knowledge depth will cause core rigidities and internal resistance, hampering its ability to create value through R&D alliances. Besides, the risks of opportunism increase with knowledge depth, therefore firms with really high knowledge depth face more risks regarding opportunistic behaviours from partners and form less R&D alliances. Thus, knowledge depth first increases R&D alliance formation frequency, but after the turning point increasing knowledge depth leads to a decreasing R&D alliance formation frequency. This can be hypothesized as:

H2: The depth of a firm’s knowledge base has an inverted u-shape effect on the frequency to which that firm forms R&D alliances.

2.5 Knowledge Breadth and R&D Alliances: The Moderating Role of Firm Size

Increasing a firm’s knowledge base in terms of its breadth, is accompanied by significant challenges. Especially, in terms of the firm’s coordination capabilities to deal with the diversified and distinctive knowledge while preventing information overload (Laursen & Salter, 2006). Managing broad knowledge and the ties between the different knowledge elements can be considered as a complex activity, making it difficult to utilize diversified know-how throughout the firm (Katz & Du Preez, 2008). In general, smaller firms can be more nimble, flexible and adaptive to the external environment (Gopalakrishnan & Bierly, 2006). Whereas, large firms experience institutional insulation and bureaucratization, decreasing the firm’s ability to respond to technological and environmental changes (Haveman, 1993). Furthermore, small firms are recognized for having simpler structures allowing for better internal communication processes, facilitating rapid and effective communication throughout the organizations’ functional areas (Cordero, 1991; Gopalakrishnan & Bierly, 2006 Meyer, 1993; Wheelwright & Clark, 1992).

(14)

13

evaluate R&D opportunities. In terms of absorptive capacity it becomes harder to assimilate and recombine knowledge elements because of the firm’s knowledge dispersion.

Moreover, as firm size increases, the flexibility and adaptability to transfer outside knowledge throughout the organization decreases because of internal resistance (Rosenkopf & Nerkar, 2001). When large firms engage in external knowledge sourcing, they are hampered by social conflicts including the social and psychological costs associated with altering habits and routines (Oliver, 1997). Firm size influences the firm’s flexibility and adaptability to form R&D alliances in case of knowledge breadth as described in dimension 3. Increasing firm size will cause conflict and core rigidities in linking internal knowledge with external knowledge. This will affect the firm’s ability to create value through R&D alliances and hence the formation frequency.

Thus, firm size negatively moderates the relationship between knowledge breadth and the formation frequency of R&D alliances by affecting the ability to create value through R&D alliances. Due to decreasing flexibility, adaptability and internal communication efficiency, the firm’s ability to scan, recognize and evaluate R&D opportunities decreases. Furthermore, the recombination opportunities from absorptive capacity diminish and more conflicts and core rigidities will arise. Although large firms have advantages due to a higher resource availability, these resources might not be employed efficiently due to core rigidities and lacking coordination. Thus, the following hypothesis is proposed:

H3: The relationship between the breadth of the firm’s knowledge base and the formation

frequency of R&D alliances is negatively moderated by firm size.

2.6 Knowledge Depth and R&D Alliances; The Moderating Role of Firm Size

Especially when firms have a deep knowledge base, they are at risk because of environmental or technological change, possibly destroying existing competences and undermining the usefulness of the deeply embedded knowledge base (Henderson & Clark, 1992; Tusham & Anderson, 1986). Firm size affects the relationship between knowledge depth and the formation frequency of R&D alliances by affecting the firm’s ability to create value through R&D alliances. Firm size affects dimension 1b, the firm’s ability to scan, recognize and evaluate opportunities for R&D alliances within technological fields and dimension 2b, the firm’s absorptive capacity to utilize the knowledge for recombination. As with knowledge breadth, increasing firm size results in coordination problems and decreases the efficiency of internal communication processes. This can result in knowledge being too dispersed across the organization in order to utilize it for creating value through R&D alliances.

(15)

14

allocation or internal power structures will therefore be less likely to be considered (Oliver, 1997). Thus, firm size affects the flexibility and adaptability of the firm by strengthening the core rigidities and internal resistance that go with knowledge depth, which decreases the ability to create value through R&D alliances.

Firm size also affects dimension 4, the firm’s knowledge protection and perceived risks of opportunistic behaviour. In this paper, it was already identified that as the firm’s knowledge depth increases they face greater risks of knowledge appropriation and opportunistic behaviour from partners. However, when firm size increases firms have more resources available to increase their levels of knowledge protection (Norman, 2002). Therefore, when firm size increases firms with a deep knowledge base face less risks of opportunistic behaviour and increase their ability to create value through R&D alliances.

Inverted u-shaped relationships can be moderated in two distinctive ways. First, the turning point can shift either to the left or the right. Second, the elasticity of the U-shape can be affected in such a way that it flattens or steepens (Haans et al., 2016). This study argues for an additive steepening where the moderator strengthens the curvilinear mechanism involved in the relationship between knowledge depth and the formation frequency of R&D alliances.

Increasing knowledge depth is especially beneficial for small firms as they do not yet encounter severe core rigidities and internal resistance in their external knowledge sourcing activities. At the same time increasing firm size allows them to better protect their knowledge due to growing resource availability. Together, these forces increase the smaller firm’s ability to create value through R&D alliances and hence R&D alliance formation frequency, steepening the positive curvilinear effect until the turning point.

However, as size increases firms encounter severe core rigidities and internal resistance. At higher levels of knowledge depth core rigidities and a dominant mindset are already present in the firm, the core rigidities and internal resistance strengthen when firm size increases. Thus, increasing firm size strengthens the firm’s core rigidities and internal resistance caused by a lack in adaptability and flexibility. The resource availability of a firm increases with firm size, however due to a lack of flexibility and adaptability larger firms are limited in utilizing these resources efficiently. In conjunction, these forces decrease the larger firm’s ability to create value through R&D alliances and hence decrease R&D alliance formation frequency, steepening the negative curvilinear effect after the turning point. Thus, the moderation effect can be hypothesized as:

(16)

15

3. METHODOLOGY

3.1 Research Setting

In order to test the proposed conceptual model, this study takes the biotechnology and pharmaceutical industry as its research context. These industries are characterised by the intensive usage of new technological knowledge, spread among incumbent companies, new biotech firms, universities and research centres. Due to the dispersion of knowledge, organizations need to rely on each other’s knowledge to develop and commercialize products, resulting in very high levels of alliance activity (Powell, Koput & Smith-Doerr, 1996). Besides, the biotechnology and pharmaceutical industries are characterized by a high reliance on knowledge that is captured in patents (Cohen, Nelson & Walsh, 2000; Levin et al., 1987; Mansfield, 1986), allowing to identify the knowledge profiles of the sample firms (Prabhu et al., 2005).

3.2 Sample and Data Collection

Taking the biotechnology and pharmaceutical industry as the research context, this study focuses on firms in the United States in the period between 2005-2012. The focus is on the United States because it is the largest market for the biotechnology/pharmaceutical industry, and therefore firms often patent there, before they patent in other countries (Albert, Avery, Narin & McAllister, 1991; Rothaermel & Boeker, 2008). Furthermore, this study uses the Securities Data Company (SDC) database (SDC, 2019) and as this database originates from the United States it might be biased toward including firms from the United States (Schilling, 2009). The availability of the SDC database was limited since I only had access to alliance data up to the year 2012, therefore this period was chosen since it included the most recent data.

Panel data was gathered for the 250 largest firms in terms of their turnover in 2012, collected through the Orbis database. Gathering data about the alliance activities of small or privately-owned firms is considered difficult and hence the focus of this study is on leading companies in the industry with a large turnover (Ahuja, 200: Gulati, 1995; Lin & Wu, 2010). The firms were identified by using the industrial classification codes, in this case focussing on the firms with SIC code 283. This code corresponds to the industry ‘’drugs’’.

In order to map the firm’s knowledge profile, patent data is collected from the United States Patent Office (USPTO), specifically through USPTO’s PatentsView. The USPTO PatentsView offers opportunities for exploring 40 years of patent data. It is a data visualization and analysis platform aiming to increase the value, utility and transparency of US patent data (USPTO, 2019). In the section discussing the operationalization of the variables it will be discussed why this study uses patent data.

(17)

16

Oxley & Sampson, 2004). SDC contains data from SEC filings, trade publications, wires and news sources. Furthermore, it provides insights in a wide range of collaboration types amongst which joint ventures, strategic alliances, R&D agreements, marketing agreements, manufacturing agreements, supply agreements and licensing pacts (Schilling, 2009). For this study the focus is on the strategic alliances containing an R&D agreement.

For identifying data on the firm characteristics of 250 companies, the S&P Compustat database was used. This database was established in 1962 and provides data over 99 000 global securities, covering 99% of the world’s total market capitalization. In sum, this study combines data from USPTO, SDC and Compustat in order to provide panel data about the pharmaceutical and biotechnology industry in the period from 2005 until 2012.

The sampling procedure started with collecting data from the SDC database and USPTO’s PatentsView for the 250 selected firms. A few firms were removed because in the period before 2012 they were still a subsidiary of a larger company. Abbvie was for instance a subsidiary from Abbot till 2012. Then, when using Compustat to find firm data, many missing data were encountered. The firm year observations that included missing data had to be removed from the sample and this resulted in 1038 firm year observations and 161 firms in the final sample. The following section will present how the variables used in this study were operationalized.

3.3 Operationalization and Measurement of the Variables

3.3.1 Dependent Variable

R&D Alliance Formation Frequency

In order to capture the firm’s R&D alliance formation frequency, this study measured the firm’s number of alliances that were characterised by an R&D agreement, as specified in the SDC database. This implies that the dependent variable in this study is a count variable, as it refers to the number of alliances which can only take discrete non-negative integer values (McCullagh and Nelder, 1989). The variable was operationalized as the cumulative number of alliances in the period from t to t-2 (i.e. sum of alliances formed in t, t-1 and t-2) (Lin & Wu, 2010). Thus, a cumulative measure was taken to capture the variation across years, since the number of R&D alliances in one single year may not correspond with the long-term collaboration strategy of the focal firm (Lin & Wu, 2010). Alliance data was collected between 2003 and 2012, where for example the firm’s R&D alliance formation frequency in 2005 was measured as the sum of alliances in 2003, 2004 and 2005.

3.3.2 Independent Variables

(18)

17

knowledge. Accordingly, having multiple patents represents a firm's collection of knowledge elements. Identifying the collection of patents that a firm is familiar with, assists in identifying the revealed knowledge base of a firm. The patents that a firm possesses show the knowledge that a firm has accumulated and for which it is acknowledged as the creator (Jaffe, Trajtenberg & Henderson, 1993)’’.

For the variables knowledge breadth and knowledge depth, a 3-year window was adopted. Meaning that a firm’s knowledge base in year t includes all the granted patents (i.e. cumulative number of patents) in the past 3 years operationalized as the sum of patents in t-1, t-2 and t-3. Year t was not included in our measure, since it takes time before knowledge is assimilated in the organization’s knowledge stock and can be efficiently utilized in the firm’s collaboration strategies. A window was adopted to gather data more accurately, since annual fluctuations exist in patent data. Additionally, it is likely that firms rely on their stock of knowledge instead of their knowledge in one single year (De Carolis & Deeds, 1999; Grant & Baden-Fuller, 2004). Some previous studies adopted a 5-year window for the knowledge measures (Wu & Shanley; 2009; Zhang & Baden-Fuller, 2010) others adopted a 3-year window (Lin & Wu, 2010; Prabhu et al., 2005). In this study a 3-3-year window is adopted since the value of knowledge tends to decay fast over time in high-technology industries, where there is uncertainty and rapid technological change (Prabhu et al., 2005). To measure a firm’s cumulative number of granted patents in for instance the year 2012, I calculate the sum of number of granted patents in 2009, 2010, and 2011. This study follows previous work and applies four digit application codes to capture the technological patents classes, in this case the CPC classifications which are highly comparable to IPC classifications are used (Lerner, 1994; Wu & Shanley; 2009). The four digit codes show the technological subclasses (See appendix 2).

Knowledge Breadth

In this study knowledge breadth is operationalized by measuring the cumulative number of technological subclasses of the granted patents. This is in line with previous work researching the effects of knowledge breadth (George et al., 2008; Prabhu et al., 2005; Zhang & Baden-Fuller, 2010). Previous work uses different windows to operationalize knowledge breadth, as mentioned in this study a window of 3-years is adopted. Appendix 1 shows exactly how the measure for knowledge breadth was obtained. Knowledge breadth can be operationalized in different ways, however the measure used here is found to be most intuitive (Prabhu et al., 2005). The number of different technological classes in which a firm possesses patents, serves as a proxy to determine the broad distribution of technological knowledge (Agyres & Silverman, 2004).

Knowledge Depth

(19)

18

al., 2005; Wu & Shanley, 2009). This was done by computing the sum of patents in all technological subclasses divided by the total number of subclasses in the firm’s knowledge base in a 3-year window. By doing this a single value for knowledge depth can be obtained. For example, if the firm’s knowledge depth is 5, this means that on average it has 5 patents in all its subclasses. Appendix 1 clarifies how the measure for knowledge depth was computed.

3.3.3 Moderating Variable Firm Size

The current study measures firm size as the log of the number employees in thousands (e.g. Bianchi et al., 2016; Lin & Wu, 2010; Zhang & Baden-Fuller, 2010; Zhou & Li, 2012). A log transformation was used to normalize the data. In this study number of employees in logs was adopted because it can be considered more stable over time and is less hampered by macroeconomic changes than for instance firm sales (Bianchi et al., 2016). Applying a lag-structure for variables assist in alleviating endogeneity concerns (Belderbos, Gilsing, Lokshin, Carree, Sastre, 2018 ), therefore firm size is operationalized as the firm’s number of employees in year t -1. As a robustness check, firm size will also be operationalized as the firm’s total assets.

3.3.4 Control Variables

In the proposed model a number of control variables are included, in line with previous research. First, the effect of firm age is controlled for by adding a variable indicating the number of years since the company’s establishment (Rothaermel & Boeker, 2008). Firm age changes the organizational context in which firms collaborate for R&D. Some scholars argue that the organization’s competence improves over time ( e.g. Henderson, 1993; March, 1991), and thus its competence in R&D collaboration might also increase over time. However, aging can also hinder effective R&D collaboration due to organizational inertia that might arise (Barron, West & Hannan, 1994).

Furthermore, R&D intensity was included as a control variable, measured as the R&D expense in millions divided by the number of employees in order to take into account size effects of the variable (Zhang & Baden-Fuller, 2010). R&D intensity is included because multiple studies found that firms with high levels of R&D intensity are more capable to establish and exploit external linkages and create knowledge (Lin et al., 2012). Other studies use firm sales to compute a measure for R&D intensity, however firms may encounter fluctuations in sales in highly competitive industry resulting in a less reliable indicator. For this variable a lag-structure was adopted to alleviate endogeneity concerns (Belderbos et al., 2018), through measuring R&D intensity in year t-1.

(20)

19

Also, several industry dummies are incorporated into the model since the firms propensity to form alliances might differ between industries. All the sample firms are active in the industry with SIC code 283, corresponding to ‘’drugs’’. As a control variable this category will be broken down in sic codes 2833, 2834, 2835 and 2836 (see appendix 3), 2836 being the omitted dummy variable. Besides a dummy variable indicating whether a firm is a manufacturing firm or whether is solely focusing on service, wholesale or retail is added. Information about these dummy variables was gathered from Orbis.

Finally, economic conditions and market environments especially in fast changing industry change over time, which might influence their propensity to form R&D alliances and engage in external collaborations. Therefore, year dummies are introduced to control for this time effect. The omitted year was 2012.

3.4 Estimation Approach

In order to test the proposed conceptual model, this study uses Poisson regression with random effects. Poisson regression is chosen because the dependent variable in this study is the formation frequency of alliances, which is a count variable only taking discrete non-negative integer values (McCullagh and Nelder, 1989). Moreover, the dependent variable included a substantial amount of zero values, for firm’s that did not engage in any alliances in the particular years of our sample. When using Poisson regression it is ensured that zero values are incorporated into a model, instead of being implicitly truncated as they are in the OLS regression (Katila & Ahuja, 2002).

This study uses random effects instead of fixed effects because the independent variables for some firms include only zero’s in some years of the time frame. This is possible as some firms do not have any patenting activity, making these variables time invariant. This is problematic as the estimator of a fixed effects model does not allow for estimating time-invariant coefficients (Allison, 2009). When using fixed effects, these will automatically be removed from the analysis and result in far less observations and this study wants to take into account that some firms do not have any patenting activity. In addition, random effects allow for cross-firm comparison over time which is more suitable as this study investigates if the knowledge structuration, in terms of breadth and depth, explains why some firms engage more in the formation of R&D alliances compared to firms with a different knowledge structuration. In contrast, fixed effects use within-firm effects to explain how changes within the firm’s knowledge breadth and depth influence its R&D alliance formation frequency, and thus align less with the conceptual framework by not including cross-firm differences. However, using random effects also has its limitations which will be discussed in the limitations section.

(21)

20

equal, however in this data set the variance of the dependent variable (V=7.3169) is larger than the mean (M=0.8999). Therefore, the data will also be analysed using a negative binominal model in order to ensure robustness of the findings. The negative binominal regression relaxes the assumption of equal mean and variance while allowing for a direct measure of heterogeneity (Cameron and Trivedi, 1986).

4. RESULTS

4.1 Descriptive Statistics and Correlations

The results section of this paper starts with describing the descriptive statistics of the dependent variable and the explanatory variables in this unbalanced panel data set. The final data set includes 161 firms and 1038 firm year observations, on average there were 6.4 observations per firm. Table 3 includes the correlation matrix and the descriptive statistics, describing the mean, standard deviations, minimum and maximum. The table indicates that the typical firm in our sample accumulates around 0.9 alliances in a 3-year period, has an age of approximately 23,7 years and on average has patents in 5.8 different CPC classes (breadth) and approximately 3.3 patents in each subclass of the firm’s knowledge base (depth). The firm size variable has a mean and minimum that are negative, this was the result of the log transformation that was applied to the number of employees.

Table 3: Descriptive Statistics and Correlations

Mean Std. Dev. Min Max. 1 2 3 4 6 7 R&D Alliance Formation Frequency .900 2.705 0 35 1.000 Knowledge Breadth 5.804 9.431 0 65 0.605*** 1.000 Knowledge Depth 3.323 4.672 0 41 0.446*** 0.644*** 1.000 Firm Size (ln) -1.966 2.034 -6.908 4.758 0.517*** 0.715*** 0.557*** 1.000 R&D Intensity 282.173 339.744 0 4515 -0.093*** -0.108*** -0.028 -0.337*** 1.000 Firm Age 23.674 24.235 1 163 0.599*** 0.669*** 0.434*** 0.615**** -0.223*** 1.000 Non- R&D Alliances 0.418 1.178 0 15 0.702*** 0.578*** 0.373*** 0.450*** -0.107*** 0.533*** 1.000 Note: *p<0.1 **p<0.05 ***p<0.01

(22)

21

cut-off value of 4 (Pan & Jackson, 2008). In the current model, the mean VIF is 2.10 and the highest VIF value is 3.3. These VIF values indicate the multicollinearity among the variables is not an issue.

Table 4: Variance Inflation Factors

Variables VIF 1/VIF

Knowledge Breadth 3.30 0.3031

Firm Size (ln) 2.61 0.3831

Firm Age 2.09 0.4793

Knowledge depth 1.80 0.5560

Non R&D alliances 1.60 0.6268

R&D intensity 1.21 0.8244

Mean VIF 2.10

4.2 Hypotheses Testing: Poisson Regression Results

Table 5 and 6 display the results from the Poisson regression with random effects performed on the panel of US pharmaceutical/biotech firms. Table 5 present the results from model 1-4, aiming to show the main effects of knowledge breadth and depth on the frequency of R&D alliances. Table 6 present the results from models 5-8, including the direct effect of the moderator variable and the results of the tests for interaction effects.

Model 1, includes the linear terms of the control and dummy variables used in the analysis. The table indicates that firm age is positive and significant as a control variable (β= 0.022, p < 0.001) . Furthermore, the year dummies from the year 2005 until 2009 are significant (p<0.01). The significance of the year dummies indicates that the formation frequency R&D alliances is among others explained by year effects. The control variables R&D intensity (β= -0.0003, p>0.1) and accumulated non-R&D alliances (β=0.022, p>0.1) are not significant in explaining the formation frequency of R&D alliances. Furthermore, the dummy variable for manufacturing and the industry dummy variables for the SIC codes are also insignificant (p>0.1).

(23)

22

same and that both knowledge breadth (β =0.022, p=0.006) and knowledge depth (β =-0.100, p=0.001; β =-0.003, p=0.003) are significant in explaining R&D alliance frequency.

In order to test hypothesis 1, models 2 and 4 are analysed. Models 2 and 4 indicate that knowledge breadth is a significant factor in explaining the formation frequency of R&D alliances, with p-values( p < 0.01) and therefore the effect of knowledge breadth as hypothesized in hypothesis 1 is supported. For hypothesis 2, models 3 and 4 are analysed. The models indicate that the positive linear term and the negative squared term are significant with p-values (p< 0.01) and therefore the results indicate that hypothesis 2 might be supported. However, a significant positive linear effect and a significant negative quadratic effect alone is not enough to provide evidence for an inverted u-shape relationship. In determining whether the inverted u-shape relationship between knowledge depth and the formation frequency of R&D alliances is present the recommendations of Haans et al., (2016) are followed. Next to identifying the positive significant linear term and the negative significant quadratic term, I locate the turning point from which knowledge depth starts to have a negative effect on the R&D alliance formation frequency. I locate the turning point at −𝛽1/2𝛽2. The turning point is located at 17,031 with a 95% confidence interval of [16,747;17,315] (See appendix 4). Since this confidence interval is within the data range, as the minimum is 0 and the maximum is 41, the inverted u-shape relationship in model 4 is significant (Haans et al., 2016). The inverted u-shape effect of knowledge depth on the formation frequency of R&D alliances is plotted in figure 2. The plot confirms that the turning point is at 17,031 and that it is within our data range. Therefore, providing support for the hypothesized inverted u-shape relationship between knowledge depth and the formation frequency of R&D alliances. This implies that as the firm’s knowledge depth increases, they first tend to increase their frequency of R&D alliances, whereas this decreases with higher levels of knowledge depth. Since, both hypothesis 1 and 2 receive support, the baseline model with the independent variables presented in this paper is supported.

(24)

23

Table 5: Poisson Regression Results (Model 1-4) DV: R&D Alliance Formation

Frequency Model 1 Model 2 Model 3 Model 4

Independent variables

Knowledge Breadth 0.027(0.008) *** 0.022 (0.008) ***

Knowledge Depth 0.118 (0.031)*** 0.100 (0.031) ***

Knowledge Depth (squared) -0.004 (0.001) *** -0.003 (0.001) ***

Firm Size (ln)

Knowledge Breadth x Firm Size Knowledge Depth x Firm Size

Knowledge Depth (squared) X Firm Size Control variables

Firm Age 0.022 (0.005) *** 0.015 (0.005) *** 0.017 (0.005)*** 0.012 (0.005)** R&D Intensity -0.000 (0.000) -0.000 (0.000) -0.000 (0.000) -0.000 (0.000) Non R&D Alliances 0.022 (0.019) -0.019(0.022) 0.013 (0.019) -0.019 (0.022) Manufacturing Firm ( 0 or 1) 0.029 (0.342) -0.008 (0.330) 0.002 (0.323) -0.021 (0.319) Industry dummy’s:

SIC Code 2833 -0.210 (0.581) -0.124 (0.561) -0.092 (0.550) -0.053 (0.5419) SIC Code 2834 -0.395 (0.339) -0.355 (0.328) -0.2560 (0.322) -0.257 (0.319) SIC Code 2835 -0.536 (0.637) -0.461 (0.617) -0.473 (0.610) -0.426 (0.602)

Year Dummy’s: YES YES YES YES

Constant -1.591 (0.468)*** -1.597 (0.451)*** -1.988 (0.452)*** -1.915 (0.446)*** No. of Firms 1038 1038 1038 1038 No. of observations 161 161 161 161 Log likelihood -839.884 -833.745 -832.368 -828.560 Wald chi^2 236.28 246.45 254.90 260.07 Prob > chi^2 0.000 0.000 0.000 0.000 Note, *p<0.1 **p<0.05 ***p<0.01

Table 6: Poisson Regression Results (Model 5-8) DV: R&D Alliance Formation

Frequency Model 5 Model 6 Model 7 Model 8

Independent Variables

Knowledge Breadth 0.019 (0.008) ** 0.009 (0.016) 0.020 (0.009) ** 0.011 (0.019) Knowledge Depth 0.071 (0.032) ** 0.085 (0.038)** 0.079 (0.033)** 0.089 (0.039)** Knowledge Depth (squared) -0.002 (0.001) ** -0.003 (0.001)** -0.002 (0.001)** -0.002 (0.001)** Firm Size (ln) 0.239 (0.073) *** 0.228 (0.074)*** 0.209 (0.079)*** 0.208 (0.079)***

Knowledge Breadth x Firm Size 0.003 (0.004) 0.003 (0.005)

Knowledge Depth x Firm Size 0.014 (0.013) 0.011 (0.014)

Knowledge Depth (squared) X Firm Size -0.000 (0.000) -0.000 (0.000) Control variables

Firm Age 0.003 (0.005) 0.001 (0.006) 0.000 (0.006) 0.000 (0.006) R&D Intensity -0.000 (0.000) -0.000 (0.000) -0.000 (0.000) -0.000 (0.000) Non R&D Alliances -0.015 (0.022) -0.020 (0.023) -0.019 (0.023) -0.021 (0.023) Manufacturing Firm ( 0 or 1) -0.057 (0.301) -0.054 (0.300) -0.047 (0.300) -0.042 (0.300) Industry dummy’s:

SIC Code 2833 0.076 (0.515) 0.081 (0.513) 0.118 (0.512) 0.115 (0.511) SIC Code 2834 -0.218 (0.306) -0.213 (0.305) -0.213 (0.304) -0.212 (0.304) SIC Code 2835 -0.488 (0.585) -0.485 (0.584) -0.469 (0.583) -0.475 (0.583)

Year Dummy’s: YES YES YES YES

(25)

24

In model 5, the moderating variable firm size is included as an explanatory variable. The results indicate a positive significant effect of size on the formation frequency of R&D alliances (β = 0.2389, p<0.001). When adding this variable, the baseline model proposed in model 4 remains significant, with knowledge breadth (β = 0.019, p=0.016) and depth (β = 0.071, p=0.028; β = -0.002, p=0.023). As firm size is added as a control variable, firm age becomes an insignificant control variable (β = 0.003, p>0.1). In model 5 the turning point is located at 15.517 with a 95% confidence interval of [15,233;15,801] (appendix 4). This is within the data range and thus providing evidence for a significant inverted u-shape relationship between knowledge depth and R&D alliance formation frequency (Haans et al., 2016). Thus, the baseline model remains significant when adding firm size as another explanatory variable.

In models 6, 7 and 8, the interaction effects of firm size on the relationships of knowledge breadth and depth on R&D alliance formation frequency were added. Model 6 tests for the interaction of firm size on the relationship between knowledge breadth and R&D alliance formation frequency. The coefficient is positive and insignificant (β =0.003, p=0.471), indicating that there is no significant moderation effect in the model. In model 7, the interaction effect of firm size on the inverted-U shape relationship between knowledge depth and R&D alliance formation frequency is included. The linear term of the moderation has a positive coefficient which is insignificant (β=0.014, p=0.260). The squared term of the moderation has a negative coefficient and is not significant (β=-0.000, p=0.129). This indicates that there is no significant moderation effect in model 7. Model 8, includes both the interaction effects. When adding both moderation effects, the results remain insignificant. The moderation effect including knowledge breadth has a positive coefficient (β 0.003, p=0.597) and the moderation effects including knowledge depth were also insignificant for the linear term (β =0.011, p=0.409) and the squared term (β =-0.000, p=0.207).

In order to test whether hypothesis 3 receives support, model 6 and 8 should be analysed. Since in both models the coefficient has a p-value that is insignificant (p>0.1), no support is received for the moderation effect as hypothesized in hypothesis 3. Regarding the moderation effect of firm size on knowledge depth and its influence on the formation frequency of R&D alliances, model 7 and 8 indicate that in both models the coefficient has a p-value (p>0.1) that is insignificant, therefore not supporting hypothesis 4. As there is no significant moderation effect, I do not further investigate the steepening of the inverted-u shape effect.

4.3 Robustness Checks

(26)

25

Furthermore, the R&D intensity of the sample firms was measured in three ways, as including the absolute R&D expense, as the R&D expenses divided by the firm revenue, and finally and the one included in the main models as the R&D expense divided by the number of employees. The different operationalization for the firm’s R&D intensity did not change the significance of the results. Therefore, the operationalizations of firm size and R&D intensity can be considered robust.

For the dependent variable, R&D alliance formation frequency, two different measures were tested. The first one, was the number of alliances in year t and the second one was the cumulative number of alliances in year t, t-1 and t-2. Regarding the significance of the results, they were clearly significant when using the cumulative number of alliances, whereas knowledge breadth (β=.00145, p=0.273) became insignificant when using the number of alliances in year t in model 5. This study adopted the cumulative measure since it captures the firm’s long-term collaboration strategy instead of the short-term strategy.

Furthermore, the independent variables knowledge breath and knowledge depth were measured both as t, t-1 and t-2 as well as in t-1, t-2, t-3. There were some differences in the significance of the results, but in the models the independent variables were operationalized as the latter. This is because it takes time before knowledge is assimilated in the organization’s knowledge stock. New knowledge needs to spread throughout the organization and including in the organizational memory is time consuming. When it is assimilated in the organization’s knowledge stock it can be better utilized in the firm’s collaboration strategies.

Finally, the models were tested with both Poisson regression as well as Negative Binominal regression. The difference with a Poisson regression is that the negative binominal regression relaxes the assumption of equal mean and variance while allowing for a direct measure of heterogeneity (Cameron and Trivedi, 1986). When performing the Negative Binominal regression the significance of the results remained unchanged, therefore this study stuck to using the Poisson regression for the main analysis.

5. DISCUSSION

5.1 Key Findings

(27)

26

and how this process is moderated by firm size. By explaining how these internal characteristics influence external knowledge sourcing, clarification is provided about why some firms engage more in R&D alliances than others, and what causes this increased R&D alliance activity.

Using a panel dataset from firms active in the biotechnology and pharmaceutical sector, empirical evidence is found on the role of the firm’s internal knowledge base. First, this study finds that the firm’s knowledge breadth positively influences the firm’s R&D alliance formation frequency as hypothesized. Increasing the firm’s knowledge breadth influences the firm’s ability to create value through R&D alliances compared to firms with lower knowledge breadth. Value is created because the firm’s ability to scan, recognize and evaluate R&D collaboration opportunities is enhanced, absorptive capacity is increased, the firm knowledge breadth allows flexibility and adaptability in forming R&D alliances and the firm is not largely affected by risks of opportunistic behaviour. Second, regarding the role of knowledge depth this study hypothesized and found significance for an inverted u-shape relationship between knowledge depth and its formation frequency of R&D alliances. As with knowledge breadth, it increases the firm’s ability to scan, recognize and evaluate R&D collaboration opportunities and absorptive capacity. But in contrast with knowledge breadth, when increasing knowledge depth, the firm’s flexibility and adaptability to form R&D partnerships decreases because of core rigidities and internal resistance. Moreover, as a result of increasing knowledge depth firm’s find that their knowledge base is more at risk for opportunistic behaviour by partners. Hence, as knowledge depth increase after the turning point, firms begin to engage less in R&D alliances indicating an inverted u-shape relationship.

(28)

27

Regarding the relationship between knowledge depth and R&D alliance formation frequency it was argued that firm size moderates it in such a way that it leads to a steepening of the inverted u-shape. It leads to a steepening as smaller firms do not face internal resistance and core rigidities in external knowledge sourcing yet, while having the advantages of getting more resources. Larger firms do increasingly face core rigidities and internal resistance and because of this are also not able to utilize their increased resource availability efficiently. Also here, an alternative explanation is that firm size is associated with higher resource availability. This could be more important than the core rigidities and internal resistance in influencing the firm’s capability to create value through R&D alliances.

Thus, the moderation effects were not found as it is not clear whether the advantages of firm size outweigh the disadvantages or the other way around. Whereas, the moderation effect of firm size was not found, it is found that firm size has a direct positive effect on the firm’s formation frequency of R&D alliances, which could be explained by increasing resource availability.

5.2 Implications for Research

This study adds to the emerging stream of literature about the linkages between the firm’s internal knowledge base and its external knowledge sourcing activities (e.g. Lewin et al., 2011; Wuyts & Dutta, 2014), by providing a knowledge-based view perspective (Grant, 1996) on the role of internal knowledge base influencing the firm’s ability to create and leverage value through R&D alliances. In contrast to looking at internal R&D as a knowledge base proxy (e.g. Cassiman & Veugelers, 2006), which neglects the configuration of the firm’s knowledge and is an input variable instead of an output variable (Zhang & Baden-Fuller, 2010), this study takes into account the structuration of knowledge (George et al., 2008) by reconfiguring firm knowledge in terms of knowledge breadth and depth. This knowledge structuration clarifies how the distinctive knowledge elements influence external knowledge sourcing, specifically the formation of R&D alliances, separately. It enhances the understanding on the role of the firm’s knowledge breadth on R&D alliance formation frequency by providing support for a positive effect of knowledge breadth. Regarding knowledge depth, previous work has argued for a negative relationship between knowledge depth and R&D alliance formation frequency (Zhang & Baden-Fuller, 2010). This study enriches the understanding of the phenomenon by testing and confirming, following Haans et al. (2016), an inverted u-shape effect of knowledge depth. This implies that research should take into account that too much knowledge depth might have negative effects on external knowledge sourcing activities.

Reconfiguring the Firm’s Knowledge Base: The Effects of Knowledge Breadth and Knowledge Depth on the Formation Frequency of R&D Alliances.