Master Thesis
Alliance Partner Characteristics in Interorganisational Collaboration
Net-works: Does Alliance Composition matter for Breakthrough Innovation?
Sarah Neumanna, *, supervised by Pedro de Fariaa, co-assessed by Thijs Broekhuizena
a Faculty of Economics and Business, University of Groningen, The Netherlands
THESIS INFO: ABSTRACT
Interorganisational collaboration has been acknowledged as beneficial for firms’ innovativeness. Pre-vious works in this field underline the notion that aside from the company’s own central position in its network, the position of its partners also plays a pivotal role. This thesis aims to ascertain to what extent the position of a firm and its partners - referred to as ‘partner-weighted alliance centrality’ - in the interorganisational alliance network in combination with certain alliance partner characteristics im-pact the development of breakthrough innovations. Building on recent theoretical development of part-ner-weighted alliance centrality and breakthrough innovation, I propose that the industry and the ge-ography type of partners in the firm’s alliance portfolio will indirectly influence the inverted U-shape relationship so that it will shift its turning point to the left. Using a data sample of 196 U.S. pharmaceu-tical companies that spans from 1985 to 2001, empirical evidence identifies partial support for the in-teraction effect of industry and geography alliance type. My findings provide important implications for research on alliance portfolios, network centrality and breakthrough innovation. The data shows no evidence that unrelated industry alliances in an alliance portfolio have any impact on realising break-through innovation when those collaboration partners are central positioned in the network. Whereas, choosing more cross-border relative to domestic partners in an alliance portfolio causes the optimal innovation performance to be reached at lower levels of partner centrality.
Keywords: Alliance Portfolios × Interorganisational Collaboration × Unrelated Industry Alliances × Cross-border Alliances × Alliance Network Centrality × Breakthrough Innovation
July 20, 2018
Word count 13,033 (excluding references and appendices). ______________________________________
* Corresponding author of MSc. Strategic Innovation Management; Student Number: S3073297. E-mail: s.neu-mann.2@student.rug.nl.
Table of Content
1. INTRODUCTION 3
2. THEORETICAL BACKGROUND AND HYPOTHESES 5
2.1. Alliances 5
2.2 Alliance Portfolios and Configuration 7
2.3 Alliances as Interorganisational Collaboration Networks: The Role of Network Centrality and
Innovation 9
2.4 Hypotheses Development 12
2.4.1 The Moderating Effect of Unrelated Industry Alliances (Alliance Industry Type) 12 2.4.2 The Moderating Effect of Cross-Border Alliances (Alliance Geography Type) 14
3. METHODOLOGY 16
3.1 Empirical Setting 17
3.2 Data Collection and Sample 17
3.3 Measurements 18 3.3.1. Dependent Variable 18 3.3.2 Independent Variable 18 3.3.3 Moderating Variables 18 3.3.4 Control Variables 19 3.4 Analytical Method 20 4. RESULTS 21
4.1 Descriptive Statistics and Correlations 21
4.2 Regression Results and Hypothesis Testing 23
4.3 Additional Analyses 27
4.4 Robustness Checks 27
5. DISCUSSION 28
5.1 Theoretical Implications 28
5.2 Managerial Implications 31
5.3 Limitations and Future Research 31
6. CONCLUSION 32
7. REFERENCES 34
1. Introduction
It is becoming increasingly critical for companies to innovate in order to sustain a competitive advantage
(Schumpeter, 1939; Dunlap-Hinkler et al., 2010). Companies endeavour to develop breakthrough inno-vations as these will allow them to outperform the competition and secure their long-term survival (Hill and Rothaermel, 2003; Tushman and Anderson, 1986). Breakthrough innovations which are operation-alized as high-impact innovations can introduce new technological trajectories and lead to paradigm shifts (Ahuja and Lampert, 2001; Phene et al., 2006). Access to new and broad knowledge domains, the exposure to emergent technologies as well as the successful ()combination thereof with existing re-sources are crucial for attaining breakthrough innovation (Rosenkopf and Nerkar, 2001; Schoenmakers and Duysters, 2010; Dong and Yang, 2015; Un et al., 2010). Given the strategic importance of knowledge for the development of breakthrough innovation and the role of interorganisational networks as conduits of such knowledge, information exchange and learning are of general interest to both schol-ars and managers (Dong and Yang, 2015; Un et al., 2010). Past research has suggested that the quality and quantity of a firm’s available resources to generate innovation is affected by their involvement in strategic alliances (Deeds and Hill, 1996; Wassmer and Dussauge, 2012). However, setting up partner-ships requires time and effort to find suitable candidates to collaborate with as well as to assimilate and comprehend their knowledge in order to utilise it effectively. As the whole process is filled with com-plexity and uncertainty, coordination between companies can be difficult but critical (Zollo, 2002; Dun-lap-Hinkler, 2010). In this light, research on alliance portfolio configuration has tried to determine the optimal composition of alliance portfolios that facilitate the best performance for the focal company (Wassmer, 2010). Thus, successfully building and maintaining an alliance portfolio becomes a key stra-tegic issue when aiming to achieve superiority in novel product development (Powell et al., 1996; Faems et al., 2005; Teng and Das, 2008).
centrality, more precisely partner-weighted alliance centrality, shows an inverted U-shape relationship with breakthrough innovation. The conclusions by Dong et al. (2017) are, to some extent, singularly focused on only private and public partners. As such, it does not analyse in more depth how specific partner characteristics can play a decisive role in the creation of breakthrough innovation and affect the role that partner centrality plays in this regard. A supplementary examination thereof can prove further the state of research on this topic. After all, such relationships do not operate in a vacuum, and contextual factors can refine or even shift conclusions entirely. Consequently, this study aims to fill this gap in the alliance network literature by focusing on the moderating effect of the industry and country in which partners are active on the relationship between partner-weighted alliance centrality and innovation per-formance of a firm. It was designed to produce new insights on the composition of interorganisational collaboration networks in order to enhance the probability to create breakthrough innovations.
This leads to the following research question:
Research question: How does the position of a firm and its partners in the interorganisational alliance network in combination with certain industry and geographical alliance partner char-acteristics impact the development of breakthrough innovations?
Regarding the research question, I hypothesise the following: First, a greater share of unrelated relative to related industry alliances in a firm’s alliance portfolio will ease the diversity of knowledge inputs from a company’s alliance network, but it will also cause more coordination and communication prob-lems due to their inherent divergence. This indicates that when those partners are positioned more cen-trally in the network, attention problems will diminish the benefits of partner-weighted centrality even further in advancing breakthrough innovation. Second, even though engaging into more cross-border relative to domestic partnerships offers access to broader and differing knowledge bases, they are com-prised of various cultures and work ethics, which may cause miscommunication and other managerial problems. Hence, I argue that the point at which a higher degree of partner-weighted alliance centrality will start to become detrimental for developing breakthrough innovations will come at an earlier stage as a firm engages in more cross-border partnerships, such that it would be more beneficial to engage in less of these partnerships altogether.
The findings contribute to literature on alliance portfolios, network theory and breakthrough innovation. I posit that only specific partner characteristics impact the extent to what a firm’s and its partners’ posi-tions in the interorganisational alliance network lead to the creation of breakthrough innovation. While the industry in which partners operate was not shown to be meaningful, their geographical origin does influence the effect of alliance network positioning and the relevance thereof in achieving breakthrough innovation in different ways.
The paper is structured as follows: Section two provides an overview of the relevant literature which I used to theoretically ground the hypotheses that drive the analysis. The subsequent section outlines the data and elaborates on the research methodology. Next, section four presents the empirical results as well as additional analyses and robustness checks. This is followed by a discussion on the findings including theoretical and managerial implications as well as limitations of and future research based on this study, after which the conclusions are drawn in the final section of this paper.
2. Theoretical Background and Hypotheses
This section introduces relevant research streams on alliances, alliance portfolios and their configuration as well as network centrality which will help to provide the reader with the necessary understanding of the contemporary discussions amongst academics. Building mainly on the knowledge-based view and network theory, I will further elaborate on alliance partner type characteristics that are related to network centrality and innovation performance differences leading to the development of my hypotheses.
2.1. Alliances
Over the last decades, alliances have become a ubiquitous phenomenon (Gulati, 1998) and are defined as voluntarily and cooperative arrangements between at least two independent companies (Gulati and Singh, 1998; Dussauge et al., 1999; De Man and Duysters, 2005). Scholars increasingly posit that alli-ances can be used as powerful tools to improve the companies’ performance and ultimately their overall value (Das and Teng, 2000; Wassmer, 2010; Dong et al., 2017).
creation, integration, transfer as well as transformation of knowledge that result in new or innovative products (Nonaka, 1988, 1994; Inkpen, 1998; Kogut and Zander, 1992). As such, this underlines the importance of knowledge itself and knowledge generation as a company’s key asset to in sustaining a competitive advantage and superior performance (Kogut and Zander, 1992). However, knowledge bases and competences are imperfectly distributed across individuals and organisations in society as well as across individuals within firms (Tsoukas, 1996; Un and Asakawa, 2015). Transferring knowledge through and across organisations is often characterised as difficult due to the fact that it resides in or-ganisations members, tasks or tools and often appears to be tacit or difficult to articulate (Nonaka, 1991; Szulanski, 2003; Ambrosini and Bowman, 2001). When taking alliances into consideration, the com-plexity of transferring knowledge will increase since it needs to be spread beyond the boundaries of a single company (Eisenhardt and Santos, 2002). Even though individuals within organisations have the ability to leverage their knowledge in order to create innovative output, firms often lack the capacity to self-supply all relevant knowledge required to innovate (Deeds and Hill, 1996; Eisenhardt and Schoon-hoven, 1996).
In this light, companies form frequently alliances. Prior research has provided a variety of reasons to explain why firms engage in collaborative agreements. Salient among the incentives to partner is the opportunity to exchange and share capabilities and resources, such as financial and human capital, tech-nology or even firm-specific assets (Gulati, 1998; Lavie et al., 2007). Thus, companies are able to access, integrate and leverage a multitude of the partners’ valuable resources which usually serves as comple-mentarities to the focal firm (Teece, 1986; Mowery et al., 1996; Ahuja, 2000a; Hoffmann, 2007; Wang and Zajac, 2007). The interplay of internal and external knowledge flows allows the firm to broaden its own knowledge base. This, in turn, provides opportunities to learn and facilitates the creation of new ideas and innovation (Kale et al., 2000; Cassiman and Veugelers, 2006; Lavie and Miller, 2008; George et al., 2001; Faems et al., 2012).
In this process, by means of alliances, companies can stimulate their productivity, make use of econo-mies of scale and scope as well as shorten the development time of products (Sampson, 2007). Alliance participants do not need to directly compensate for an inability to develop the knowledge and resources internally or alternatively gain them through an acquisition (Ahuja, 2000a; Hoffmann, 2007). Therefore, collaborations might enable participants to grow faster and more effectively.
to imitate products or services (Lorenzoni and Lipparini, 1999). In this context, Kale et al. (2002) state that strategic alliances have become vital to the overall performance of a firm.
Moreover, collaborations can reinforce the firm’s strategic position against competitors through combi-native market power or efficiency (Eisenhardt and Schoonhoven, 1996; Katila et al., 2008).
Furthermore, (international) alliances ease the business expansion into new, foreign markets and appear to be a less risky and capital-intensive investment in comparison to conventional mergers and acquisi-tions (Duysters and Lokshin, 2011).
Even though alliances exhibit considerable benefits, organisations encounter additional challenges in managing alliances successfully. Heimericks and Duysters (2007), for example, claim that the success rate of alliances is only around 50 percent. Specifically, the necessity of coordinating multiple alliances simultaneously is stated as a pitfall (Kale et al., 2002). Firms also face considerable risks when forming alliances due to the chance of opportunistic behaviour and conflicts with partners (Williamson, 1985; Silverman and Baum, 2002).
In sum, the literature has elaborated on the firm’s incentives to engage in (strategic) alliances extensively as well as the consequent effects on an organisation’s performance. Nevertheless, it is important to take into consideration that companies usually participate simultaneously in multiple alliances in an attempt to multiply the above-mentioned effects, indicating an alliance portfolio approach.
2.2 Alliance Portfolios and Configuration Alliance Portfolios
While there is a general consensus among scholars about the essence of (strategic) alliances, existing literature on the conceptualisation of alliance portfolios varies in its conclusions, depending on the the-oretical focus scholars have chosen when investigating them (Wassmer, 2010). In general, the additive perspective is most frequently utilised by researchers. Here, an alliance portfolio is defined as a collec-tion of all strategic alliances of a company (e.g. Hoffmann, 2007; Lavie, 2007; Lavie and Miller, 2008). Researchers employing the organisational learning perspective include the aspect of experience in their consideration. As such, they concentrate either on both current and past alliances or solely on the latter (e.g. Simonin, 1997; Hoang and Rothaermel, 2005). Publications embedded in the network literature identify portfolios as focal companies’ egocentric alliance networks, or the total sum of direct ties with collaborative partners (e.g. Baum et al., 2000; Rowley et al., 2000).
inherent restrictions across different alliances in a portfolio influence the focal companies’ innovative-ness, which is impossible to observe when focusing merely on individual alliances (Parise and Casher, 2003). Besides, a more wholistic approach enables companies to strengthen their market position and ability for sustained innovativeness (Faems et al., 2005, 2012). Additionally, the alliance portfolio has an impact on the network position of the focal firm (Hoffmann, 2007). By composing and managing their alliance portfolio, companies can play an active role in influencing the network in which they are embedded in (Lavie, 2006).
Hence, it becomes apparent that adopting an alliance portfolio approach is advisable because it allows researchers to take all relevant and possibly instrumental contextual factors of alliances into considera-tion (Faems et al., 2005, 2012). On these grounds, this thesis will focus on the focal firm’s alliance portfolio and how its portfolio composition impacts its position in the network and will further expand the established research in this field.
Alliance Portfolio Configuration
Several research streams are devoted to the configuration of alliance portfolios, or more specifically to the constituents and arrangement of collaborations (Wassmer, 2010). They build upon a number of dif-ferent theories like the resource-based view (Ahuja, 2000a, 2000b; Lavie, 2006), organisational learning theory (Deeds and Hill, 1996; Lavie and Miller, 2008) or the social network theory (Gulati, 1999; Goerzen and Baemisch, 2005). The variation in configurations can substantially influence the com-pany’s access to (new) information and resources in terms of quality, quantity and diversity (Hoffman, 2007; Ahuja, 2000a; Koka and Prescott, 2008; Beers and Zand, 2014) and hence impact the company’s performance (Lavie, 2007; Duysters and Lokshin, 2011).
and distribution alliances are a common practice (Levinthal and March, 1993; Faems et al., 2005; Jiang et al., 2010), wherein adjusting, refining, upscaling and the efficiency of already available products is of significance. Collaborations with downstream partners, such as customers or suppliers as well as with horizontal partners like competitors, frequently represent exploitative-orientated alliances (Faems et al., 2005; Kim and Higgins, 2007; Un et al., 2010). Those exploitative partnerships provide opportunities to share managerial, industry- or market-specific knowledge as well as offer insights into the develop-ment process of products (Gnyawali and Park, 2011; Tether and Tajar, 2008; Sobrero and Roberts, 2002).
Furthermore, in regards to the structural and relational characteristics of a company’s network, scholars have found that the position of the focal firm within the network is critical for gaining access to external knowledge in terms of quantity as well as quality (Koka and Prescott, 2008), which I will further elaborate on in the next sections.
2.3 Alliances as Interorganisational Collaboration Networks: The Role of Network Centrality and Innovation
Literature conventionally elucidates that for innovation, interorganisational collaboration networks and their structure play an essential role (e.g. Powell et al., 1996; Ahuja, 2000a). Recent innovation studies also underline the importance of the firm’s central position in the network as the advantages of an alli-ance network are not shared amongst its participants equally (e.g. Dong and Yang, 2016; Dong et al., 2017). Network centrality, therefore, refers to the degree to which an actor is positioned centrally as well as exposed to valuable knowledge in the network (Freeman, 1977; Tsai, 2001; Robinson and Stuart, 2006).
To understand the significance of network centrality for the companies’ innovation capacity it is im-portant to note that this advantage has also been researched from a variety of angles. Ahuja et al. (2003), for example, provide evidence that centrality of an individual in virtual and knowledge-intensive R&D groups predicts innovation performance. Tsai (2001) in turn, draws on the learning perspective and shows that organisational unit’s network centrality impacts innovativeness, explaining differences due to the absorptive capacity of the units. Other scholars shed light on firm-level network centrality and its effect on innovation outcomes (e.g. Gilsing et al., 2008; Dong and Yang, 2016).
In conclusion, the position of a company in a network plays a crucial strategic role. It can either facilitate or hamper innovative activities (Tsai, 2001; Dong et al., 2017) on which the latter will be elaborated on hereafter, demonstrating that the relationship is not always linear. Positive and negative consequences can be observed, depending on the degree of centrality.
When looking at lower levels of centrality, an increase therein will yield more benefits than disad-vantages, in turn leading to an improvement in innovation performance. Central network members pos-sess many connections to other participants and hence, expose the focal company to a higher degree of diverse knowledge sources (Gilsing et al., 2008; Cui and O’Connor, 2012). In this respect, central com-panies gain experience from current and prior relationships as well as overall value as potential collab-oration partners (Sampson, 2005; Zheng and Yang, 2015; Dong and Yang, 2016). On the downside, those companies spend a considerable amount of time on other members in the network. At a certain level of centrality, innovation performance reaches a threshold and starts to decrease. A partner that is very central in the network can be rather saturated in this regard as their abundance of connections to other organisations eventually lead to problems with attention allocation, timing and coordination (Laursen and Salter, 2006; Jiang et al., 2010). Thus, central partners in the network cannot allocate their resources optimally and might not be able to provide a focal company with all the attention required to facilitate effective mutual knowledge exchange in order to develop new products. In contrast, whereas firms that are not centrally positioned in the network are exposed to a far smaller range of different knowledge sources due to their lack of connections, they may be able to focus more on the focal firm. In turn, this can create reciprocal learning benefits (Tsai, 2001; Phelps, 2010; Dong et al., 2017). Con-sequently, there is a trade-off between attention and the access to more resources a central partner can give to a focal firm. Thus, a moderate level of centrality in the network is likely to facilitate learning and the pursuit of innovation the most (Dong et al. 2017; Gilsing et al., 2008).
network in terms of ties with central and peripheral partners. Figure 1 illustrates this concept and shows that a focal firm is connected with a more central positioned company in the network which in turn has multiple ties with other organisations in the network. It becomes apparent that aside from a firm’s own location in the network, the position of its partners plays a pivotal role.
Figure 1: Partner-weighted alliance centrality in an interorganisational collaboration network
However, even though Dong et al. (2017) focused on the varying influence of public and private part-ners, they take only a few other partner characteristics into consideration. Specifically, the question remains what other alliance partner attributes impact the effects of network centrality on a firm’s inno-vativeness. The authors conclude by explicitly proposing this avenue for future research in this field (Dong et al., 2017). This is an important research gap, as various types of partners grant access to unique kinds of knowledge and influence innovation outcomes in different ways and to varying degrees. It can either provide opportunities for or limitations to a firm when it comes to the development of break-through innovations. Therefore, in order to achieve superior innovation benefits, it is important to un-derstand how to organise and compose an alliance portfolio in the interorganisational collaboration net-work more effectively. By taking a contingency approach, I aim to provide a better understanding of the boundaries and contextual factors which are affecting the relationship between alliance network cen-trality and breakthrough innovation. As such, I will attempt to illustrate that this direct relationship does not constitute a universal truth for all firms, as it depends on their environment, partners and other con-tingent factors. For that purpose, I recognise the baseline relationship as a given precondition since it has been well-established by Dong et al. (2017). To expand on this, I will extensively examine the interaction effect of specific partner type characteristics on partner-weighted network centrality and breakthrough innovation. According to Schoenmakers and Duysters (2010), a combination of knowledge from various domains that are normally not interlinked tends to deliver more breakthrough innovations. In this light, I will investigate two relevant dimensions: the industry and location (or
Eigenvector Centrality
The (small) focal firm connects with the more central firm in the network.
geography) that a firm and its partners operate in, as those are likely to be sources of differing and broader knowledge and resources compared to the focal firm, which I will elaborate on this in the sub-sequent sections.
2.4 Hypotheses Development
2.4.1 The Moderating Effect of Unrelated Industry Alliances (Alliance Industry Type)
Several studies have provided evidence that collaborations with external parties benefit a firm’s inno-vativeness (e.g. Ahuja, 2000b; Baum et al., 2000). This demonstrates that companies make successful use of external knowledge sources in order to cope with the accelerating pace of technological changes and complexity in their environment (Hill and Rothaermel, 2003).
However, not all partnerships in the portfolio are equally valuable to the development of new and inno-vative products (Deeds and Hill, 1996). Scholars have emphasised that the selection of suitable alliance partners is a determinant factor for the success of the alliance and its innovative performance (Hitt et al., 1995; Faems et al., 2005). Such a choice will consist of partners with same (homogeneous) or unre-lated (heterogenous) characteristics compared to the focal firm (Un et al., 2010; Faems et al., 2012). One of those characteristics is the distinction between different industry backgrounds among collabora-tion partners in the alliance portfolio.
Low levels of alliance portfolio partner diversity indicate that companies are connected to partners from the same or related industries which are likely to share similar knowledge bases, operational routines and languages. In the case where all alliance portfolio members operate in close distance to each other, an additional similar partner will not yield any extra value and will only be able to contribute marginally in exploring new technological domains and when developing innovations (Luo and Deng, 2009). In this light, companies have only limited opportunities to access complementary assets or new knowledge (Faems et al., 2005; Knudsen, 2007; Noseleit and de Faria, 2013), which are important in enhancing the speed of new product development and innovation performance (Jiang et al., 2010). When knowledge and capabilities are similar, redundancy problems might occur as the possibilities of knowledge recom-bination can quickly become exhausted and learning opportunities from feedback as well as synergies among partners are restrained (Ruef, 2002; Vasudeva and Anand, 2011). Furthermore, partners who operate in the same field also tend to fulfil similar customer needs, which might give rise to conflicts (Knudsen, 2007). Especially, when cooperating with competitors, firms may obstruct knowledge
trans-fer to avoid unintended knowledge spillovers (Parise and Casher, 2003; Un et al., 2010). Sub-optimal and lower innovation outcomes may be a consequence (Dussauge et al., 2000; Vasudeva and Anand, 2011).
linkages can enlarge the focal firm’s knowledge pool, which may reduce the threat of core rigidities and positively influence the innovation output (Katila and Ahuja, 2002; Jiang et al., 2010). However, high levels of (technological) diversity associated with unrelated industries in a focal company’s alliance portfolio hamper the ability to absorb new knowledge, reduce the efficiency of internal R&D efforts as well as routine building (Vasudeva and Anand, 2011). Dissimilarities in orientation, interests and rou-tines as well as the usage of largely tacit knowledge might complicate learning for companies (Lane and Lubatkin, 1998; Bruneel et al., 2010; Estrada et al., 2016). Consequently, this results in a significant rise in coordination, monitoring and communication costs which, in turn, affects the firm’s corporate per-formance negatively (Combs and Ketchen, 1999; Goerzen and Beamish, 2005) and thus, decrease the rate of (breakthrough) innovations.
In this light, a surge in scholarly work finds consensus that a moderate level of distant partners (in terms of technological domains) seems to be an ideal situation to unlock the focal company’s innovativeness (e.g. Laursen and Salter, 2006; Nooteboom et al., 2007; Sampson, 2007; Keil et al., 2008; Leiponen and Helfat, 2010; Duysters and Lokshin, 2011).
From this perspective, the composition of the alliance partners of a firm in terms of partner knowledge diversity has a direct effect on its innovativeness. However, the indirect effects are also relevant in elucidating innovation performance. Particularly for the generation of breakthrough innovations, a large number of different and broader knowledge domains are required (Laursen and Salter, 2006; Schoenmakers and Duysters, 2010). This promotes the inclusion of more unrelated industry alliances in a portfolio. Companies in networks that are knowledge-diverse will therefore reach the same level of innovation performance as firms that have more central partners instead. However, allying with compa-nies that only diverge moderately from their partners can yield the most benefits from the collaboration (Sampson, 2007). As such, it may be the case that if there is a high share of unrelated industry partners within the alliance network the positive effects of partner centrality are more difficult to realise, and the latter detrimental effects are expedited. In other words, the peak innovation performance is achieved at lower levels of partner centrality for firms in highly heterogeneous networks, explained by the follow-ing.
encounter difficulties in communicating effectively as the partner is too distant. In this connection, a higher degree of unrelated industry alliances increases the complexity of managing and coordinating alliances. This will ultimately lead to a decrease in innovation performance due to management capacity limits (Duysters and Lokshin, 2011). Second, the partner gives a high degree of attention and commit-ment to upkeep their existing relationships in the network. A connection to central organisations affords the focal firm a more direct route to acquiring the requisite knowledge for developing breakthrough innovations. However, as per the established relationship between partner centrality and innovation per-formance, there are limits to the absorptive capacity of a firm (Cohen and Levinthal, 1990; Zahra and George, 2002), and so the focal firm can only take advantage of so much centrality before an even higher degree becomes detrimental. As such, the complexity derived from an increased share of unrelated in-dustry partners only adds to the cognitive limitations of the focal company’s absorptive, attention and management capacity, lowering the optimal point of centrality at which peak performance is reached. Hence, the balance between access to more resources and knowledge as well as attention the partner can provide to the focal firm shifts. The advantages of partner centrality will diminish earlier than they would in a situation wherein partner knowledge is generally more aligned. Therefore, I suggest that the turning point, or optimum, of partner centrality for a portfolio with a higher share of unrelated industry alliances will come at an earlier stage.
In sum, these arguments support the following hypothesis:
Hypothesis 1: The share of unrelated industry alliances moderates the inverted U-shaped re-lationship between partner-weighted alliance centrality and breakthrough innovation, such that higher levels of unrelated industry alliances in a firm’s alliance portfolio shift the turning point to the left.
2.4.2 The Moderating Effect of Cross-Border Alliances (Alliance Geography Type)
Additionally, I delineate one contingency that underlines the tension between the advantages and disad-vantages related to partner-weighted alliance centrality and innovation performance.
Facilitated by mounting pressures caused by globalisation and rapid technological advances, companies are looking for business opportunities outside their national borders and thereby form alliances with partners located in different countries (Sivakumar et al., 2011). This phenomenon becomes saliently relevant nowadays because technological knowledge is increasingly dispersed across the globe (Duysters and Lokshin, 2011).
Rosenkopf and Almeida, 2003; Zhang et al., 2010). For example, Kim and Inkpen (2005) discovered that in the chemical-pharmaceutical industry, among alliances, cross-border R&D collaborations have the strongest influence on learning. They provide arguments that due to greater worldwide harmonisa-tion of procedures in drug research, relevant informaharmonisa-tion become more codified, which in turn facilitates worldwide knowledge transfer and saves development time (Kim and Inkpen, 2005). An additional ben-efit has been pointed out by a study from Eisenhardt and Schoonhoven (1996) who argue that interna-tional alliance portfolios possess greater flexibility and responsiveness to market conditions.
The presented arguments extent our understanding of the direct effects of cross-border alliances on firms’ innovativeness. In addition to this, I suggest that cross-border alliances also indirectly influence the relationship between the centrality of partners in a network and innovation performance, due to the following reasons:
By means of cross-border alliances, the focal company aims to expand its existing knowledge stock which might strengthen its innovative performance. Access to resources and knowledge from the central partners as well as to the partners’ various connections is vitally important and the main reason for collaboration. Attention provided by the central partner - which is already limited as central partners need to look after a lot of connections in their network - is of less concern as the focal company is already exposed to a lot of difficulties in managing those international alliances. Differences in common goals, cultures or values can complicate the knowledge transfer due to the higher complexity of tacit knowledge (Mowery et al., 1996). Various studies presented evidence that an increased internationali-sation of an alliance portfolio negatively influences the innovation output of the focal firm (e.g. Siva-kumar et al., 2011). Research by Mowery et al. (1996) supports this argument by stating that cross-border alliances within the biotechnology industry seem to be characterised by a lower level of knowledge transfers than alliances within the same country. As a consequence of geographical and cul-tural distance, as well as challenges due to acculturation, coordination and communication problems may arise (Gulati, 1998; Hitt et al., 1996; Von Zedtwitz et al., 2004; Lavie and Miller, 2008). The ability to assess the quality of partner knowledge or building knowledge sharing routines may be reduced as companies experience a lack of common ground, which will eventually impairs learning (Ganesan et al., 2005; Jiang et al., 2010).
Therefore, the following hypothesis can be derived:
Hypothesis 2: The share of cross-border alliances moderates the inverted U-shaped relation-ship between partner-weighted alliance centrality and breakthrough innovation, such that higher levels of cross-border alliances in a firm’s alliance portfolio shift the turning point to the left.
The conceptual model, depicted in Figure 2, provides an overview of the hypotheses examined in this study. The two hypothesised interaction effects between the various constructs are tested at the firm-level.
Figure 2: Conceptual model
3. Methodology
The following sections detail the (1) industry and sample data that has been used in this study, (2) how the data has been collected as well as (3) how the variables have been constructed. The final part (4) explains the analytical strategy to test the hypotheses. Within this section, I also elaborate on the meth-odological choices that I have made. I chose a longitudinal approach to capture the moderating effect of specific alliance partner characteristics on partner-weighted alliance centrality and firm’s innovativeness over time.
∩
Centrality Partner-weighted Alliance Centrality Innovation Performance Breakthrough InnovationAlliance Industry Type
Unrelated Industry Alliance Share
Control Variables 1. Number of Partners 2. R&D Intensity 3. Financial Leverage 4. Firm Performance (RoA) 5. Firm Size 6. R&D Agreement Share
7. Year Dummies
Alliance Geography Type
3.1 Empirical Setting
The empirical setting of this study is the U.S. pharmaceutical industry primarily due to the following primarily reasons: The pharmaceutical industry represents a very knowledge-intensive field where in-novation is key in order to build and maintain a competitive advantage (Dong and Yang, 2016). Firms in this sector are well-known to have a high tendency to patent their output (Lanjouw and Schankerman, 2004). Moreover, a large number of pharmaceutical companies engage in strategic alliances to make extensive use of external sources to develop new products (Hoang and Rothaermel, 2005). This makes the pharmaceutical industry an interesting field of study for alliance portfolio research. Furthermore, the pharmaceutical sector is frequently used for empirical studies in innovation research (e.g. Dong and Yang, 2015; Rothaermel and Deeds, 2004), allowing for a high degree of comparability with previous findings.
3.2 Data Collection and Sample
This study builds on data from three archival sources utilised by Dong et al. (2017). Besides that, I collected additional information on alliance characteristics such as the firm’s primary industry and ge-ographical origin, as well as the objective of the alliance (e.g. R&D agreements) for my hypothesis testing.
First, I used the Thomson Reuters Securities Data Corporation’s (SDC) Platinum database, which con-tains the most comprehensive information on strategic alliances and its partners (Sampson, 2007), and collected data on alliance portfolio levels for the years 1985 to 2005. This resulted in 107,964 announced alliances. After that, I processed those data to only include alliances wherein at least one of the cooper-ations partners operates in the pharmaceutical sector (SIC code: 2834-2836) which lead to 8,260 part-nerships. Finally, in order to retrieve financial data for the control variables, I further reduced the sample to 2,281 pharmaceutical alliances at firm-level, including only cooperation agreements where at least one partner was a publicly listed company.
Second, for almost all control variables, Dong et al. (2017) obtained financial firm-level data for each alliance partner from the Standard and Poor’s COMPUSTAT database which includes financial, statis-tical and market information of companies worldwide. By means of using the Committee on Uniform Security Identification Procedures (CUSIP) number, a 6-digit company code, the authors matched those financial data with the sample on a firm-level basis (Dong et al., 2017).
final constructed sample includes 722 firm-year observations in a time period from 1985 to 2001, en-compassing 196 unique U.S. companies in the pharmaceutical sector.
Finally, I merged all obtained data for each firm in the final sample into on single Excel-file and ulti-mately imported them into STATA15.1 for further investigation which will be elaborated on in the following sections.
3.3 Measurements
The following subsections provide an extensive description of the variables and their measurements. Apart from the moderating variables unrelated industry alliance share, cross-border alliance share as well as the control variable R&D agreement, all other variables and the measurements thereof originated from the database of the focal paper of Dong et al. (2017).
3.3.1. Dependent Variable
Breakthrough innovation. Dong et al. (2017) followed recent research on breakthrough innovation (e.g. Srivastava and Gnyawali, 2011; Zheng and Yang, 2015) and used the count of patents granted to a company each year that received forward citations above the 97th percentile of all patents within a
spe-cific patent class as the operationalised measurement. The authors estimated a three-year time lag for breakthrough innovations since there is a time delay between knowledge inflows and subsequent results in the pharmaceutical sector (Dong et al., 2017).1
3.3.2 Independent Variable
Partner-weighted alliance centrality. Dong et al. (2017) specified partner-weighted alliance centrality as the extent to which a focal company’s partner works together with other central organisations in the alliance network, using an eigenvector centrality measure. This measure stresses the relevance of other partners that are connected to the focal company and weighs a firm’s centrality by its partners’ centrality in the network (Bonacich, 1972, 2007). Before converting to a firm-level, the authors constructed the complete alliance network in order to calculate the centrality measure (Dong et al., 2017).
3.3.3 Moderating Variables
Unrelated industry alliance share. Even though an alliance portfolio of diverse industry partners will increase complexity and coordination costs, it will also allow firms to get access to various resources and opportunities to learn (Jiang et al., 2010). As such, I compared the four-digit numerical SIC (Stand-ard Industry Classification) code of the focal firm and those of the partner companies in order to identify the primary business of the participants and used ‘Participant Primary SIC Code’ in the SDC database.
As Noseleit and de Faria (2013) found similar beneficial effects of same (intra) and related industry alliances on innovation performance, I combined those two industries (same four- and two-digit SIC codes) and only differentiated between related and unrelated industry alliances (all others). Therefore, the unrelated industry alliance share is the ratio of unrelated relative to related industry alliance partners in the focal firm’s alliance portfolio. In particular, I divided the number of unrelated alliance partners by the total number of all alliance partners a focal company had in a specific year.
Cross-border alliance share. Participants from different countries are more likely to encounter difficul-ties in collaboration due to geographical distance and cultural differences (Dong and Glaister, 2006). I extracted information regarding the partners geographic origin from the SDC database, using the column ‘Participant Nation’. Same-nation alliances are classified as domestic alliances and the remaining ones as cross-border alliances. Hence, the cross-border alliance share is the ratio of international relative to domestic alliance partners in the focal firm’s alliance portfolio. Specifically, I divided the number of cross-border alliance partners by the total number of all alliance partners a focal company had in a specific year.
3.3.4 Control Variables
On the basis that there are diverse other factors that may also affect the dependent variable, I included several control variables.
Number of partners. The number of partners provides insight into the company’s overall access to ex-ternal knowledge sources and refers to the number of unique partners in the alliance portfolio of a firm (Phelps, 2010). However, a growth in alliance partners increases the difficulty to manage alliances suc-cessfully and might void possible benefits (Doz and Hamel, 1998).
R&D intensity. Investments into R&D can significantly affect a firm’s innovation activity since they foster the absorption, creation, exploitation and the transformation of knowledge (Cohen and Levinthal, 1990; Lin et al., 2012). In order to control for the effect of a firm’s R&D intensity, Dong et al. (2017) measured the variable as the overall R&D expenditures divided by the firm’s total sales.
Financial leverage. As another control, Dong et al. (2017) established the financial leverage of a firm as being its long-term debt over total assets. This measurement may illustrate the companies’ risk thresh-old when trying to facilitate innovation activities (Wang et al., 2016).
Firm size. The size of a firm impacts the likelihood of engaging in collaboration with other companies (Ahuja, 2000b; De Leeuw et al., 2014). Moreover, size might influence the companies’ ability to assim-ilate external generated knowledge (Dong and Yang, 2016). Additionally, firm size may lead to scale effects which affects the innovation performance of that company (Lahiri and Narayanan, 2013; Cui and O’Connor, 2012). Thus, Dong et al. (2017) operationalised firm size as the natural logarithm of total sales (to account for skewness of large outliers).
R&D agreement share. It is anticipated that an agreement which explicitly aim to develop new products will foster the firms’ innovativeness. Following existing literature, explorative activities are represented in R&D alliances (Tether, 2002; Rothaermel and Deeds, 2004; Lin et al., 2012). Hence, a higher share of R&D alliances should suggest the intention to discover and create innovations. The SDC Platinum database provides such information by categorising alliances in terms of its R&D flag simply by ‘Yes’ or ‘No’. Accordingly, I calculated the R&D agreement share by counting all alliances of a focal firm which are associated with a positive R&D flag divided by the total number of a firm’s yearly alliances.
Year dummies. Lastly, I included year dummies for each year to control for the fixed effects. 3.4 Analytical Method
The dependent variable, breakthrough innovation, is a count variable and can only take non-negative integer values. Poisson or negative binominal regression models are usually suitable to evaluate count variables. However, the Poisson model requirements indicate that the mean of the distribution should be equal to its variance and over-dispersed data provides inconsistent results (Hausman et al., 1984). In most of these datasets, the value of the standard deviation exceeds the value of the mean of the dependent variable. If over-dispersion is present, it is recommended to use a negative binominal regression instead (Wooldridge, 2002). Table 1 shows that the mean of breakthrough innovation is 1.615, whereas the standard deviation is higher with a value of 5.403. Therefore, a negative binomial distribution likely describes the data well, and will be assumed from henceforth; as such, nbreg was used in Stata to regress the variables.
Given the panel structure of the data and to control for potential endogeneity and unobserved heteroge-neity between companies over time (Noseleit and de Faria, 2013), I adopted a fixed-effect specification in my analysis, treating each company as a single subject. Hence, companies were clustered (clus-ter(gvkey) in Stata). Furthermore, in order to correct for the correlation inherent to the presence of the same companies throughout the analysed timeframe, robust standard errors are applied in all analyses (Dong et al., 2017).
Furthermore, in Table 2, I used the log-pseudolikelihood as well pseudo-R2 ratios to determine whether
In order to test the interaction effect of the unrelated industry alliance share as well as cross-border alliance share on the inverted U-shape relationship between partner-weighted alliance centrality and breakthrough innovation, the following regression model for a firm i in a year t has been used:
Breakthrough innovationit+3
= b0 + b1 Partner-weighted alliance centralityit
+ b2 Partner-weighted alliance centrality2it
+ b3 Partner-weighted alliance centralityit x Unrelated industry alliance shareit [Cross-border
alliance shareit]
+ b4 Partner-weighted alliance centrality2it x Unrelated industry alliance shareit [Cross-border
alliance shareit]
+ b5 Unrelated industry alliance shareit [Cross-border alliance shareit]
+ bj Controlsit + eit
The results of this analysis will be explained in the next sections in more detail.
4. Results
This section presents the statistical results derived from this study. First, the descriptive statistics and correlations are shown. Second, the results of the negative binominal regression to test the hypotheses are described. The third part depicts the results of additional analyses that I performed to identify whether there are differences in results for other types of industry alliances. Lastly, procedures to in-crease the robustness of the analyses are presented.
4.1 Descriptive Statistics and Correlations
Table 1 reports the summary statistics and correlation matrix of the study, showing the mean values, standard deviations and the correlations between all variables.
Table 1: Descriptive statistics and correlations
Variable Mean S.D. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
(1) Breakthrough Innovation 1.615 5.403 1.000
(2) Partner-weighted Alliance Centrality 0.009 0.053 0.031 1.000
(3) Unrelated Industry Alliance Share 0.343 0.390 -0.050 -0.043 1.000
(4) Cross-border Alliance Share 0.466 0.419 -0.081* 0.036 -0.043 1.000
(5) Number of Partners 2.093 0.445 -0.034 0.305** 0.084* 0.052 1.000
(6) R&D Intensity 5.965 51.352 -0.030 -0.017 0.132** -0.068 -0.019 1.000
(7) Financial Leverage 0.106 0.176 0.057 0.030 -0.116** 0.078* -0.018 -0.051 1.000
(8) Prior Performance -0.130 0.625 0.134** 0.041 -0.060 0.097** 0.042 -0.047 -0.294** 1.000
(9) Firm Size 3.981 3.505 0.381** 0.063 -0.107** 0.138** 0.074* -0.232** 0.120** 0.461** 1.000
(10) R&D Agreement Share 0.540 0.412 -0.053 -0.028 -0.022 -0.144** 0.032 0.071 -0.072 -0.055 -0.170** 1.000
To fully preclude the presence of multicollinearity for the sample, I furthermore conducted the Variance Inflection Factors (VIF) test. The test in Table 2 shows a mean VIF of 1.190 for all variables, with prior performance presenting the highest value at 1.530, well below the suggested maximum threshold of 10 (see e.g. Robinson and Schumacker, 2009). These results indicate that no multicollinearity is present in the data and hence, corroborates the previously drawn conclusion.
Moreover, Table 1 shows that, on average, a firm is granted 1.615 patents for breakthrough innovations per year (with a standard deviation of 5.403), indicating that about 1,166 patents for breakthrough inno-vations were granted in the time between 1985 and 2001. However, there is a strong concentration thereof for certain individual firms. Whereas the majority of companies are unable to produce break-through innovations (about 74%), only a few companies frequently produce breakbreak-through innovations. Furthermore, for a given year, the average firm has 2.093 partners with an average share of 0.343 and 0.466 for unrelated industry alliances and cross-border alliances, respectively. This indicates, on aver-age, a higher degree of collaborations with related industry partners as well as with domestic partners on average.
Table 2: Variance inflection factors
4.2 Regression Results and Hypothesis Testing
Table 3 displays the results of the negative binominal regression, performed for the purpose of testing the hypotheses. The pseudolikelihood (LL) as well as Pseudo R2 confirmed that the model improved
successively from Model 1 (LL = -787.224; R2 = 12.730) to Model 6 (LL = -775.738; R2 = 14.000).
Model 1 - the baseline model - simply includes the control variables and measures their impact on break-through innovation. Number of partners displays negative and significant results (b = -0.394, p < 0.1).
Variable VIF 1/VIF
Prior Performance 1.530 0.655
Firm Size 1.510 0.662
Financial Leverage 1.230 0.812
Number of Partners 1.130 0.887
Partner-weighted Alliance Centrality 1.110 0.899
R&D Intensity 1.080 0.927
Unrelated Industry Alliance Share 1.060 0.947
R&D Agreement Share 1.050 0.947
Cross-border Alliance Share 1.050 0.954
This implies that an increase in alliance partners has diminishing effects on the innovation performance. In contrast, firm size (b = 0.434, p < 0.01) as well as R&D share (b = 0.634, p < 0.05) demonstrate a positive and significant effect on breakthrough innovation. Firm size consistently shows evidence for a positive and strong significant (p < 0.01) relationship with breakthrough innovation - in all six models. Thus, an increase in firm size (i.e. measured by total sales) triggers an increase in innovation perfor-mance. The positive results of R&D share and breakthrough innovation are also not surprising as R&D alliances are usually employed for explorative activities.
In Model 2, I added the independent variables weighted alliance centrality and partner-weighted alliance centrality squared. As already proven by the focal paper by Dong et al. (2017) and therefore used as baseline relationship for my analysis, partner-weighted alliance centrality squared exhibits a negative and strong significant relationship with breakthrough innovation (b = -44.523, p < 0.01), indicating a curvilinear relationship between those variables. Except in Model 4, this inverted U-shape relationship is present in all subsequent models. I conducted additional tests in order to validate these findings as further described in Model 6. Moreover, aside from firm size (b = 0.429, p < 0.01), R&D share (b = 0.531, p < 0.1) also remains positive and significant, but the latter effect disappears in the models that follow.
Model 3 incorporates the moderator variables. Unrelated industry alliance share (b = -0.523, p < 0.1) and cross-border alliance share (b = -0.845, p < 0.01) both show negative and statistically significant values.
In Model 4, I added the first order and quadratic interaction term for the moderator unrelated industry alliance share only. Surprisingly, the baseline relationship between partner-weighted alliance centrality as well as its squared term and breakthrough innovation demonstrate no significance (b = 6.249, p > 0.1 and b = -25.295, p > 0.1). Unrelated industry alliance share (b = -0.571, p < 0.1) as well as cross-border alliance share (b = -0.856, p < 0.01) both show negative and statistically significant values. However, the interaction term of unrelated industry alliance share and partner-weighted alliance cen-trality squared is non-significant (b = -74.825; p > 0.1).
Whereas in Model 5, I repeated the same procedure as in Model 4, exchanging just the moderators and including the first order and quadratic interaction term for cross-border alliance share. Unrelated in-dustry alliance share still demonstrates statistically significant values (b = -0.527, p < 0.1). A positive and strong relationship can be identified of the second moderator cross-border alliance share on the relationship between partner-weighted alliance with breakthrough innovation (b = 69.191, p < 0.01).
Table 3: Negative binominal regression results for breakthrough innovation
Curvilinear
Model Full Model
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 95th Percentile 99th Percentile Independent Variable
Partner-weighted Alliance Centrality 15.623*** 14.397*** 6.249 26.110*** 21.474** 20.160* 21.609*
(4.487) (4.359) (7.712) (6.472) (9.089) (11.077) (11.647)
Partner-weighted Alliance Centrality Squared -44.523*** -40.235*** -25.295 -76.123*** -74.318*** -72.952* -75.436*
(12.735) (12.958) (28.233) (18.714) (27.550) (39.409) (38.988) Moderator Variables
Unrelated Industry Alliance Share -0.523* -0.571* -0.527* -0.565* -0.418 -0.401
(0.304) (0.305) (0.303) (0.305) (0.304) (0.317)
Unrelated Industry Alliance Share x Partner-weighted Alliance Centrality 35.067 21.669 16.999 8.464
(22.715) (17.907) (20.864) (24.485)
Unrelated Industry Alliance Share x Partner-weighted Alliance Centrality Squared -74.825 -29.585 2.126 5.233
(74.160) (52.407) (74.720) (74.210)
Cross-border Alliance Share -0.845*** -0.856*** -0.820** -0.824*** -0.694** -0.907***
(0.319) (0.316) (0.319) (0.317) (0.336) (0.303)
Cross-border Alliance Share x Partner-weighted Alliance Centrality -24.290*** -24.642** -24.429** -24.028**
(8.103) (9.961) (10.185) (11.139)
Cross-border Alliance Share x Partner-weighted Alliance Centrality Squared 69.101*** 69.942*** 64.768** 69.712**
(20.901) (25.926) (29.850) (28.484) Control Variables Number of Partners -0.394* -0.359 -0.296 -0.287 -0.284 -0.276 -0.026 -0.271 (0.222) (0.251) (0.243) (0.249) (0.252) (0.258) (0.372) (0.222) R&D Intensity 0.003 0.003 0.003 0.003 0.003 0.003 0.002 0.003** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.001) Financial Leverage -0.532 -0.658 -0.450 -0.449 -0.471 -0.469 -0.890 0.162 (0.877) (0.880) (0.832) (0.839) (0.826) (0.827) (0.768) (0.941) Prior Performance -0.181 -0.207 -0.146 -0.144 -0.148 -0.146 -0.157 -0.159 (0.191) (0.186) (0.193) (0.193) (0.193) (0.194) (0.228) (0.189) Firm Size 0.434*** 0.429*** 0.422*** 0.421*** 0.423*** 0.422*** 0.423*** 0.451*** (0.056) (0.055) (0.052) (0.052) (0.052) (0.052) (0.048) (0.070)
R&D Agreement Share 0.634** 0.531* 0.268 0.269 0.272 0.285 0.289 0.266
(0.296) (0.279) (0.251) (0.249) (0.250) (0.252) (0.262) (0.217)
Year Dummies Yes Yes Yes Yes Yes Yes Yes Yes
Constant -5.375*** -5.971*** -5.189*** -5.624*** -5.141*** -5.453*** -5.946*** -6.032*** (0.739) (0.648) (0.805) (0.778) (0.802) (0.796) (0.961) (0.909) Wald Chi-Square 462.880*** 570.740*** 648.480**** 769.910*** 640.040*** 755.170*** 737.270*** 571.010*** Log Pseudolikelihood -787.224 -784.287 -777.000 -776.325 -776.238 -775.738 -905.113 -581.348 Pseudo R2 [%] 12.730 13.050 13.860 13.940 13.950 14.000 12.340 18.780 Observations 722 722 722 722 722 722 722 722
Note: Significance levels at * p < 0.1; ** p < 0.05; *** p < 0.01. Clustered robust standard errors in parentheses. Dependent variable is the number of breakthrough innovations (97th percentile).
However, the moderator cross-border alliance share and partner-weighted alliance centrality squared has a positive and strongly statistically significant coefficient (b = 69.942, p < 0.01). However, it is premature to subsequently conclude that Hypothesis 2 is supported.
Previous studies have pointed out that the interpretation of the interaction terms in non-linear regressions necessitate a thorough examination (e.g. Hoetker, 2007; Haans et al., 2016). Hans et al. (2016) empha-sise that even though the regression results of the quadratic relationship present a significant b2
coeffi-cient other tests need to be considered to derive a reasonable inference. Following Lind and Mehlum (2010), I executed a three-step procedure to test if an inverted U-shape truly exists. I found evidence and significant results in examination of the U-shape, which strengthen the regression results (see the calculation in Appendix A). Moreover, I needed to evaluate if a shift in the turning point of the inverted U-shape relationship occurs due to the statistically significant moderator. Haans et al. (2016) suggest the following equation to test the significance, which I adjusted to my case.
d Turning Point b1b4 - b2b3
d Cross-border alliance share = 2 (b2 + b4 Cross-border alliance share)2
Appendix B displays the complete calculation based on the moderator. The direction of the turning point shift is only dependent on the sign of the numerator, indicating that if b1b4 - b2b3 is positive (negative),
the turning point will move to the right (left) (Haans et al., 2016) as the moderator cross-border alliance share increases.
Following the guidelines of Haans et al. (2016), results show that the numerator of the formula is nega-tive (b1b4 - b2b3 = -329.383). The turning point is shifted to the left as the cross-border alliance share
increases. Put differently, as the cross-border alliance share increases for a given company, the point at which a higher degree of partner-weighted alliance centrality will start to have negative consequences for the ability of the company to produce breakthrough innovations will be reached progressively earlier. Furthermore, I checked the significance of the interaction term for different levels thereof, in particular for low, mean and high values, albeit I mainly found significance for low values at the 0.1 level. The low number of observations may thereby have hampered the robustness of these results. Figure 3 visu-ally displays the effect of partner-weighted alliance centrality on breakthrough innovations. As the three curves reveal, the impact on the level of cross-border alliance share and the curvilinear relation-ship shift to the left at high levels of cross-border alliance share. In this context, I also observed that the inverted U-shape relationship flattens as the cross-border alliance share increases.
cross-border alliance share. Therefore, I am confident to conclude that Hypotheses 2 is supported, and I will proceed with the interpretation of the results as is, even without absolute confirmation.
Figure 3: Moderating effect of cross-border alliance share
4.3 Additional Analyses
I further substantiate my empirical results by conducting additional tests. Surprisingly, the effect of the unrelated industry alliance share was shown not to be significant. Therefore, I examined other industries in more detail in order to evaluate if my measurement of the related industry alliance share was too imprecise. As such, I broke down related industry alliances into two variables: First, same industry alliances (same four-digit SIC code compared to the focal firm), which are most likely comprised of direct competitors of the focal firm. Second, related industry alliances without competitors (same two-digit SIC code compared to focal firm, excluding firms which share the four-two-digit SIC code compared to focal firm). Subsequently, I conducted the same negative binominal regression analysis for Model 6 (see Appendix C and Appendix D). The results are in line with the main analysis. Hence, I can assume that in my sample there is no effect of industry alliance type on the relationship partner-weighted alli-ance centrality and breakthrough innovation. However, one must be careful to interpret results that are not statistically significant. It simply means that a relationship cannot be inferred from this particular data set.
4.4 Robustness Checks
I experimented with a few alternative measures for the variable breakthrough innovation, specifically different percentile cut-offs for the forward citations of their associated patents. I used 95th and 99th
percentiles instead of 97th to examine possible divergent outcomes compared to the main model (see
Table 3). The relationship between partner-weighted alliance centrality squared and breakthrough in-novation was still negative and significant for 95th and 99th percentile (95th: b = -72.952, p < 0.1; 99th: b
0 0,5 1 1,5 2 2,5 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 N um be r of B re akt hr ough Innova ti ons
Partner-weighted Alliance Centrality
= -75.436; p < 0.1). The interaction term of unrelated industry alliance share and partner-weighted alliance squared was still not significant for either percentile (95th: b = 2.126, p > 0.1; 99th: b = 5.233;
p > 0.1). The moderating effect of cross-border alliance share on the inverted U-shape relationship between partner-weighted alliance centrality and breakthrough innovation showed consistently positive and significant coefficient values (95th: b = 64.768, p < 0.05; 99th: b = 69.712; p < 0.05). Thus, those
similar outcomes to the main estimate underline the robustness of my results.
5. Discussion
The following section discusses in more depth the empirical results and their interpretation. Further, I provide the implications of the main findings for theory and practice which might be useful for both researchers as well as managers of alliance portfolios. The section concludes with a discussion on the limitations of this study and give suggestions for future research.
5.1 Theoretical Implications
This thesis underlines the importance of the firm's own position in the network and the strengths of its ties (Gulati, 1999; Tsai, 2001) and also of the central position of its partners and their own links in the network (Stuart, 2000). Innovative developments depend on the successful exchanging and sharing of generated resources and knowledge among other partners in the firm’s network (Cui and O’Connor, 2012). However, the topics of centrality and partner selection have mostly been examined separately. This research aims to integrate both streams, based on the pre-existing notion that while a moderate level or partner centrality results in the highest innovation outcome (Dong et al., 2017), some partner types are more beneficial than others in order to access valuable external knowledge, and thereby im-proving the companies’ innovation performance.
This study was designed to produce new insights on the organisation of interorganisational collaboration networks in order to enhance the probability to create breakthrough innovations. In particular, I investi-gate the manner in which alliance partner characteristics and positionings of a firm and its partners in the interorganisational alliance network can impact breakthrough innovations. In this light, I examine the heterogeneity of partners in the alliance portfolio of the focal firm in regard to its industry and ge-ography attributes.
As such, I can contribute two major findings to this field of study, as well as a number of less significant but nonetheless insightful observations that have emerged throughout my research.
moderating effect of the share of unrelated industry alliances on the inverted U-shape relationship be-tween partner-weighted alliance centrality and breakthrough innovations. Moreover, even supplemen-tary testing based on more a granular definition of related and unrelated industry alliances provided no support for any such relationship. In general, it is impossible to say whether these results demonstrate the absence of any relationship altogether, or just that the data or analysis are insufficiently robust to establish its presence. Literature has suggested that access to a diverse array of knowledge sources is a positive determinant of generating breakthrough innovations for focal firms (Schoenmakers and Duysters, 2010). Even though distant partners offer such variety knowledge, cooperation with such part-ners suffer from complicated communication as well as a lack in knowledge overlap (Sampson, 2007; Luo and Deng, 2009; Nooteboom et al., 2007; Noseleit and de Faria, 2013). Subsequently, this leads to higher learning costs and a decreased absorptive capacity for the focal firm (Sampson, 2007; Dong et al, 2017). Unrelated industry alliances can also be less beneficial for innovation performance as it di-minishes the efficiency of internal R&D efforts (Noseleit and de Faria, 2013; Sampson, 2007). However, the locus of analysis in this study is the relationship of such knowledge sources on the effects of partner centrality. As such, a possible explanation for the absence of proof for any such effect is that the influ-ence of the share of unrelated partners is inherently complex in nature. More specifically, a high share may have an entirely different impact than a low share, but ultimately equally positive in its eventual outcome. Thus, as it was not shown to have a significant influence, it can be concluded that the detri-mental and advantageous influence which is gained through the diverse knowledge sources from the partners network can be offset by a lack of attention the partner grants to the focal company. At least in this data set, it cannot be concluded that the access to knowledge is more important than the attention of the partner.
