Corporate Venture Capital and Innovation: The Role of Networks –
a quantitative empirical study
Richard Groenewegen S2890488 January 2020
Master thesis MSc Strategic Innovation Management University of Groningen
Supervisor: Dr. K. J. McCarthy Co-assessor: Dr. K. R. E. Huizingh
Abstract
Corporate venture capital (CVC) is often used by corporations to acquire novel knowledge and foster innovation. Vital for CVC entities is occupying the optimal position in a network in order to receive information and establish access to desired resources and thereby improving the innovation performance of the corporation. Network studies have shown ambiguous results in which the theorized benefits of being central are not reflected empirically. Through the efforts of the attention-based view, this quantitative empirical study explores this established phenomenon in a different manner by applying a new lens. With the attention-based view the potential benefits and downsides become considerably more in balance. By constructing a dataset of 1292 firm-year observations between 2002 - 2018 the hypotheses are tested. Results show that network centrality does impact innovation performance and that actors occupying a moderate network position enjoy better innovation performances than actors occupying a low or high network position.
Key words: Network centrality, attention-based view, innovation performance, corporate venture
1. Introduction
Corporations face an increasingly growing demand for innovation in order to emulate the rapid developments in the world (Ahuja and Lampert, 2001). In fact, the continuity of success for corporations often depends on innovation. To develop innovations and stay ahead, organizations need novel and valuable knowledge, resources and capabilities (Teece, 2007; Grant, 1996). However, corporations encounter difficulties in finding these novel and valuable assets (Gompers, 2002). As a consequence, corporations feel the urge to seek these vital assets externally because firms outside the boundaries of the corporation do possess the needed knowledge, resources and capabilities (Gompers, 2002). The ability of a corporation to develop and acquire valuable innovations, is often related to finding knowledge external to the firm and integrating it with internal knowledge (Teece, Pisano and Shuen, 1997; Henderson and Cockburn, 1994; Galunic and Rodan, 1998).
In order to conquer the difficulties of finding novel and valuable knowledge, resources and capabilities corporations started with corporate venture capital (CVC). CVC is defined by Dushnitsky and Shapira (2010: 991) as ‘‘the practice of minority equity investment by established firms in entrepreneurial ventures, that is, innovative companies that seek capital to continue operations. CVC investments opens a window onto new markets and novel technologies, thus offering established firms an opportunity to advance their innovation efforts.’’. The reason for this is that start-ups do possess novel knowledge that can enhance the business of a corporation but have often limited resources to finance their venture (Gompers, 2002). The CVC entity is a strategic corporate investment arm of a large corporation that identifies potential valuable start-ups and provides venture capital. This venture capital can be a source for entrepreneurial ventures to obtain the needed financial resources, while corporations use it as mechanism to obtain novel knowledge (Gompers, 2002). With CVC are corporations able to develop innovations more easily and endure the development pace in the world better (Gompers, 2002). Given the rapid pace of developments and in order to ensure the aforementioned continuity of success, the relevance of venture capital for corporations has never been this high as it is in recent times.
network of a CVC is important for receiving information and identifying potential opportunities (Bonacich, 1987; Rogers, 1995; Valente, 1995). The position in the network decides how fast and effective the information is received due to the degree of reach and access in the network. Nonetheless, little research is conducted about the network conditions in corporate venture capital and the performance. There is a gap in the literature of which, if answered, practitioners could benefit in daily practice. This study aims to fulfill this gap by researching the antecedents of the relationship between the network conditions of a CVC organization and the innovation performance of the corporation.
Furthermore, a large body of research has directed a great deal of attention to network aspects and the effect on performance in general (Nerkar & Paruchuri, 2005; Sorenson & Stuart 2001; Koka & Prescott, 2008; Gulati, Nohria & Zaheer, 2000; Kilduff & Tsai, 2003).Many studies who researched this relationship use network theory as their focal theory. According to network theory, centrality implies better access to information and resources in the network while simultaneously having more control. Nevertheless, academics have found results that have been ambiguous, which results in confusion among scholars (Mariotti & Delbridge, 2012; Reinholt, Pedersen, & Foss, 2011; Lechner, Frankenberger and Floyd, 2010; Koka et al., 2008; Gnyawali & Madhavan, 2001).
Correspondingly negative (Mariotti et al., 2012; Lechner et al., 2010) and positive (Koka et al., 2008; Gnyawali et al., 2001) effects of high centrality are found. In addition, Reinholt et al., (2011) found that a central position is beneficial, but far from enough and proposes a moderate influence of centrality. In addition to scholars, the mixed results have led to confusion in managerial practice as well and shows that scholars lack the understanding of why the theorized results are not reflected in practice. Therefore, there is a call for more investigation to solve this puzzle (Gibbons, 2004). This paper answers this call by applying a new lens on the network conditions of the investment arm under which corporations have better innovation performance.
in entrenched phenomena (Bouquet and Birkinshaw, 2008; Chen, Bu, Wu, and Liang, 2015; Zhao, Chen, and Xiong, 2016).
The attention-based view (ABV) theory as is known today is developed by Ocasio (1997) and built his idea of how firms distribute and regulate attention of decision-makers on three core principles. Namely, focus of attention, situated attention and structural distribution of attention. These forms of attention are closest to the attentional engagement, which is considering attention as a process concerning the application of time, energy and effort (Ocasio, 2011). ABV studies firms’ decision-maker’s attention and the consequences of this attention on firm behavior.
I propose that due to the third principle of the ABV a positive effect is present on the innovation performance by an increase in centrality, while the first principle has a negative effect on the innovation performance by an increase in centrality. The ABV suggests that the principles are equally important in allocating attention (Ocasio, 1997). Based on this, one can hypothesize an inverted U-shape relationship between network centrality of the CVC entity and the innovation performance of the corporation, according to Haans, Pieters and He (2016). Therefore, we argue that CVC entities should occupy a moderate central position in its network in order to maximize the innovation performance of the corporation.
This relationship will be moderated by geographic distance and relatedness in such a way that it flattens the U-shape relationship by an increase in geographical distance and relatedness. A CVC entity’ aim is to find what novel knowledge is available outside the corporation and where this novel knowledge located is, that can enhance the corporate business (Maula, 2007; Dushnitsky, 2011). Following this finding, two logical important contextual factors CVC entities face are relatedness and geographical location. This paper will provide an answer to the main research question: ‘What is the influence of the CVC’ network on the innovation performance of
the corporation?’
I show that network centrality is related with innovation performance in an inverted U-shape relationship. Therefore, CVC entities should aim for a moderate central position in its network. Relatedness is found to decrease the innovation performance, and hence should CVC entities search beyond the scope of related ventures when they seek to stimulate breakthrough inventions. Unlike relatedness, there is no support for geographical distance as a moderator.
This paper contributes to the already existing body of literature in two ways. First, practitioners are given insights in the relationship between the network of a corporate venture capital investment arm and the innovation performance of the corporation, which will help them in decision making. Furthermore, this paper provides new insights for research on network aspects in general and for CVC entities and their performance in particular.
2. Background
2.1 Corporate venture capital
Corporations have been attracted to venture capital investments for decades and started in the mid-l960s (Gompers and Lerner, 2000). These organizations saw often that start-ups accomplished great successes with opportunities they identified first (Teece, 1986; Gompers, 2002). However, due to insufficient protection and organizational myopia they were inadequate to exploit and/or reap the benefits from these opportunities (Teece, 1986; Gompers et al., 2000). In order to overcome these start-ups, corporations started with corporate venture capital. With the increasingly competition in research and development and the considerable reduction in product life cycles, demand for an increase in the speed of innovation has never been this high (Fulghieri and Sevilir, 2009). Corporations use CVC as a tool to answer this demand, efforts have been motivated primarily for strategic reasons by a desire to gain access to innovations (Gompers, 2002). The corporation influences the investment decisions the CVC takes and ensures that the strategy is aligned with the corporation (Gompers et al., 2000; Dushnitsky, 2012).
start-up through three different manners, namely by providing financial resources (Gompers, 2002), leveraging corporate resources (Maula and Murray, 2002; Dushnitsky and Lenox, 2005b) and advocating the start-up to new potential partners (Stuart, Hoang and Hybels, 1999).
2.2 Corporate venture capital and innovation performance
As stated in the previous paragraph, CVC is primarily used by corporations for strategic reasons in order to enhance and increase innovation, which is contrary of financial motivations like independent venture capital organizations. Exposure to novel, pioneering technologies has the effect that the likelihood of creating innovations for established firms would increase (Ahuja et al., 2001). A large body of literature has conducted research to find the relationship between CVC investments and innovation performance. Results show a positive linear relationship (Fulghieri et al., 2009; Dushnitsky and Lenox, 2005a) or an inverted U-shape relationship (Lee, Kim and Jang, 2015; Wadhwa and Kotha, 2006) between CVC investment and innovation performance.
Apparent is the positive effect on innovation performance since there are no studies that show significant negative results. According to these studies, CVC fosters corporate innovation. Important to mention here, is that most research dives into the contextual factors of CVC to examine the effects on innovation performance. For example, Wadhwa, Phelps and Kotha (2016) investigated the diversity of CVC investments and found that corporate investors maximize their innovation performance when they invest in moderately diverse portfolios of start-ups. Dushnitsky et al., (2005b) found that corporate venture capital programs may be instrumental in harvesting innovations from entrepreneurial ventures, especially in weak intellectual property regimes.
2.3 Network centrality
Network centrality is one of the most frequent used measures in network studies, it refers to how well connected an actor is to the parts of the network with the greatest connectivity (Bonacich, 1987) and the extent to which an actor occupies the central position of a network (Freeman, 1979). Centrality is an attribute of social networks that is imperative (Freeman, 1979). Actors occupying such position are responsible for maintaining and establishing communication through the network (Cote, 2019). Actors can refer to individuals, groups, departments and organizations and they have strong or weak ties. The strength of the ties between the actors determines the value of the relationship (Cote, 2019). Centrality has both positive and negative impact on the innovation performance based on several studies which will be outlined in the two paragraphs below.
2.4 Positive effects of high network centrality on innovation performance
Whereas the effect of CVC networks on innovation performance have had limited attention in the literature, there are studies with a focus on the effect of network centrality on innovation performance in general. Scholars have used various theories besides the obvious social network theory to research this relationship. For example, the knowledge-based theory found that firms closer to the center of the network have better innovation performance (Brennecke and Rank, 2017; Guan and Liu, 2016). The social capital theory found that individuals and organizations that are better-connected tend to perform better (Tan, Zhang and Wang, 2015). Another used theory is the transaction cost theory that suggest that a central position in the network results in a reduction of transaction costs, which will lead to a positive effect on the innovation performance (Williamson, 1985; Maskell, 2000). Relationships can be seen as a resource to obtain access to the needed assets (Pfeffer and Salancik, 1978). Therefore, the position in a network can be seen as a resource to obtain assets that can increase the innovation performance by the resource dependency theory.
increase compared with less central actors due to the number of connected partners (Freeman, 1979). Accordingly, given this information asymmetry, centrally positioned actors could more rapidly exploit potential resources and seize opportunities that pass by (Gulati et al., 2000).
Not alone from an information perspective is it beneficial to posit a central position, according to Soh (2010) a centralized organization has greater proximity to all other organizations. With this proximity the organization is able to exert greater influence on its network (Koka and Prescott, 2002; Sorenson et al., 2001).
The third aspect is that centrally positioned organizations are seen as more credible and external actors are willing to exchange additional resources with the central organization (Podolny, 1993; Podolny, 2001; Reinholt et al., 2011). Therefore, the central actor has greater access to critical resources and valuable potential partners (Koka et al., 2008; Knoke and Burt, 1983).
Fourth, information and knowledge spillovers are more easily achieved when an organization is more central in its network (Kilduff and Brass, 2010). It is known that information exchange and knowledge diversity fosters innovation (Hargadon, 2002; Schoenmakers and Duysters, 2010). In addition, cohesive, connected, concise, valuable and knowledge rich networks can positively influence the innovation performance of a firm (Muller and Peres, 2019; Wang, Chen and Fang 2018; Kim, 2019).
The final aspect is the quality level of the organization, resources are seen as a signal of quality of the organization that possess a central position in the network (Nerkar et al., 2005; Sorenson et al., 2001). This quality signal results in additional power over other network actors while these actors rely on the organization (Brass, 1992; Stevenson and Greenberg, 2000). Thus, the influence of the network is a very important factor in this relationship that needs to be well understood.
2.5 Negative effects of high network centrality on innovation performance
is that the knowledge received by the organization will decrease in timeliness and effectiveness (Mariotti et al., 2012).
Another negative effect is that over-search occurs when the organization has a high centrality (Koput, 1997). Over-search can result from multiple sources, like absorptive capacity problem, timing problem and attention allocation problem. Based on the absorptive capacity theory it might occur that the marginal returns might not be sufficient when the organization maintains high centrality (Koput, 1997; Cohen and Levinthal, 1990; Gilsing, Nooteboom, Vanhaverbeke, Duysters and Van den Oord, 2008).
However, these arguments are seen as less prevail by scholars in the majority of the used theories and cannot overcome the benefits. Even when empirically there is no unequivocal evidence that high centrality is beneficial, is the tendency in literature still that high network centrality has a positive effect on innovation performance. Nevertheless, scholars have observed that the results have been largely inconclusive (Mariotti et al., 2012; Reinholt et al., 2011; Lechner et al., 2010; Koka et al., 2008). The mixed results have led to confusion among scholars and managerial practice as well. Therefore, this paper answers the call (Gibbons, 2004) for more investigation on this phenomenon while simultaneously filling a gap in research between network centrality and the innovation performance by conducting this study in a CVC setting.
As stated in the two paragraphs above, various theories have been used to research this relationship. Nevertheless, there are still ambiguous results that have led to confusion. For CVCs it is critical to identify valuable opportunities to foster innovation (Maula, 2007; Dushnitsky, 2011). By doing this, managers need to allocate their attention to the right places. Since the allocation of attention is vital, I introduce that managers attention is a key aspect within a CVC organization in order to identify promising opportunities. Therefore, in this paper I will use the attention-based view (ABV) theory instead of more common used theories in order to provide novel insights in this established phenomenon. This is the first time that the ABV will be applied on this relationship. By applying the ABV lens it might be possible to clarify, on a theoretically solid manner, why the empirical findings differentiate from the theorized findings in past research.
2.6 Attention-based view
introduction in explaining firms’ strategic decision making and adaption (Ocasio, 1997) is it nowadays visible in an increasingly number of research. For example, Bouquet et al., (2008) researched the relationship between the headquarter and subsidiaries. Furthermore, the ABV is used in the field of social issues in management (Zhao et al., 2016) and stakeholders (Bundy, Shropshire and Buchholtz, 2013). While it is present in organizational change (Eggers & Kaplan, 2009), management innovation (Chen et al., 2015) and internationalization strategies (Levy, 2005) as well. As showed, the ABV provides a new lens for scholars to research entrenched phenomena which brings new understanding among these phenomena. ABV studies firms’ decision-maker’s attention and the consequences of this attention on firm behavior (Ocasio, 1997).
The ABV finds its origin in the early work of Simon (1947) when he introduced a new perspective, which calls for attention to explain how firms make decisions based on the limits of human rationality of the labor force. Actually, decades later it is Ocasio (1997) who develops the attention-based view as is known today. The ABV describes firms as systems of structurally distributed attention in which the cognition and action of individuals are not predictable from the knowledge that the individual possesses, but are derived from the particular context-depending situations that an individual in the organization faces (Ocasio, 1997). Here, I follow the definition of Ocasio (1997: 189) of attention: ‘‘Attention is here defined to encompass the noticing, encoding, interpreting, and focusing of time and effort by organizational decision-makers on both (a) issues; the available repertoire of categories for making sense of the environment: problems, opportunities, and threats; and (b) answers: the available repertoire of action alternatives: proposals, routines, projects, programs, and procedures.’’.
The second principle is situated attention, which illustrates that attention is context-dependent (Ocasio, 1997). Depending on the context or situation decision-makers face, they focus and react on particular issues and answers. The characteristics of the situation has a direct influence on individuals’ behavior (Ross and Nisbett, 1991). The premise of this principle is that behavior of decision-makers is depending more on the characteristics of the situation than the characteristics of the individual (Ocasio, 1997). In addition, this principle operates at the level of social cognition (Fiske and Taylor, 1991) which can be described as how organizations and its environment shape the situation of the individual and how the individual thinks and decides.
The last and third principle is the structural distribution of attention. Here, the focus of attention depends on ‘‘how the firm distributes and controls the allocation of issues, answers and decision-makers within specific firm activities, communications and procedures.’’ (Ocasio, 1997: 191). The latter firm specific characteristics can also be referred to as attention structures and are defined as “the social, economic and cultural structures that govern the allocation of time, effort and attentional focus of organizational decision-makers in their decision-making activities.” (Ocasio, 1997: 195).
According to Ocasio (1997), these attention structures have an active role in the valuation and legitimization of issues and answers and the development of procedures and communication paths, as well as in shaping decision-makers’ interests and identities that guide them. In the attention structures there are four factors that explain the allocation of attention. It is important to mention that they are interconnected with each other. These factors are the rules of the game, players, structural positions and resources and can each influence the focus of attention. These factors will be discussed in more detail in the next paragraph.
2.7 An ABV lens on the network centrality of the CVC and innovation performance
The ABV lens has not been applied in current literature by scholars to research the relationship between network centrality and innovation performance, in particular the CVC’ network and innovation performance of the corporation. As previously discussed, the ABV has three main principles. These principles will be applied in order to research the relationship between the network centrality of the CVC and the innovation performance of the corporation.
processes focus the energy, effort, and mindfulness of organizational decision-makers on a limited set of elements that enter into consciousness at any given time (Ocasio, 1997). In addition, it illustrates that people focus on ideas that have legitimacy, value and relevance and are drawn away from other ideas. In this case decision-makers focus their attention on aspects that have legitimacy, value and relevance for the CVC and can only focus on a particular number of ideas. Usually, the strategy of the CVC is well aligned with the strategy of the corporation, since the CVC’ goal is to support the corporation by acquiring valuable novel knowledge (Gompers, 2002). Organizations that are central in a network may receive incorrect information from other actors in the network (Lechner et al., 2010). Based on this information, the organization chooses to follow a certain path. Given the cognitive capacity and their ability to focus on a limited number of ideas only, they miss out on other valuable information in the network. Because of the diminished accuracy and reliability potential promising opportunities can be missed.
The second aspect based on the first principle is that organizations with high centrality are incapable to process all incoming information flows properly because their number of ties has increased enormously (Mariotti et al., 2012). As a result, timeliness and effectiveness will decrease for the focal organization. Finally, a central organization has problems with over-search due to problems with the allocation of the attention (Koput, 1997). The resources of the firm become ‘overheated’, as a consequence the firm has lower performance and misses out on promising information and investment opportunities. These problems stated above arise when an organization is in the situation of high centrality and are less respected and given limited value in other theories. However, they become much more essential when an ABV lens is applied. These deficiencies result from the first core premise of the ABV, while other theories have downplayed its value.
and communication channels constructed and decide decision-makers where to allocate their attention. Following this reasoning, two important contextual factors CVC entities face are relatedness and geographical location. First, relatedness of the knowledge, does the knowledge add value to the current knowledge base and is there a fit with the corporate strategy. Second geographical location, where can I find this knowledge and can it easily be transferred to the corporation, which is seen as the distance between actors as well.
Within the CVC decision environment these two contextual characteristics, relatedness and geographical distance, are important factors. Scholars found that geographical distance is an important aspect in the venture capital industry (Maula, 2007; Dushnitsky, 2011). Geographical distance is the physical distance between the investment entity and the entrepreneurial venture. According to Cumming and Dai (2010) this industry has a forceful local bias in their investment decisions. In addition, organizations have a difficult time monitoring investment opportunity when it is outside their locality (Chen, Gompers, Kovner and Lerner, 2010). Especially when the competition is fierce, because in a study of Lindgaard Christensen (2007) is found that when the competition increases, organizations tend to confine themselves to investments within a closer geographical distance. When applying the ABV lens on geographical distance it is apparent that by an increase in distance, there is less attention for organizations located further away due to the fact that decision-makers in organizations react to situations that are shaped by the organization and its environment. Therefore, its attention is focused on start-ups nearby since they are demanding attention. Because the cognitive capacity of the organization is limited, it is hard to allocate attention to opportunities and the corporation located further away because the capacity limit is reached. Neglecting opportunities nearby is often not an option since start-ups demand attention regardless, due to their proximity. Given their locally distributed attention it is harder to align well with their parent organization located further away and be aware of the needs that the corporation seeks.
the entrepreneurial venture and the corporation are too closely related, the situation might occur that the venture possess little novel knowledge for the corporation to learn from (Sapienza, Parhankangas and Autio, 2004). This might have the effect that fewer innovations will be introduced by the corporation (Ahuja and Katila, 2001). Research that incorporated the CVC setting found that CVC investments in related industries results in greater innovation performance, thus the acquirer and the target should be in related industries (Keil, Maula, Schildt, Zahra, 2008).
When applying the ABV lens on relatedness, the procedural and communication channels are essential. This implies that decision-makers attention is situated in the procedural and communication channels of the firm (Ocasio, 1997). In other words, the availability and saliency to issues and answers depends on the context and characteristics of the procedural and communication channels of the firm. Based on these ‘routines’ within the corporation and between the investment arm and corporation, they are directed to particular issues and answers. These issues and answers are opportunities that fit well with the strategy of the corporation. Therefore, the procedural and communication channels are composed in order to ensure that the investment arm focuses on opportunities in the best interest of the corporation. As a result, it will miss out on opportunities that are beyond these ‘routines’ and therefore miss opportunities that are unrelated in terms of the industry of the corporation that could facilitate breakthrough inventions.
The third and last principle is structural distribution, which is based on how the organization is formed (Ocasio, 1997). This principle distinguishes four important aspects, namely rules of the game, resources, the players and social positions. Together, these four constitute the attention structures of the organization and generate a distributed focus of attention among decision-makers participating in the firm’s procedural and communication channels (Ocasio, 1997). Attention structures regulate the valuation and legitimization of issues and answers (Ocasio, 1997). In addition, they influence both individuals and organizations in the network of the organization.
succeed (Ocasio, 1997). The rules of the game structure and regulate the motives of actors within the focal firm by providing logic of action together with cultural and material values as well.
The second attention structure is resources, which are tangible and intangible assets that enables a firm to perform its activities by producing its goods and services (Wernerfelt, 1984). These resources are embedded in the routines and capabilities of the organization and allows the firm to perform a wide variety of tasks (Nelson and Winter, 1982). Resources can consist of human, physical, technological and financial capital that are available to the firm in order to reach the objectives (Ocasio, 1997).
The third attention structure is players and is according to Ocasio (1997) a vital aspect of the attention regulation. Players are individuals or groups that affect the firm’s attention regulation by bringing specific skills, beliefs and values to the firm (March and Olsen, 1976). They can influence others to certain behavior and therefore regulate the decision and activities of other decision-makers (Ocasio, 1997).
Lastly, the attention structure social position which is called the structural position as well. This is defined as: ‘‘the roles and social identifications that specify (a) the functions and orientations of decision-makers, and (b) their interrelationships with other structural positions internal and external to the firm.’’ (Ocasio, 1997: 197). There is an interaction between the social position and the rules of the game to provide decision-makers with the interests, values and identities that regulate their thinking and doing in organizations (Ocasio, 1997). It is the position in the organization in which an actor has power over others and can influence them in their behavior.
3. Hypothesis development
3.1 CVC’ network centrality and innovation performance
the principles are equally important for the allocation of attention within a firm, according to Ocasio (1997).
Based on the third principle of Ocasio (1997), a positive linear function is presumed in the independent variable. The attention structures of an organization illustrate the positive effect the characteristic of high centrality has for an organization. When the organization occupies a central position in its network, it has through their players and social positions informal influence on partners and other parties in the network to steer decision making (Lechner et al., 2010). The more central the firm becomes, the greater the influence of the players will be on other organizations. In addition, the central position of the organization results in greater and direct access to resources within the network (Ibarra,1993). These resources enhance the answers to issues the organization faces in their daily business. Because the answers to issues is shaped and partially determined by existing organizational resources (Ocasio, 1997). Contingent upon the access of resources is a firm able to have superior answers. Thus, by having high centrality the firm is likely to have better answers to certain issues due to greater access of resources.
The rules of the game and social position are meaningful by high centrality because an actor that occupies a strategic position characterizes a specific ego’s power relative to alternative network alters (Gnyawali et al., 2001; Kenis and Knoke, 2002; Wasserman and Faust, 1994). Therefore, it is able to transfer its superior structures to actors that are less central and the rules of the game of a central firm are seen as more paramount. Lastly, the central positioned player in the network is seen as the most important player in the network (Kilduff et al., 2003). As a result, it enjoys the benefit of receiving information sooner while having a particular status as well. The importance created by the central position of the firm has the result that the firm becomes the overlying organization in the network.
The second aspect, based on the first principle, is that organizations with high centrality are incapable to process all information properly because their number of ties increased dramatically (Mariotti et al., 2012). As a consequence, the timeliness and effectiveness of information processing decreases. The last aspect is that a central organization has problems with over-search, due to problems with allocation of the attention (Koput, 1997). The resources of the firm become ‘overheated’, which results in lower performance by missing out on promising information and opportunities. Based on the first and third principle of Ocasio (1997) there are two functions presumed in the independent variable, one positive and one negative. According to the paper of Haans et al: (2016: 1180) ‘‘One may construct an inverted U-curve by interacting two latent linear functions, one positive and one negative in the independent variable.’’.
To summarize, this paper hypothesizes that CVC network centrality has a positive - due to the third principle of Ocasio (1997) - and negative - due to the first principle of Ocasio (1997) - function which results, according to Haans et al., (2016) in an inverted U-shape relationship between CVC’ network centrality and the innovation performance of the corporation. The following hypothesis can be formulated.
H1: CVC’ network centrality has an inverted U-shape relationship with the innovation performance of the corporation.
3.2 Moderating effect 1
The first moderating effect in this paper is geographical distance between the corporation and the CVC entity. Organizations that are proximate to each other potentially benefit due to less information asymmetry and close monitoring (Cumming et al., 2010; Boschma, 2005). Therefore, the CVC entity is likely better aware of what the corporation might need instead of CVC entities located further away. With an attention perspective, attention is more easily given and maintained when there is little distance. The reason for this is that less cognitive capacity is needed than by greater differences in distance.
has only a limited amount attention to provide, it has difficulties in allocating attention to actors further away. This does not only account for entrepreneurial ventures, but for the corporation as well. When the geographical distance increases between the investment arm and the corporation, contact and alignment decreases. The amount of attention a CVC assigns to the corporation and therefore to the needs is less for entities located further away than adjacent, because it is drawn away by network actors nearby.
As a result, geographical distance affects the positive and negative latent mechanisms of the inverted U-shape. The positive latent mechanism - in which a higher level of centrality results in more access and control and thereby increasing the innovation performance - is affected by geographical distance in such a way that an increase in distance is likely to result in lower benefits. The reason for this is that the investment arm is situated further away from the corporation and is less able to select promising opportunities and align with the corporation than CVC entities nearby. The negative latent mechanism - in which higher levels of centrality will subsequently lead in reaching the boundaries of the human capacity and therefore unable to provide attention to all - is affected by geographical distance in such a way that investment arms located further away have more difficulties in maintaining attention on the corporation than entities located nearby. This is due to the fact that the entity’ attention is more often drawn away locally and is reaching their attention limits.
Therefore, this paper hypothesizes that geographical distance negatively moderates - due to the inability to overcome boundaries and better valuation of issues and answers with greater legitimacy and value for the corporation - the inverted U-shape relationship between CVC’ network centrality and the innovation performance of the corporation in a way that an increase in geographical distance flattens the inverted U-shaped relationship.
H2: Geographical distance moderates the inverted U-shape relationship between CVC’ network centrality and the innovation performance of the corporation in such a way that an increase in geographical distance flattens the inverted U-shaped relationship.
3.3 Moderating effect 2
Ocasio (1997) illustrates that decision-makers their attention is situated in the procedural and communication channels of the organization. In other words, the availability and saliency to issues and answers the organization provides attention to, depends on the context and characteristics of the procedural and communication channels of the organization (Ocasio,1997). Based on these ‘routines’ within the corporation and between the investment arm and corporation, is the CVC directed to particular issues and answers.
Normally, these issues and answers align employees with the strategy of the corporation well. However, in order to ensure breakthrough inventions for the corporation the CVC entity should think and act outside the box as well, instead of following opportunities of the current corporate strategy alone. Due to the procedural and communication channels the attention of the CVC is focused on opportunities that fit well with the business strategy of the corporation, while its attention is drawn away from opportunities outside the procedural and communication channels of the corporation. As a result, it will neglect opportunities that are beyond these ‘routines’ and therefore focuses on opportunities that are related in terms of industry of the corporation because the CVC is unable to provide attention to all opportunities. The consequence is that the CVC is unable to provide novel knowledge to create breakthrough inventions due to their related focus.
As a result, relatedness affects the positive and negative latent mechanisms of the inverted U-shape. The positive latent mechanism - in which a higher level of centrality results in more access and control and thereby increasing the innovation performance - is affected by relatedness in such a way that a higher level of relatedness is likely to result in a decrease of the benefits. The reason for this is that the knowledge bases will be too similar, which will lower the chance on breakthrough inventions. The negative latent mechanism - in which a higher level of centrality results in reaching the boundaries of the human capacity and therefore unable to provide attention to all - is affected by relatedness in such a way that it enhances the negative latent mechanism due to bounded procedural and communication channels that are connected to the corporate strategy. Therefore, the CVC entity is assigning their attention to incremental opportunities instead of breakthrough opportunities.
H3: Relatedness moderates the inverted U-shape relationship between CVC’ network centrality and the innovation performance of the corporation in such a way that an increase in relatedness flattens the inverted U-shaped relationship.
Following the hypotheses stated above the conceptual model shown in figure 1 can be made.
Figure 1: Conceptual model
4. Methodology 4.1 Empirical setting
4.2 Sample
This study constructed a data set from three archival sources at the firm level. Data for the independent variable is obtained from Thomson Reuters Eikon database. This database contains various data aspects of firms and countries. For the dependent variable data from the European Patent Office (EPO), extracted from the PATSTAT database, is used to collect patent information of the firms in the sample. In existing literature, it is more common to use the USPTO database instead of data from the EPO, however using the EPO has three advantages. Namely, the filing fees are higher at the EPO than the USPTO. This results in a decrease in the number of preemptive patenting and increases the accuracy of following the inventiveness of the firms in the sample. Second, the EPO makes a distinction between granted and non-granted patent applications, whereas other patent offices are often neglecting this. This distinction provides a more comprehensive view of the actual inventive activities of the firm. Finally, the EPO provides detailed information about the applicants’ credentials which will help in matching firms to patent applicants. In addition, the EPO includes information based on the harmonization procedure of the K.U. Leuven (Magerman, Van Looy and Song, 2006) which improves the matching procedure.
The control variables were obtained from the Standard and Poor's Compustat database. This database contains financial data and is one of the most used databases for financial information among scholars. This database has high-quality data and has long historical data. The final data set consists of pharmaceutical industry firms dated between 2002 - 2018. After matching and constructing the panel data set 93 pharmaceutical firms with 1292 firm-year observations remain, which is used for the analysis.
4.3 Independent variable
Hirotaka (2012). The data for the independent variable is retrieved from the Thomson Reuters Eikon database.
Degree centrality is the simplest measures and considers the number of ties of the focal firm, compared to the size of the entire network (Freeman, 1979). This measure acknowledges the number of ties a firm has in a network. Therefore, it is a good measure to observe how networked the firm is. However, this measure does not account for inequalities in the ties of the firm. In addition, the aggregate of the weights of the ties connected to the actor is taken into account if the network is weighted (Barrat, Barthelemy, Pastor-Satorras and Vespignani, 2004). For an undirected network, degree centrality for an actor i is defined as:
In this formula the leading divisor is adjusted for excluding the j = i term. Directed networks may require actors having a different number of incoming and outgoing ties, and therefore there is an out-degree and in-degree centrality. Out-degree centrality for actor i is defined equivalently to the above formula, while for the in-degree centrality the formula is simply adjusted by the adjacency matrix:
This centrality measure reflects how actors with smaller average shortest path lengths receive higher centrality scores than those that are situated farther away from members of their network. One problem that arises from this method is the problem of infinite distances to unreachable actors. A solution for this problem is considering only actors that are reachable. Nevertheless, important to mention is that ties tend to be less and shorter of actors who are less central. This possibly results in actors receiving higher closeness centrality scores than actors more central in the network, while this is against the premise that actors in this situation are less central in the network (Newman, 2010).
Betweenness centrality is a measure that provides larger centrality scores to actors that lie on a larger proportion of the shortest path, that connects pairs of other actors. It can be described as the position of parties in a certain network that are able to intercept or influence the communication within that network in which communication follows the shortest paths (Freeman, 1979). In this formula Pij stand for the number of shortest paths from actor i to j. Pij(k) stand for the number of shortest paths from actor i to j including actor k. Following Anthonisse (1971) and Freeman (1977), the measure for betweenness centrality for vertex k can be defined as:
In order to normalize the maximum number of paths a given actor could have, the formula should be divided by (|V | − 1)(|V | − 2). Betweenness centrality reflects diverse knowledge flows well, since the focal firm is the knowledge broker for different parties with different knowledge bases, due to the characteristic of the shortest path. This measure does not differentiate in significance of network actors as it assumes that all partners are equal.
(1972) was the first one to introduce this measure and defines the centrality of actor i, xi, as the aggerate of the centrality of its neighbors which is then multiplied by a constant, which is A. The formula for eigenvector centrality is:
This formula can be rewritten as:
Vector x in the rewritten formula is an eigenvector of adjacency matrix A, while λ is its corresponding eigenvalue. Normally, the eigenvector is used matching the dominant eigenvalue of A. In case the network is directed, the overall problem is obtaining a centrality measure based on how often an actor is being pointed to, while the importance of neighbors associated with the incoming ties is taken into account as well. Therefore, with a minor adjustment to the rewritten formula, eigenvector centrality is redefined as a vector x where A′ is the transposed adjacency matrix:
It is important to mention that with the eigenvector centrality an actor that has no incoming ties will always have a centrality of zero. In fact, even when neighbors do not have any incoming ties the centrality will be zero because the sum will not contain any nonzero terms.
4.4 Dependent variable
used. Using a simple patent count is insufficient to differentiate in importance and the impact of a patent and therefore the quality of the innovation performance of the corporation. Forward citations per patent is seen as a measure of technological impact of inventions (Yan et al., 2019). In addition, research indicates that the number of citations a patent receives is associated with its importance and estimated economic value (Harhoff, Scherer and Vopel, 2002; Tseng, Hsieh, Peng and Chu, 2011). Usually, a certain time period is used to calculate the forward citations of a patent. However, within the SQL coding that is used to retrieve data in PATSTAT it is not possible to give restrictions in time periods for citations. Therefore, a relative measure is used to calculate the difference in quality of a patent. The relative measure is calculated using the fixed-effects approach by the OECD and can be found in the patent statistics manual (OECD, 2009).
This approach involves scaling citation counts by dividing them with the average citation count for a group of patents to which the patent of interest belongs, in this case the year. The benefit of using this approach is that all variations over time in citation intensities are being eliminated (OECD, 2009). Therefore, it enables researchers to compare citation intensity of different cohorts. Another benefit of this measure is that comparisons over the years can be fairly made while recent years can be taken in account as well, which enables this study to incorporate recent years. This would not be possible when a year lag of normally five years would have been used to indicate the number of forward citations. Then, forward citations of patents in the 97th percentile or greater are used to measure breakthrough inventions. Each firm is assigned with a value that reflects the total number of breakthrough inventions in that year. When the firm has no breakthrough inventions in a specific year, the value zero will be assigned.
4.5 Moderating variables
4.6 Control variables
Furthermore, a set of four relevant control variables plus dummy variables are included in this study to eliminate alternative explanations for innovation performance. In addition, these variables will subsequently increase the validity of the results. The control variables are retrieved from Compustat database, which is a database where information is stored about financial, statistical and market aspects on active and inactive companies.
The first control variable is R&D intensity, it is important to control for this because firms that have higher R&D intensity are more likely to patent frequently (Wang, Choi, Wan and Dong, 2016). In addition, Cohen et al. (1990) found that firms with higher R&D intensity are able to combine novel acquired knowledge to the existing knowledge base better and enjoy higher inventive output than firms with lower R&D intensity. R&D intensity is calculated by the R&D expenditures divided by the total sales. Second, firms financial leverage is included because the financial leverage of a firm mirrors the risk tolerance of the innovation activity of a firm (Wang, et al., 2016). The firm’s financial leverage is measured by long-term debt divided by total assets. The third control variable is firm’s prior performance. The prior performance of a firm can generate the search for innovation within the firm (Dong, 2016). This is calculated by operating income before depreciation divided by total assets. The final control variable in this study is firm’s
size, which is calculated by the natural logarithm of the total sales. According to Dong et al., (2016)
the size of the firm may influence the ability to assimilate external knowledge and use it for innovation. In addition to these control variables are year, country and firm dummies included in the analysis.
4.7 Method of analysis
assumption is that the mean and variance of the data are equally distributed. As can be seen in the descriptive statistics of the data it shows significant evidence of overdispersion, therefore I will use a negative binomial regression (Cameron and Trivedi, 1998; Greene, 2003).
In order to test hypothesis 2 and 3 new variables are created in the panel data set by multiplying the moderating variables with the four network centrality measures. Thereafter, these new variables are included in the analysis to analyze their moderating effect. Control variables and dummy variables are included in the analyses as well.
5. Results
5.1 Sample statistics and correlations
Prior to the main test results will be the descriptive statistics discussed, the descriptive statistics including the correlations can be found in table 1. The independent variable, dependent variable, moderating variables and control variables are included in table 1. The mean of the dependent variable, breakthrough inventions, is 0.52 while the standard deviation is 1.45 with a minimum of 0.00 and a maximum of 16.00. The independent variable, network centrality, is presented by four different measures. Apparent are the differences between these measures, the correlation of the centrality measure closeness is positive but not significant, while the other measures are positive and significant. In addition, closeness centrality has a particular low correlation (0.04) contrary to the other measures; degree centrality (0.33), betweenness centrality (0.34), eigenvector centrality (0.40). Therefore, closeness centrality is the singular centrality measure that is not correlated with innovation performance, while the remainder centrality measures are correlated and more similar to each other.
*= p<0.05, significance in parenthesis
Table 1: Correlation matrix
1 2 3 4 5 6 7 8 9 10 11 1. Breakthrough innovation 1.00 2. Degree centrality 0.33* (0.00) 1.00 3. Betweenness centrality 0.34 * (0.00) 0.81 * (0.00) 1.00 4. Closeness centrality 0.04 (0.15) (0.47) -0.02 (0.10) -0.05 1.00 5. Eigenvector centrality 0.40 * (0.00) 0.94 * (0.00) 0.75 * (0.00) -0.06 * (0.03) 1.00 6. Relatedness 0.13* (0.00) 0.25 * (0.00) 0.22 * (0.00) -0.06 * (0.05) 0.26 * (0.00) 1.00 7. Geographical distance 0.02 (0.58) 0.13 * (0.00) 0.19 * (0.00) (0.69) -0.01 0.11 * (0.00) 0.14 * (0.00) 1.00 8. R&D intensity 0.04 (0.12) -0.00 (0.73) -0.01 (0.68) 0.08 * (0.00) -0.01 (0.63) 0.03 (0.36) 0.02 (0.53) 1.00
9. Focal firm financial leverage 0.03 (0.37) -0.04 (0.13) -0.05 (0.07) -0.07 * (0.01) -0.04 (0.16) -0.01 (0.63) -0.17 * (0.00) -0.13 * (0.00) 1.00
10. Focal firm prior performance -0.07 * (0.01) 0.03 (0.29) -0.00 (0.94) 0.02 (0.54) 0.04 (0.13) 0.06 * (0.04) 0.03 (0.33) -0.03 (0.22) -0.07 * (0.01) 1.00
11. Focal firm size -0.02
5.2 Main results
The negative binominal regression test is used to analyze the dataset, the main effects of the analyses can be found below in table 2. Model 1 consist of all the baseline attributes of the data set and includes all control variables and their effect on the dependent variable. Thereafter, model 2 through model 5 reflects the effect of the different centrality measures on breakthrough inventions. The continuation of table 2 starts with models 6 through 9 in which the tests for hypothesis 1 are displayed. The following continuation table, number 3, consists of models 10 through 13, which are the models for hypothesis 2. The fourth and final continuation table of the main results consist of models 14 through 17 and are the test models for hypothesis 3.
Table 2 – Regression results
* =p<0.1 ** = p<0.05 *** = p<0.01 Standard deviation in parenthesis
Model 1 Model 2 Model 3 Model 4 Model 5
1. R&D intensity 0.023 0.030** 0.027* 0.023 0.030**
(0.104) (0.043) (0.052) (0.105) (0.035)
2. Focal firm financial leverage 0.270 0.587 0.608* 0.233 0.609
(0.465 (0.110) (0.098) (0.534) (0.086)
3. Focal firm prior performance -0.098 -0.086 -0.071 -0.101 -0.099
(0.347) (0.361) (0.458) (0.336) (0.288)
4. Focal firm size 0.003 -0.050 -0.050 0.003 -0.050
(0.925) (0.114) (0.115) (0.936) (0.104) 5. Degree centrality 6.230*** (0.000) 6. Betweenness centrality 14.167*** (0.000) 7. Closeness centrality 0.010 (0.285) 8. Eigenvector centrality 6.253*** (0.000) Constant -0.748** -0.847*** -0.745** -0.761** -0.880*** (0.020) (0.003) (0.010) (0.018) (0.002) Observations 1292 1292 1292 1292 1292
Year dummy YES YES YES YES YES
Firm dummy YES YES YES YES YES
Country dummy NO NO NO NO NO
ll -1070 -1033 -1037 -1070 -1024
The results show that three out of the four centrality measures included in this study show a significant relationship with breakthrough inventions. Closeness centrality is the only centrality measure that is not significant and has little to no effect on breakthrough inventions (β 0.010, p = 0.285). Betweenness centrality is the measure with the greatest effect on the dependent variable (β 14.167, p < 0.01) while respectively degree centrality (β 6.230, p < 0.01) and eigenvector centrality (β 6.253, p < 0.01) have relatively strong effects as well.
Table 2 continued – Regression results
* =p<0.1 ** = p<0.05 *** = p<0.01 Standard deviation in parenthesis
Model 6 Model 7 Model 8 Model 9
1. R&D intensity 0.030** 0.027** 0.024* 0.030**
(0.035) (0.046) (0.098) (0.033)
2. Focal firm financial leverage 0.588 0.600* 0.282 0.574
(0.122) (0.097) (0.456) (0.104)
3. Focal firm prior performance -0.089 -0.071 -0.103 -0.096
(0.350) (0.453) (0.329) (0.303)
4. Focal firm size -0.057* -0.054* 0.001 -0.054*
(0.070) (0.085) (0.971) (0.080) 5. Degree centrality 12.625*** (0.000) 6. Betweenness centrality 22.766*** (0.000) 7. Closeness centrality 0.054 (0.243) 8. Eigenvector centrality 10.181*** (0.000) 9. Degree centrality2 -15.221*** (0.000) 10. Betweenness centrality2 -42.937*** (0.000) 11. Closeness centrality2 -0.001 (0.322) 12. Eigenvector centrality2 -10.122*** (0.001) Constant -0.907** -0.763** -0.778* -0.900*** (0.002) (0.007) (0.016) (0.001) Observations 1292 1292 1292 1292
Year dummy YES YES YES YES
Firm dummy YES YES YES YES
Country dummy NO NO NO NO
ll -1025 -1032 -1069 -1022
In order to test hypothesis 1, the quadratic term of each centrality measure is included in the negative binomial regression test in models 6 through 9 (Haans et al., 2016). Similar to the regular tests described above, three out of the four measures show a significant result while closeness centrality shows no significant result. The first centrality measure, degree centrality,
shows in model 6 that the coefficient for network centrality is statistically significant and positive (β 12.625, p < 0.01). Whereas the other side, the coefficient for network centrality squared, is
statistically significant and negative (β -15.221, p < 0.01). Models 7 and 9, respectively betweenness centrality and eigenvector centrality, show a coefficient that is statistically significant and positive (β 22.766, p < 0.01; β 10.181, p < 0.01) as well, while the squared term is statistically significant and negative (β -42.937, p < 0.01; β -10.122, p < 0.01). Model 8 shows no significance, similar to the first test and therefore closeness centrality is the only centrality measure that is not significant. Taken together, these results support the hypothesized inverted U-shaped relationship between network centrality and breakthrough inventions for the centrality measures degree, betweenness and eigenvector (Haans et al., 2016). For the network centrality measure closeness, the findings do not support the hypothesized inverted U-shaped relationship between network centrality and innovation performance.
Table 2 continued – Regression results
* =p<0.1 ** = p<0.05 *** = p<0.01 Standard deviation in parenthesis
Model 10 Model 11 Model 12 Model 13
1. R&D intensity 0.030** 0.029** 0.024* 0.030**
(0.034) (0.037) (0.096) (0.032)
2. Focal firm financial leverage 0.535 0.626* 0.302 0.522
(0.133) (0.082) (0.424) (0.131)
3. Focal firm prior performance -0.090 -0.093 -0.102 -0.092
(0.338) (0.325) (0.329) (0.319)
4. Focal firm size -0.058* -0.049 0.001 -0.057*
(0.061) (0.115) (0.979) (0.063) 5. Degree centrality 16.182*** (0.000) 6. Betweenness centrality 53.653*** (0.000) 7. Closeness centrality 0.047 (0.381) 8. Eigenvector centrality 16.096*** (0.000) 9. Degree centrality2 -30.003*** (0.00) 10. Betweenness centrality2 -346.447*** (0.000) 11. Closeness centrality2 -0.001 (0.410) 12. Eigenvector centrality2 -32.066*** (0.000)
13. Degree centrality x geographical distance -0.001* (0.094)
14. Degree centrality2 x geographical distance 0.003**
(0.016)
15. Betweenness centrality x geographical distance -0.006** (0.002)
16. Betweenness centrality2 x geographical distance 0.052***
(0.001)
17. Closeness centrality x geographical distance
18. Closeness centrality2 x geographical distance
19. Eigenvector centrality x geographical distance
20. Eigenvector centrality2 x geographical distance
0.000 (0.963) 0.000 (0.807) -0.002** (0.014) 0.005** (0.002) Constant -0.916*** -0.854** -0.775** -0.889*** (0.001) (0.003) (0.016) (0.001) Observations 1292 1292 1292 1292
Year dummy YES YES YES YES
Firm dummy YES YES YES YES
Country dummy YES YES YES YES
ll -1021 -1026 -1069 -1017
The final hypothesis, number 3, is tested by models 14 through 17 and tests the moderating effect of relatedness on the relationship between network centrality and breakthrough inventions. The greatest moderating effect relatedness has is on the relationship with the centrality measure betweenness (β 361.725, p = 0.11). However, this measure is not significant although it is close to
p < 0.1. Degree centrality shows a significant and positive result for the moderating effect (β 45.532, p < 0.01). Since the coefficient of the interaction term between degree centrality squared
and relatedness is positive, the inverted U-shaped relationship is flattening with an increase in degree network centrality (Haans et al., 2016). In addition, the same effect occurs when relatedness is added as a moderator by eigenvector centrality (β 33.323, p < 0.1), although this moderating
effect is less strong. Relatedness has no significant moderating effect on closeness centrality (β -0.001, p = 0.694). Therefore, for the network centralities degree and eigenvector hypothesis 3
Table 2 continued – Regression results
*=p<0.1 ** = p<0.05 *** = p<0.01 Standard deviation in parenthesis
Model 14 Model 15 Model 16 Model 17
1. R&D intensity 0.032** 0.028** 0.024* 0.031**
(0.030) (0.045) (0.093) (0.031)
2. Focal firm financial leverage 0.633* 0.599 0.344 0.607*
(0.084) (0.101) (0.377) (0.088)
3. Focal firm prior performance -0.090 -0.096 -0.103 -0.093
(0.349) (0.333) (0.328) (0.322)
4. Focal firm size -0.061* -0.056* -0.001 -0.055*
(0.050) (0.075) (0.988) (0.076) 5. Degree centrality 26.921*** (0.000) 6. Betweenness centrality 55.480** (0.014) 7. Closeness centrality 0.040 (0.427) 8. Eigenvector centrality 18.034*** (0.001) 9. Degree centrality2 -55.220*** (0.000) 10. Betweenness centrality2 -404.036* (0.076) 11. Closeness centrality2 -0.000 (0.600) 12. Eigenvector centrality2 -41.311** (0.020)
13. Degree centrality x relatedness -15.253** (0.011)
14. Degree centrality2 x relatedness 45.532**
(0.006)
15. Betweenness centrality x relatedness -32.759 (0.150)
16. Betweenness centrality2 x relatedness 361.725
(0.112)
17. Closeness centrality x relatedness 18. Closeness centrality2 x relatedness 19. Eigenvector centrality x relatedness 20. Eigenvector centrality2 x relatedness
0.030 (0.791) -0.001 (0.694) -8.648* (0.084) 33.323* (0.068) Constant -0.953*** -0.762** -0.775** -0.916*** (0.001) (0.007) (0.016) (0.001) Observations 1292 1292 1292 1292
Year dummy YES YES YES YES
Firm dummy YES YES YES YES
Country dummy NO NO NO NO
ll -1021 -1031 -1069 -1021
5.3 Robustness check
In order to verify the robustness of the previous outcomes, alternative analyses are applied. In the regular analysis a negative binominal regression test is used to test the hypotheses. Since the variance and mean are divergent this test was the most suitable, however for the robustness check one can assume that the variance and mean are more or less equal to each other. When assuming this condition a poisson regression is a suitable test. In addition, the inventive quantity variable is winsorized in order to exclude that outliers will influence the outcoming results. As appendix A shows is breakthrough inventions, the innovation performance variable, non-normal distributed with outliers. As a result, these outliers might influence the results. Therefore, breakthrough inventions is winsorized in such a way that the observations outside the 10th and 90th percentile are equal to the 10th and 90th percentile (Ghosh and Vogt, 2012). Furthermore, an additional regression test is conducted because the dependent variable contains many zeros. To handle the issue of having a highly skewed dependent variable with many zeros, a tobit analysis is a suitable test to use due to censoring (Greene, 2003). After all the robustness checks the results remain largely consistent as in the first analysis. In the tobit analysis was hypothesis 3 the only different result because this hypothesis is not confirmed for the four centrality measures. The other test results were consistent to the regular analysis in this study. Therefore, this study found robust results that hold across multiple tests, which can be found in appendix B.
6. Discussion
This study is the first to research the relationship between network centrality and innovation performance in a CVC setting. The foregoing findings indicate that the network centrality of a CVC entity does affect the innovation performance of the parent organization. Three out of the four centrality measures, namely degree, betweenness and eigenvector show an inverted U-shape relationship with innovation performance consistent with hypothesis 1. An interesting finding is that the other centrality measure, closeness centrality, is found to not correlate with innovation performance. In addition, all hypotheses are rejected for closeness centrality.
knowledge base of the corporation. As a result, the venture might possess little novel knowledge for the corporation to learn from and foster innovation due to similarities in knowledge bases (Cohen et al., 1990; Sapienza et al., 2004). Another possible explanation might be that the problem of Newman (2010) influenced the results by assigning higher centrality scores to actors less central due to short paths, which is explained in the methodology.
This paper proposed two moderating effects on the inverted U-shape relationship between network centrality and innovation performance. Hypothesis 2 cannot be confirmed, the results show that geographical distance does not moderate the inverted U-shape relationship. Hypothesis 3, the moderating effect of relatedness, does moderate this relationship by flattening the U-shape relationship by two centrality measures and can partially be confirmed. Betweenness centrality had a moderating effect but came just short to be significant.
A possible explanation why hypothesis 2 cannot be confirmed is that the transfer and absorption of knowledge, which is needed for fostering innovation, is not depending on the physical distance but on the knowledge bases. With the rapid technological developments in the world, physical boundaries might become irrelevant. More important is that the acquired knowledge is suitable to the corporation. With modern technologies the physical boundaries that played an important role decades ago become trivial in more recent time.
Hypothesis 3 can partially be confirmed given the results. The centrality measures degree and eigenvector show results that relatedness does moderate the inverted U-shape, while betweenness shows a strong moderation but is short from significance. Closeness centrality shows no moderation for relatedness, as it shows no relationship in the other hypotheses as well. Similar root causes arise for closeness centrality by hypothesis 3 as in the remainder hypotheses.
Therefore, I can state that relatedness does moderate this relationship by two centrality measures and is an important factor. A potential explanation why betweenness centrality is not significant in hypothesis 3 but is significant in hypothesis 1 might be because betweenness assigns higher scores to actors who are knowledge brokers. In case of a network with related organizations there might be less need for a knowledge broker because they are more familiar with the actors in the network. Therefore, in case of unrelated ventures the position of knowledge broker is more important. When relatedness is used as a moderator this might influence the results.
