Research Study on Preferences for the Formation of University and Industry Partnerships

Description
Research Study on Preferences for the Formation of University and Industry Partnerships in Knowledge Transfer Networks: Money Matters:- A partnership is an arrangement where parties agree to cooperate to advance their mutual interests

Research Study on Preferences for the Formation of University and Industry Partnerships in Knowledge Transfer Networks: Money Matters
Abstract We examine network members' collaboration preferences between and within academic and commercial organizations that are members of a UK-based Knowledge Transfer Network. By introducing conjoint analysis to innovation studies, we provide insights into the relative importance of various task and partner characteristics of potential collaborations in determining preferences. Research funding is found to be the most important predictor of network members' preferences to collaborate, but its effect is moderated by firm size, prior experience and the goal of the collaboration. We suggest that the desire to control and diminish both behavioural and outcome uncertainty explains these moderating effects. KEYWORDS: Preferences, Collaboration, Matching, Research Funding, R&D, Electronics

2

1. Introduction There is little doubt that collaborations between and within industry and universities are critical for innovation. Relationships between firms are increasingly important sources of competitive advantage (Dyer and Singh, 1998), and firms that draw from university research are known to foster economic growth (Mueller, 2006). However, opportunities to form collaborative partnerships are unequally distributed across organizations (Ahuja, 2000; Mitsuhashi and Greve, 2009). The network literature suggests that being part of a network increases the chances of forming a tie with a member within that network, especially when these networks are dense (Goerzen, 2007; Schilling and Phelps, 2007). Structural positions within a network are found to be important in tie establishment, as well as spatial proximity and collocation (Narula and Santangelo, 2009; Stuart, 1998). We study collaboration within a UK-based plastic electronics knowledge transfer network (KTN). Plastic electronics is an emerging industry in which innovations in solution-based chemistry and materials science are brought together to create lightweight, robust and disposable electronic devices on flexible surfaces. New product development in plastic electronics spans the disciplines of chemistry, physics and engineering, and the electronics, printing and chemicals industries, requiring new varieties of partnerships that have not been observed before. Research has shown that especially when the knowledge base of an industry is complex and expanding, as is the case for our study, the locus of innovation tends to be found in learning networks (Powell et al., 1996). Additionally, this novel context provides an interesting extension to research that has largely focused on US biotechnology and pharmaceutical collaborations (McKelvey et al., 2003). Even within established learning networks, differences between SMEs, large organizations and universities are likely to affect chances of tie formation. We know that the institutional norms of the academic and the commercial sphere diverge (Dasgupta and David, 1994), but there is also evidence that universities increasingly patent their own research and are willing to license technologies which, in combination with businesses' reliance on external R&D, implies that the divide between open and commercial science has somewhat narrowed (Mowery et al., 2001; Thursby and Kemp, 2002; Thursby and Thursby, 2002). This tendency is also
3 _ _

_

evident in the growing literature on university entrepreneurship, intellectual property protection, and research-based spinoffs (Clarysse et al., 2007; Lockett and Wright, 2005; Mustar et al., 2006; Rothaermel et al., 2007; Vohora et al., 2004). Nonetheless, it is unlikely that the apparent convergence between industry and universities has eliminated barriers to collaborate (Bruneel et al., 2010; Siegel et al., 2003). Barriers are likely to persist and to affect organizations with different resources differentially. Fontana et al. (2006) found that the propensity to forge university - industry (U-I) collaborations depends on firm size, while Santoro and Chakrabarti (2002) conclude that size influences the task-orientation of collaborations with universities, with smaller firms focusing on core technology and larger firms on non-core technological competencies. Also, despite SMEs' lack of R&D self-sufficiency, they are less inclined to work with research institutes than larger organizations (Tether, 2002; Woolgar et al., 1998). In addition to such general tendencies, SMEs are known to have lower slack resources (George, 2005) and to face liabilities of size when they enter in collaborative relationships (Rao et al., 2008). Therefore, SMEs are in general less selective when choosing partners to collaborate with and tend to stay more local (De Jong and Freel, 2010; Narula, 2004). Besides organization size, it has been shown that experience with collaboration shields organizations from behavioural and outcome uncertainties, inherent in collaborations (Inkpen and Tsang, 2005). Also, prior partners are often favoured for future interactions, although this is not always an optimal choice (Goerzen, 2007). Various characteristics of an organization will thus influence the likelihood of establishing a partnership. The same characteristics will correspondingly determine a potential partner's likelihood of establishing a tie as well. Thus, tie formation is not only about who you are but also who wants to partner with you (Mindruta, 2012; Mitsuhashi and Greve, 2009). Using insights from matching theory (Jovanovic, 1979; Logan, 1996), the formation of inter-organizational partnerships has previously been construed as a selective matching process or an assignment game, in which each player ranks others and forms a tie with the most desirable partner who is also available (Mindruta, 2012; Mitsuhashi and Greve, 2009). Matching theory asserts that, to understand the determinants of relationships that are entered into voluntarily, it is necessary to consider the characteristics and resources that each potential partner values in all other potential partners (Logan, 1996). However, because of the two-sided assignment game in
4 _ _

_

which partners preferences have to match, existing matches provide a biased sample (Mindruta, 2012; Stuart and Sorenson, 2007). To better understand the desired match characteristics, we look at preferences for collaboration before the establishment of an actual tie. Ranking potential partners on the basis of preferred characteristics is fundamental to matching theory. Preference ordering has been examined in the choice of marriage partners, entrepreneurial ties and employment relationships (Ferris and McKee, 2005; Logan et al., 2008; Vissa, 2011) but has not explicitly addressed technology development collaborations that suffer from high outcome uncertainty and potential for opportunism (Wathne and Heide, 2000). High uncertainty in collaborations is exemplified in frequent failures to create value due to unrecognized transaction-specific investments (Madhok and Tallman, 1998). We investigate to what extent partner and task characteristics of a potential collaboration partner influence the preference for such matches (Geringer, 1991), and find that the latter trump the former. As research funding is being put forward by governments and commercial organizations to stimulate growth in the plastic electronics industry, investigating the effectiveness of different sources of research funding in influencing match preferences has relevance for policymakers. This article offers various contributions to the literature. First, we introduce matching theory to innovation studies in the context of potential collaborations between SMEs, large organizations and universities under conditions of high uncertainty. Second, to the best of our knowledge, no prior research has looked at preference ordering of desired match characteristics. We examine the relative importance of various partner and task characteristics (Geringer, 1991) in determining collaboration preferences and find that in knowledge transfer networks research funding, a task characteristic, dominates. We furthermore investigate how firm size, prior experience and collaboration orientation moderate the importance of research funding. Third, by focusing on preference ordering of match characteristics, we look at the origins of alliance formation and collaborations, which have remained understudied (Stuart and Sorenson, 2007). Our research extends Vissa's (2011) work on tie formation intentions and unpacks the drivers of preferences, avoiding the bias of existing ties. Finally, we introduce conjoint analysis to innovation studies (Green et al., 2001). While conjoint analysis had been used extensively in marketing to gain insight into people's preferences for specific product attributes, this method

5 _ _

_

has had limited use in innovation studies (e.g., Riquelme & Rickard's, 1992), and can meaningfully be extended to analyze preferences for matches for U-I collaborations. 2. Theory Development Matching theory has moved from the laboratories of experimental psychology (Herrnstein, 1961, 1970) into marketing research and consumer brand choices (e.g. Foxall and James, 2003; Foxall and Schrezenmaier, 2003). The key characteristic of this branch of matching theory is that a specific reinforcement influences the choice behaviour of subjects so that at equilibrium the subject distributes its responses in proportion to the patterns of rewards obtained by their consequences. Almost in parallel, a similar matching theory emerged from a mathematical treatise to optimize partner and college allocations (Gale and Shapley, 1962) which had a significant influence on research into employer-employee relationships (Jovanovic, 1979; Logan, 1996). Here, two actors seek a match based on their preferences about the other's resources. Therefore, this theory differentiates between the determinants and the consequences of choice (does the ego want to match with the alter?) and those of opportunity (is the alter interested in matching with the ego?) (Logan, 1996). Both the non-exclusive matching theory of Herrnstein (1961, 1970) in which the subject can constantly alter its choices to optimize the outcome and the exclusive matching theory of Gale and Shapley (1962) in which a single choice must be made between 'mutually-excluding alternatives' thus share the disposition that choice or preference is subject to some kind of reinforcement, either through an exogenous characteristic or through an endogenous match criterion. Our research builds on both traditions as we look into non-exclusive preferences (matches are not established) for and from network members that could collaborate. Recently, matching theory has been introduced to management in studies of alliances (Mitsuhashi and Greve, 2009), the selection of outside CEOs (Fahlenbrach et al., 2010), entrepreneurial tie formation (Vissa, 2011) and university-firm collaboration (Mindruta, 2012). These studies have started to write on the blank slate matching theory provides in order to improve understanding of the resources and characteristics that drive partner selection and match quality. Mitsuhashi and Greve (2009) found that compatibility of observable resources, and market complementarity drive alliance formation in the shipping industry and that matches that
6 _ _

_

score high on both characteristics improved firm performance. Vissa (2011) extended research on entrepreneurial tie formation, and found that both task complementarity - the overlap between the actor's current task priorities and the resources potentially available from the partner - and social similarity are important matching criteria that influence tie formation intentions and actual match formation. Beyond matching theory, empirical research that looks at alliance formation has found that specific characteristics of the focal organization such as resource endowments, network positions, and prior familiarity inspire the formation of a tie (Ahuja, 2000; Geringer, 1991; Gulati and Gargiulo, 1999). Implicit in most alliance research is the idea that the beneficial characteristics sought for in a partnership or collaboration, are transparent in the existing matches. This is not necessarily self-evident. Existing matches may be mismatches based on incorrect assumptions or information asymmetry or may be construed by lack of alternative. Fahlenbrach et al. (2010) for instance found that CEOs are more likely to become board members of other firms if those firms are geographically close and similar with regards to governance, financial, and investment policies, but failed to find evidence of positive impact on performance or on returns from acquisitions. Goerzen (2007) found that prior experience with a specific partner increases the likelihood of entering into another collaboration with this partner, but that such collaborations experience inferior performance. There is thus evidence that existing matches are not automatically beneficial for the partners. This might be explained by cognitive limitations and local search (Cyert and March, 1963; Simon, 1947), but could also be explained by the two-sided market dynamics of match formation: the decision to tie with a partner is fundamentally constrained by the decisions of all the other partners to establish a match as well (Mindruta, 2012). Looking at existing matches to retrospectively understand underlying preferences, results thus in a biased sample. Given that ex-post matches are not self-evident proof of ex-ante desired partner combinations, it is valuable to investigate pure preference ordering. To do this, we require information about desired matches and their characteristics. This is relevant to better understand what resource-constrained organizations such as SMEs that cannot afford to be picky when it comes to available partners, are actually looking for in terms of partners (Narula, 2004).

7 _ _

_

Matching theory is particularly useful for such research because the theory implies that people or organizations will be ranked on the basis of their perceived value to the seeking actor (Gale and Shapley, 1962). Despite its importance in the formation of inter-organizational relationships, relatively little attention has been devoted to partner selection (Li et al., 2008). Therefore, extending our understanding of which characteristics augment matching preferences is relevant to improve our broader understanding of firms' relational behaviour. 2.1 Matching Preferences for Collaboration Matching theory requires the existence of "inspection characteristics" that enable an actor to rank preferences without experiencing them. Such characteristics have been described in job matching research as online and offline search attributes (Lippman and McCall, 1976), and in marketing as experience and inspection goods (Hirshleifer, 1973; Nelson, 1970). As match preference ranking occurs before the partners have any experience in working together, they have to build their preferences using cognitive rather than experiential logics (Gavetti and Levinthal, 2000). Following Geringer's (1991) distinction, we divide the inspection characteristics of potential matches in partner- and task-related criteria. Partner-related criteria relate to the potential for effective coordination of effort, while task-related criteria are concerned with the activity in which the participants share a common goal (Geringer, 1991; Simon, 1947). Partner characteristics relate to the potential for effective coordination, and encompass considerations of national and corporate culture, compatibility, past association and trust between top management teams (Geringer, 1991). A first partner characteristic affecting match quality is familiarity. Research has shown that friends are preferred to strangers and to acquaintances in R&D alliances and that this effect is strengthened by the radicalness of the innovation goals (Li et al., 2008). Also, spatial proximity and prior colocation been found to affect the propensity to engage in R&D alliances (Narula and Santangelo, 2009), although the problems of distance can be attenuated by institutional force (Hong and Su, 2012). More generally, scholars have established that past association helps develop relationship-specific heuristics that in turn facilitate communication (Uzzi, 1997), and that repeated social exchanges generate trust (Blau, 1964; Gulati, 1995; Perrone et al., 2003), with the result that prior partners are often favored for future interactions (Goerzen, 2007; Gulati and Gargiulo, 1999). Also, familiarity with focal

8 _ _

_

employees in the collaborator organization is likely to be important, as the literature suggests that inter-personal interactions often provide a basis for the emergence of inter-organizational, border-crossing relationships (Balconi and Laboranti, 2006; Marchington and Vincent, 2004; Ring and Van de Ven, 1994). Specifically for SMEs, personal and professional relationships enable innovation diffusion within networks (Ceci and Lubatti, 2012). A second characteristic, and one that could substitute for familiarity, is partner reputation (Dollinger et al., 1997), since certain symbols of esteem, such as being a university professor, provide signals of the publicly validated nature of a partner's knowledge. It is therefore not unlikely that reputation might overcome lack of familiarity in potential collaborations. A third characteristic affecting match quality is the extent of shared ties. Strong third-party connections foster shared attitudes and empathetic communication, and increase the likelihood of social sanction if a partner behaves opportunistically (Burt, 1987; Granovetter, 1985; Reagans and McEvily, 2003). Also, cooperative ties influence firms' level of innovation performance (Tomlinson, 2010). These three partner-related characteristics have been shown to influence actual tie formation. Following a planned behaviour approach (Ajzen, 1991), we hypothesize that they will also drive matching intentions and preference ordering. Hypothesis 1a: In knowledge transfer networks, prior familiarity will increase a network member's preference for the other Hypothesis 1b: In knowledge transfer networks, partner reputation will increase a network member's preference for the other Hypothesis 1c: In knowledge transfer networks, mutual ties will increase a network member's preference for the other

In the context of exploratory R&D collaborations within a triple helix network of government, industry and university members (Etzkowitz and Leydesdorff, 2000), three task characteristics are likely to drive tie formation. A first and vital task characteristic affecting match quality is the anticipated nature of research funding between the parties (Lee, 2000). Funding has been shown to be a primary reason for universities to collaborate with industry (Meyer-Krahmer and Schmoch, 1998) and to positively contribute to academic research publications (Ubfal and Maffioli, 2011). Our empirical context allows for funding to be provided by the government, the partner, the seeking organization or by no one and thus allows
9 _ _

_

differentiating between different funding sources. Some collaboration benefits, such as access to basic science knowledge in universities, may be accessible through informal forms of cooperation or subsidized government funding, while in other cases it may be necessary to invest own funds in order to access the partner's knowledge. A second task characteristic relates to the institutional similarity, or the institutional context within which the partner's knowledge is embedded. Matching theory research has found that social similarity is beneficial to tie formation (Vissa, 2011), echoing the concept of homophily which states that similar actors will be drawn to each other (Lincoln and McBride, 1985). Despite possible beneficial effects of establishing ties between companies and universities (George et al., 2002), there is evidence that different institutional norms that govern public research organizations and private enterprises form barriers to collaboration (Dasgupta and David, 1994). Culture clashes, ineffective management of university technology transfer offices, bureaucracy and poorly designed reward systems are merely some examples of the problems that create effective barriers to collaborate (Kotha et al., 2013; Siegel et al., 2003). Recent qualitative research has on the contrary shown that institutional convergence occurs in established U - I collaborations so that a shared cultural space for knowledge exchange is created (Bjerregaard, 2010). However, ex post experience is unlikely to influence the cognitive, inspection-based intentions of the candidate partners. Moreover, when it comes to collaboration, research institutes are more susceptible to the geographical proximity of firms than to proximity of other research institutes (Fritsch and Schwirten, 1999). Additionally, SMEs are less likely to collaborate with universities (Tether, 2002; Woolgar et al., 1998). We therefore argue that companies will prefer to work together with companies and universities with other universities. A third task characteristic is the knowledge similarity between potential collaboration partners. Inter-institutional knowledge flows have been found to be of great importance in emerging research fields (Heinze and Kuhlmann, 2008). Although it can be more beneficial to work with partners with different - rather than similar - knowledge, since investment of constrained time and resources could have a greater payoff (Grant and Baden-Fuller, 2003; McFadyen and Cannella, 2004), and might be more likely to generate high value innovations (Ahuja and Lampert, 2001; Fleming, 2001), such collaborations are associated with greater outcome uncertainty. Moreover, there seems to be an optimal level of technological and
10_ _

_

cognitive distance between firms so that very dissimilar knowledge is counterproductive in collaboration (Gilsing et al., 2008; Wuyts et al., 2005). In our context of exploratory research, it seems likely that collaboration will be facilitated by shared technology and science between the partners. Research on matching theory has found that compatible resources are beneficial to alliance formation (Mitsuhashi and Greve, 2009). Given that knowledge creation is a pathdependent process (Dosi, 1982), it is likely that compatible technological knowledge that stems from the same scientific paradigm facilitates communication and increases the collaboration potential. Moreover, a lack of common knowledge makes coordination and knowledge transfer very difficult within a single organization (Grant, 1996), so straddling scientific and technological domains between different organizations will be even more difficult. Also, research problems that involve the combination of knowledge across distant scientific domains require effort and ability to overcome a lack of common ground (Bechky, 2003, 2006; Fleming, 2004; Heath and Staudenmayer, 2000). On the other hand, it has been shown that performance of heterogeneous teams in scientific alliances is better than for homogenous teams and that the initial heterogeneity is overcome in due time (Porac et al., 2004). Given that our aim is to understand match preferences based on off-line search or inspection, organizations' tendency to search locally (Cyert and March, 1963) makes it less likely that network members will be aware of potential overlap between technologies from different scientific domains. We therefore hypothesize that task characteristics (knowledge similarity, institutional similarity, and research funding) will positively influence a member's preference for a collaboration with another member. Hypothesis 2a: In knowledge transfer networks, knowledge similarity will increase a network member's preference for the other Hypothesis 2b: In knowledge transfer networks, institutional similarity will increase a network member's preference for the other Hypothesis 2c: In knowledge transfer networks, research funding will increase a network member's preference for the other

Researchers interested in match quality have examined both task (Dong and Glaister, 2006) and partner characteristics (Li et al., 2008; Pesämaa et al., 2009), but to our knowledge, there have been no studies of the relative value placed on these two sets of criteria.
11_ _

_

Understanding the relevance of different characteristics vis-à-vis preferences is necessary if matching theory is to be applied to inter-organizational collaboration. The various participants in any joint activity need to develop a shared or mutual understanding of their common ground (Clark, 1996; Clark and Marshall, 1981). More specifically, in exploratory research, establishing common ground requires participants to bridge the gap between their separate knowledge domains through a process of synthesizing and integrating their knowledge into a commonly understood area for development and activity (Cramton, 2001). Under typical circumstances, common ground cannot be established ex ante, so that collaborators have to invest time in creating a shared understanding of the problem and must develop a common language to convey and integrate results (Bechky, 2003; Kotha et al., 2013; Vural et al., 2013). Achieving this complex coordination seems easier when partners know each other from before, which would suggest that partner characteristics would dominate task characteristics. However, it has also been shown that firms derive valuable network resources from being embedded in a network (Gulati, 1999). Given that all participants in the study are derived from a single network, designed by the government to stimulate knowledge exchange and joint R&D, it could be argued that a lot of the value normally attributed to network resources, will be perceived as self-evident by the members. In other words, the common ground vital for knowledge exchange could be embedded in the network membership, which will further increase tie formation possibilities (Schilling and Phelps, 2007). If this is the case, task characteristics such as research funding, institutional and knowledge similarity would gain prominence over the partner characteristics. We hypothesize that partner characteristics are indeed perceived as embedded in the network environment provided by the relatively small membership (less than 800 members) of our focal Knowledge Transfer Network. This leads to the following hypothesis: Hypothesis 3: In knowledge transfer networks, task characteristics will increase a network member's preference for the other more than partner characteristics 2.2 The contingent value of research funding While we assume economically rational behaviour (receiving funds is better than not receiving funds or paying for collaboration) to hold in general, there are factors that likely

12_ _

_

moderate and potentially reverse this economic truism. These factors relate to size, experience and collaboration goals. First, because of the need to leverage limited resources efficiently, we can expect research funding to be particularly important to the perceived value of potential partnerships for decision-makers in small firms and entrepreneurial ventures (Starr and MacMillan, 1990). New ventures and small firms are likely to benefit strongly from research funding to enhance innovation, because they have a liability of size when entering collaborative partnerships due to constraints on their resources (Rao et al., 2008; Tether, 2002). In addition, small firms are likely to have limited financial slack resources that can be easily diverted (George, 2005). Rothaermel and Deeds (2004) found that small biotech firms were more likely to enter into exploration and exploitation collaborations than larger firms as the latter had sufficient slack to discover, develop and commercialize projects through vertical integration. This limited financial slack is likely to mean that decision-makers in SMEs are keen to opt for partnerships in which their own financial resources are not depleted. This leads to the following hypothesis: Hypothesis 4a: In knowledge transfer networks, research funding will be more salient for an SME member than for universities and larger organizations

While hypothesis 4a is rooted in the idea that SMEs' financial constraints drive them towards collaborations in which resources are provided, there is a counterargument to be made. Despite SMEs' lack of financial slack, the costs associated with the bureaucracy embedded in projects funded by the government or the risk of misappropriation in projects funded by others might exceed the organization's administrative capabilities or their risk-aversion. Large organizations on the other hand possess these capabilities, and their decision-makers may be less likely to think that mitigating behavioral uncertainty will increase match quality, since their organizations have greater scope to absorb the consequences of unexpected outcomes (Cyert and March, 1963; Patzelt et al., 2008). As collaborations between different organizations are arenas for opportunism (Das and Teng, 1998), one of the best ways to maintain control is providing the necessary financial resources yourself. Specifically for U-I collaborations, there is evidence that academic norms evolve because of research funding and that industry-funded university-industry projects result in more applied research (Benner and Sandström, 2000; Gulbrandsen and Smeby, 2005). When organizations know this, they might be tempted to increase control over the
13_ _

_

coordination mechanisms and collaboration outcomes by holding control over or by providing the research funding. This leads to the alternative hypothesis: Hypothesis 4b: In knowledge transfer networks, research funding will be less salient for an SME member than for universities and larger organizations Further, prior experience with forming alliances or collaborations might attenuate the need to control the relationship. Relational capabilities, built up from prior experience of collaboration, compensate for behavioural uncertainty associated with a liability of disconnectedness and facilitate knowledge integration (Powell et al., 1996; Tzabbar et al., 2012). Organizations with such capabilities are therefore less likely to perceive a need for control, because they have developed strategies for managing the process of establishing common ground so that their goals are achieved (Dyer and Singh, 1998; Kale et al., 2000). Firms can leverage pre-existing networks when searching for new partners, and although there is a tendency for searches to be restricted by existing ties (Gulati and Gargiulo, 1999; Stuart and Podolny, 1996), network embeddedness appears to facilitate the identification of common ground, thereby assisting in determining potential match quality. Mitsuhashi and Greve (2009) for instance found that in the shipping industry, alliances of firms embedded in collaboration networks had better match quality. Moreover, the relational view argues that firms differ in their ability to generate rents from the complementary resources offered by partners, and that certain relation-specific assets help firms leverage the latent value in their partnerships (Dyer and Singh, 1998). Firms with relational capabilities therefore better manage outcome uncertainty as well, as they have learnt how to control and manage inherently uncertain projects. Bruneel et al. (2010) show that prior experience and trust lower specific barriers to U-I collaborations. Firms without prior experience of collaboration in a network of alliances are required to evaluate match quality without the benefit of prior knowledge provided by existing partners. This knowledge limitation creates a disadvantage for non-connected firms, since behavioural uncertainty arising from a lack of knowledge about how to establish common ground is higher. Therefore, we posit that research funding arrangements that provide a means to control the partner's use of resources are less important for firms with prior collaboration experience.
14_ _

_

Hypothesis 5: In knowledge transfer networks, experienced members will be more willing to cede control of funded projects than less experienced members Finally, while firms are increasingly looking outside their boundaries to build technological capabilities by accessing patents, know-how, equipment or materials generated as a result of the partner's R&D activity (Chesbrough, 2003; Hung and Tang, 2008; Laursen and Salter, 2005), such activities are characterized by high uncertainty as the outcomes of exploratory technology-oriented R&D are very difficult to anticipate (Shah and Swaminathan, 2008). Within an industry, it has been shown that incumbents prefer exploitation alliances that leverage complementary assets over exploration alliances that build technological capabilities that are characterized by higher uncertainty (Rothaermel, 2001). Moreover, firms seeking external sources of technological capability are typically required to invest financial and other complementary resources in order to secure the partner's cooperation. Given concerns about their resource commitments, firms seeking to build technological capabilities via partnerships are likely to face a greater need to mitigate outcome uncertainty. Therefore, such firms might be willing to sacrifice research funding to maintain or increase control over the outcome of the project. This leads to the following hypothesis: Hypothesis 6: In knowledge transfer networks, members seeking to develop new technological capabilities will desire greater control of collaboration projects 3. Research Method 3.1 Sample We collected data from participants in a technology network serving the needs of the plastic electronics industry in the UK. At the time of the study, the network was one of approximately twenty Knowledge Transfer Networks (KTNs) funded by the UK Government. The objective of KTNs is to improve the UK's innovation performance by increasing the breadth and depth of the knowledge transfer of technology into UK-based businesses and by accelerating the rate at which this process occurs (Technology Strategy Board, 2012). The KTN we studied supported organizations developing plastic electronics technology for displays and lighting, including small and medium-sized enterprises, original equipment manufacturers and universities. Between its inception in Spring 2005 and December 2008, the KTN organized 87 events, including seminars,

15_ _

_

partnering workshops, tutorials and dissemination events. Membership of the KTN was free so that barriers to join the network were low. In October 2008 it had approximately 800 members from 500 organizations listed in its database, including a strong representation of senior executives from corporations. Its membership was comprised as follows: 64% from companies, 20% from universities, with the remainder made up of consultancies, government and other business support organizations. Invitations to fill in an electronic survey were issued to 667 members of the KTN (all members with valid email addresses) in November 2008. After two email reminders, full responses were received from 201 members, a response rate of 30%, by the end of December 2008. In the early part of the survey, respondents were asked to indicate their job roles to identify whether they were managers, scientists and engineers who were in a position to exercise discretion over choice of collaboration partner. Of the 201 respondents, 50 were in technology transfer or business development roles and 151 were in research-related roles. Analysis of nonresponse using t-tests suggested that respondents were significantly more likely to have attended events organized by the network (69% of respondents had attended events, compared with 53% in the whole population p< .001), were more likely to have prior experience of Collaborative R&D (53% of respondents compared with 40%; p< .001) and were more likely to work for universities than companies (32% worked for universities, as opposed to 22% in the population as a whole). This non-response analysis suggests that the findings may reflect a bias toward those with a greater motivation for finding collaborative partners, which is not a concern, given that the purpose of the study is to identify characteristics affecting match quality amongst those who are actively considering collaboration. 3.2 Methods To test the hypotheses, we needed an analytical approach that could extend beyond a simple ranking of preferences and help examine how preferences are weighted when they potentially compete with one another. We therefore introduce conjoint analysis to innovation studies. Conjoint analysis is a measurement technique that originated in the fields of mathematical psychology and psychometrics and can assist researchers in sorting out a product's multidimensional attributes (Green and Wind, 1975). Typically, conjoint analysis is used in
16_ _

_

marketing research to understand preference structures underlying consumer buying decisions as it has great potential for measuring trade-offs between multi-attribute products and services (Green et al., 2001; Green and Srinivasan, 1990). It is based on the premise that preferences can best be determined by asking consumers to judge products defined by combinations of attributes, rather than by judging single attributes one at a time. By systematically varying attributes of the product/match and observing how respondents react to the resulting product profiles or scenarios, researchers can deduce the importance of each individual attribute. Despite its prominence in marketing research, "conjoint measurement's potential is not limited to consumer applications" (Green and Wind, 1975). The method is suitable in situations of multi-attribute decision-making and is especially useful for studying perceptions and judgements of respondents (Riquelme and Rickards, 1992). Conjoint analysis is suitable because respondents were not asked to rate or rank match characteristics explicitly; instead they were asked to rate scenarios that described a bundle of characteristics, enabling us to examine how potential competition between match characteristics is resolved. Although tailored routines exist for analysing conjoint data (such as SPSS conjoint), conjoint analysis is a special case of GLM. Since our data are nested as respondents expressed preferences about scenarios, we clustered our data by respondents1. GLM regressions are considered robust and useful as they control for unobserved heterogeneity caused by the unique characteristics of the respondents that might influence the dependent variable. Furthermore, as the effect of these unique individual characteristics might be random, we use a random-effects model. We use the XTREG command in Stata 12 to test our models. Our stepwise estimating procedure started with our baseline model (model 1), in the next step we added main effects- partner related (model 2) and task related (models 3a and 3b) and finally, followed by interaction variables (table 3). In the GLM models, we used the value of each scenario as the dependent variable, the scenario characteristics as the explanatory variables

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
1We

also clustered our data according to the different scenarios as a robustness check. The results did not change significantly

17_ _

_

and respondent fixed-effect characteristics as control variables. The robustness checks2 gave similar results and are only reported for the full model in table 3. 3.3 Dependent variable: preference score for specific scenario We examine differences between respondents' preference scores of different scenarios for exploratory collaboration. Before being shown the scenarios, respondents worked through a series of questions about exploratory collaboration. After indicating their job role and their company's expertise, the following introductory text asked them to imagine establishing a collaborative relationship with another organization in the KTN network: "Assume that you are looking for a partner to help with exploring future commercial opportunities and/or to help develop your research or technology for the future. I will refer to these projects as exploratory projects." They were asked to indicate what type of partnership would be most useful, and to identify the nature of expertise they would want the partner organization to contribute to such a project. The scenarios were then introduced. Each scenario described a potential match for exploratory collaboration in terms of six attributes of the potential collaboration: member seniority, whether they were known to the respondent, whether any ties were shared, the type of organization they worked in (university or company), whether their knowledge was similar to that of the respondent, and the research funding for the collaboration. These six attributes provide a source of data on respondents' match preferences. They are listed with their levels, which were varied across scenarios following an experimental design (described further below), in Table 1. The first set (research funding (4), organization type (2) and knowledge similarity (2)) was intended to capture task-related preferences, and the second set (member seniority (3), prior familiarity (2) and shared contacts (2)), partner-related preferences. Insert Table 1 about here Four types of research funding were included: providing one's own funding, having funding provided by the partner, receiving government funding, and an informal arrangement
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
2

F o r a d d itio n a l ro b u s tn e s s , w e ra n th e re g re s s io n s a g a in u s in g a m a x im u m lik e lih o o d e s tim a to r (S ta ta c o m m a n d :

xtreg DV IVs, mle nolog), a restricted maximum likelihood estimator (Stata command: xtmixed DV IVs), and the specific Stata command for complex survey sampling designs (svyset, psu (clustervariable) /// svy: regress DV IVs)_

18_ _

_

where no funding was available. Of these, the highest level of control over the objectives of the collaborative project would be where the respondent's organization funded the collaboration and the lowest where the partner funded it. Each respondent was presented with a set of eight scenarios, preceded by the following introductory text: "Now imagine that you've identified several different companies or universities who could provide you the expertise you requested on the previous screens. I'd now like to ask you what type of person you would find worth collaborating with, in different types of funding arrangement. The ideal partner is not always available so it is often necessary to make compromises in choosing collaborators. On the following screen, I'm going to ask you to evaluate how worthwhile you would find it to collaborate in different scenarios". Every respondent rated between 3 and 8 different scenarios (average 7.7) and each scenario was rated by 3 (due to incomplete responses) to 16 different respondents on a scale from 0 (not worthwhile) to 10 (extremely worthwhile). This score provided our dependent variable. Although some scenarios were only rated a few times, this is not a problem for the conjoint method. Hybrid forms such as the ones we used have been found to compare favourably with traditional full profile models (Riquelme and Rickards, 1992). As the goal of conjoint analysis is to compare the relative importance of specific attributes in the context of differing alternatives on all other relevant attributes, what really matters to the explanatory power of the model is how many times each attribute was judged individually. In our design, every attribute value appeared minimally 379 times in an individual scenario. Given that there were 192 different scenarios (4*2*2*3*2*2), a full factorial design in which respondents were presented with all possible permutations of match characteristics and levels would require them to rate 192 scenarios. This would have vastly decreased response rate and would have placed excessive cognitive strain on the data-supplying capabilities of respondents (Green et al., 1981). In our study, the 192 scenarios were divided into 24 blocks of eight scenarios (using a 'confounded blocks' design) and each respondent rated one block of eight, a number suggested by Green et al. (1981). 201 respondents rated a total of 1546 scenarios. The mean value across all 1546 scenarios was 4.78, with minimum 0 and maximum 10. 3.4. Independent variables

19_ _

_

Six scenario attributes. True to our initial intention of using inspection-based measures that are observable at low cost (Gavetti and Levinthal, 2000; Hirshleifer, 1973; Nelson, 1970), we use task (research funding, institutional similarity and knowledge similarity) and partner (shared ties, familiarity and seniority, which is used an inspection proxy for reputation) characteristics that can be easily obtained within the KTN. The six match characteristics are entered into the models as independent variables representing characteristics of potential collaborators (Table 1). A dummy variable for each level of each characteristic relative to a reference category was entered. Technological capability. Respondents were asked what expertise they would want a collaboration partner to contribute. They were given a choice of nine options, such as 'researches/develops materials , technology or equipment for plastic electronics' and 'manufactures and uses displays and lighting products'. The list was derived from interviews with members of the KTN, developed in consultation with industry experts and refined in response to a pilot test of the survey. Those who selected one of the first three options (seeking a partner who is researching and/or developing technology for plastic electronics, displays or lighting) were deemed to be seeking a technological capability. 80 respondents (39.80%) were seeking to build technological capabilities. SME Respondent. We control for firm size as it is known that in university-industry (U-I) collaborations, firm size - total number and R&D-specific number of employees - has a positive impact on the propensity to ally with a public research organization (Fontana et al., 2006; Laursen and Salter, 2004). Each organization was coded to indicate whether it was a university, a large company (with over $50m in sales and/or 500 employees) or a small company (less than $50m in sales and less than 500 employees). Of the 201 respondents, 115 (57.21%) worked in SMEs, 33 (16.42%) worked for large organizations and 53 (26.37%) worked for universities. Prior collaboration. A dataset from the UK Government's Technology Strategy Board was used to determine whether each organization had previously participated or was currently participating in a government-funded collaborative R&D project in a technology area relevant to plastic electronics. We obtained data on all projects funded as a result of invitations issued between April 2004 and Spring 2007, within programs that addressed plastic electronics,
20_ _

_

advanced composite materials and structures and other disruptive technologies in electronics and displays. Spring 2007 was an appropriate end point because there is typically a year's lag between the call being issued and the project getting started. Of the projects funded in the Spring 2007 call, the earliest started in January 2008, and the latest in June 2008. These projects would have been running for less than a year before the questionnaire was issued in November 2008. Of the 201 respondents, 93 (46%) worked for organizations that had been involved in governmentfunded collaborative R&D projects. Seniority. Respondents were asked to self-classify their role using a range of options. 23 respondents were working as lecturers, research associates and junior scientists in companies and were coded as Junior (11.44%). The 62 respondents who were senior lecturers or readers in universities or group leaders in companies were coded as Middle (30.85%). 66 university professors and R&D or other Directors were coded as Senior (32.84%). For respondents with hybrid roles, spanning both universities and companies, background research was conducted to determine from public sources which category represented their primary role. 50 respondents were not in research-oriented roles. 26 (12.94%) of them were in technology development roles and 24 (11.94%) were categorized as others. 4. Results Table 2 shows the results of random-effects GLS regression. Model 1 incorporates only the control variables, and shows significant differences in means between various types of respondents3. Juniors, and to a lesser extent, business developers and mid-career respondents exhibit higher preferences for collaborations. The relatively small number of junior and business development respondents might be responsible for this positive bias. There was no significant difference between SME, large organization or university respondents. Nor were there significant differences between those seeking technological capability and those who were not, or for those with and without prior collaboration experience. Insert table 2 about here
_ 3_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
_ Note that the default category is "other respondent". While it is common practice to use the largest group as default

category, we chose to insert "senior" explicitly as a control variable because partner seniority is one of the main effects we are interested in._ _

21_ _

_

We test hypotheses 1a, 1b and 1c in model 2 by adding the main effects of three variables capturing partner characteristics. Respondents prefer to collaborate with senior people (b = 0.94, p ? 0.001) more so than they do with mid-career people (b = 0.59, p ? 0.001). Familiar partners are preferred to unfamiliar partners (b = 0.82, p ? 0.001), and partners that have shared ties are preferred to those who do not have shared ties (b = 0.38, p ? 0.01). This indicates partner characteristics indeed influence network members' collaboration preferences. Hypotheses 2a, 2b and 2c were tested in models 3a and 3b. We found no support for knowledge similarity's influence on match preferences. Contrary to our theoretical reasoning, knowledge similarity is no significant predictor of network members' collaboration preferences. This could be explained through the narrow scope of the knowledge network. Given that members of the knowledge transfer network are all members of the same emerging industry, the differences in technology and scientific background might not be large enough to result in significantly different behaviour. Model 3a shows that there is a general preference to work with companies (b = 0.38, p ? 0.001) but tells us little about institutional similarity. To test explicitly for institutional similarity, we needed to exclude the "Organization Type" variable - to avoid issues of multicollinearity - and include variables for institutional similarity (respondents from universities preference for working with universities and respondents from SMEs or large firms working with companies). We had argued that homophily would increase the likelihood of companies working with companies and universities with universities, and find strong support for such an effect with companies (b=0.84, p? 0.001) but not with universities. In spite of the general belief that companies join KTNs to work together with universities, the homophily hypothesis cannot be refuted for companies while universities express no preference in working with other universities or companies (U-I collaboration is the default). Hypothesis 2c was strongly supported in that both government-funded (b=0.83, p?0.001) and partner-funded (b=1.95, p? 0.001) collaborations were preferred to the default category of no funding. We also find that providing own resources has a significant and negative coefficient (b=-0.88, p? 0.001). This supports the overall argument of economically rational behaviour in both academic and commercial organizations. The stronger coefficient in front of partner-funded schemes could be explained from an administrative slack
22_ _

_

perspective. As government funding is typically associated with bureaucratic obligations, network members' choose partner funding over government funding. Alternatively, exploratory R&D efforts funded by the partner might be preferred because of the implied conviction of the partnering organization. When a partner is willing to put her own money on the line, she is likely to be more devoted to a positive outcome, which is beneficial for all involved parties. Hypothesis 3 proposed that task characteristics would be more salient than partner characteristics in determining match preferences. The difference in total R2 of 0.14 between Models 2 and 3a4 supports this hypothesis (changing from 0.08 to 0.22). Also, inserting only the task-related characteristics and the control variables results in an increase in R2 from 0.03 to 0.17 (not reported), while inserting partner-related characteristics and control variables only increases R2 from 0.03 to 0.08 (Model 2, Table 2). This shows the dominance of task characteristics over partner characteristics in knowledge transfer networks. A visual way to illustrate the importance of task-related preferences, and specifically the importance of research funding, is to use parameter estimates to calculate the importance of each characteristic in determining the overall scores reported by respondents. We standardized the absolute value of the coefficients in model 3 (only main effects) and compared their relative contribution to the expressed preference for collaboration. We depict the three partner and three task characteristics relative to their explained variance (18.59%) with partner characteristics explaining 24.42% of explained variance (R2 from 0.033 to 0.076) and task characteristics explaining 75.58% of explained variance (R2 from 0.076 to 0.219). From Figure 1, it can be seen that funding accounts for 65.44% of the variation explained by task and partner characteristics. Insert Figure 1 about here Research funding is clearly the dominant differentiator within the six investigated characteristics. Riquelme and Rickards (1992) argued that conjoint analysis is suitable to differentiate between non-compensatory and compensatory decision rules. The dominance of research funding, which explains more than 55% of total variation (i.e. including control

_ 4_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
_ We_ focus_ the_ analysis_ on_ model_ 3a_ rather_ than_ 3b_ because_ "organization_ type"_ w

as_ an_ explicit_ part_ of_ the_ scenarios_ and_ hence_ this_ model_ matches_ better_ with_ the_ conjoint_ me thodology._ _ _

23_ _

_

variables), suggests that research funding might provide a non-compensatory decision rule for exploratory R&D collaborations. Table 3 includes the interaction terms and the robustness checks. There is very little difference between the alternative models so we focus our discussion on the GLS model (far right, last column of Table 3). Dummy variables denoting SME respondents, prior experience of collaboration and respondents seeking technological capability were interacted with research funding. The effects of respondents being either juniors or business developers and the main effects are still significant and belonging to an SME has become slightly negative (b= -0.59, p?0.10). As before, a collaboration where funds are provided by government or a partner is more attractive than a self-funded collaboration, or one without funding, although government funding has become slightly less significant. Some interaction terms are significant, confirming that there are differences in the perceived value of research funding arrangements. Insert table 3 about here Hypotheses 4a and 4b provided alternative assumptions about the effect of research funding on matching preferences of SMEs, relative to universities and large organizations. 4a stated that SMEs lack financial slack and would therefore be more willing to engage in collaborations where funding was provided, while 4b hypothesized that SMEs are more susceptible to outcome uncertainty due to limited slack and would, in order to increase control over the collaboration, be willing to accept exploratory R&D collaboration without research funding. Following the introduction of the interaction effects, only the interaction with partner funding is significant (b=0.57, p ? 0.10, Table 3) but not strong enough to overcome the negative effect. Funding is thus more important for SMEs than for large firms or universities, which provides some support for hypothesis 4a. The rather weak support for this hypothesis could potentially be explained by the finding that large firms often engage in U-I collaborations to share costs, which indicates that resource constraints on R&D are not unique to SMEs (Veugelers and Cassiman, 2005). Hypothesis 5 proposed that compared to organizations without prior collaboration experience, organizations with prior experience would care less about control through research
24_ _

_

funding and thus be more inclined to engage in research collaborations that are funded by partners or governments. In Model 4, we find partial support for this hypothesis. The interaction effect for partner funding and prior collaboration is positive and significant (b=0.74, p? 0.05), while the interaction with government funding is positive but not significant. Figure 2 depicts these two interaction effects. Insert Figures 2 and 3 about here Whereas respondents attributed more value to partner-funded arrangements than they did to not-funded collaborations (main effect for Partner Funding: b = 1.82, p ? 0.001), this was more so for respondents with prior experience (coefficient for Partner Funded x Prior Collaboration: b = 0.74, p ? .05). For government funding, the effects are in a similar direction, but the interaction term is not significant. Nonetheless, as figure 2 shows, even the plot for government versus other funding (which includes the powerful partner-funding) shows clearly different slopes. These results suggest that those with experience apportion more value to arrangements where they haveless control (partner funding and government funding). Hypothesis 5 is partially supported. Hypothesis 6 proposed that organizations seeking to develop technological capabilities will desire greater control and hence be less susceptible to funding provision. In table 3 the three interaction terms between funding and seeking technological capability have the expected sign. Although respondents attributed more value to government-funded arrangements than they did to a not-funded collaboration (main effect for Government Funded: b= 0.90, p? 0.05), this effect was attenuated for respondents seeking technological capability (Government Funded x Seeking Tech Capability: b= -0.76, p ? 0.05). Partner funding (b=1.82, p ? .001) was less attractive for those seeking technological capabilities than for those seeking other types of partner, shown by the negative interaction term (Partner Funded X Seeking Tech Capability: b = -1.30, p ? .001). The interaction between self-funding and seeking technological capabilities attenuated the negative main effect but was not significant. The two significant interactions are depicted in figure 3. Those seeking technological capabilities are less interested in external funding sources than those seeking other capabilities. We did not find significant interactions between

25_ _

_

organizations that sought market capabilities or operational capabilities and research funding (not in table). Hypothesis 6 was supported.

26_ _

_

5.

Discussion & Future Research We found that task characteristics, driven by research funding, are more important

predictors of matching preferences than partner characteristics. As argued before, this result is likely to be contingent on our context of knowledge transfer networks in which the ease of collaboration is facilitated by the common ground established in the network, which creates mutual understanding and a common language amongst members (Bechky, 2003; Clark, 1996; Kotha et al., 2013; Vural et al., 2013). Whereas research funding had the highest explanatory power, this effect was partially moderated by firm size, prior experience and seeking technological capabilities. While the theoretical rationale for our interaction effects was largely rooted in the importance of control, we find partial support for this notion. Organizations with prior collaboration experience are slightly more likely to prefer collaborations where funding is provided by others than those without collaboration experience. Firms that have been engaged in prior collaborations have developed mechanisms to deal with outcome and behavioural uncertainty that to some extent shield them from the uncertainties associated with collaborating (Inkpen and Tsang, 2005; Uzzi, 1997). Also, we found that in exploratory R&D projects where technological capabilities need be developed, high outcome uncertainty attenuates organizations' desire to receive external funding. Analysis of the marginal effects of organization type and funding source shows that:_ SMEs are significantly less likely to prefer collaborations than universities; (2) that SMEs' collaboration preferences are not significantly different from large companies; and (3) large companies' collaboration preferences are not significantly different from universities. This suggests that the traditional disadvantage of size in establishing matches (Fontana et al., 2006; Narula, 2004) is not only a consequence of partners not wanting to collaborate with SMEs, but also of a lower preference of SMEs to collaborate in general. _ Another finding suggests that when it comes to collaborations, it matters where the funding comes from. Hence, the value of money in collaborative research seems to be contingent on the perceived strings that are attached to it. The strong preference for partner funding (main (1)

27_ _

_

effect) and the three significant interaction effects with respectively SME, prior collaboration, and seeking technological capabilities indicate that the source of funding is more important than who receives it, whether the receiver has prior relational experience and what the funding will be used for. These findings are theoretically relevant because they challenge the notion that prior collaboration experience can readily be interpreted as conducive to future collaboration. Our data suggest that the importance of prior experience is contingent on where the funding comes from. In our context, in which prior collaboration is defined as "having been engaged in a government-funded project before", this finding suggests that prior experience in working in a government funded project does not make organizations significantly more likely to do this again. It however makes organizations more likely to engage in a partner funded project. This suggests that there are partnership skills, developed in previous collaboration with government funding, that are transferable to other projects without the government. The attenuating influence of "seeking technological capabilities" on funding suggests as predicted that the extent to which the reasons for collaboration (such as development of technological capabilities) influences collaboration preferences will be moderated by the source of funding. The attenuating influence is stronger for partner funding (slope -0.81) than for government funding (slope -0.27), which supports the control argument. When partners are funding the exploratory research project they are more likely to maintain control over the resources so that it becomes harder for the partner to develop technological capabilities. When governments fund projects, this effect is weaker. Taken together, our findings about the effects of firm size, seeking technological capability and prior collaborative experience on network members' collaboration preference make a contribution to the literature on how partners are selected for collaboration, addressing Hitt et al.'s (2000) call for research on the contingent nature of selection criteria. 5.1 Contribution to university - industry collaboration research This research, to our knowledge, is among the first in management to explicitly look at matching preferences rather than at established ties or partnerships. Research into alliances has commonly inferred match preferences from established matches (Mindruta, 2012; Mitsuhashi and Greve, 2009) and although some attention has been paid to matching intentions (Vissa,

28_ _

_

2011), no explicit ordering of preferences has been found in the literature so far. By focusing on desired match characteristics and their relative rank order, we respond to Sorenson & Stuart's (2007) call for more research on the origins of tie formation. Moreover, our research design does not suffer from the biased estimates and skewed conclusions associated with alliance studies that regress an innovation output on various covariates without taking into account the fundamentally strategic (and thus endogenous) nature of matching decisions on the results (Mindruta, 2012). Whereas the literature suggests that both task and partner characteristics are important (Geringer, 1991; Li et al., 2008), our finding that task characteristics are more salient in determining match preferences than partner characteristics extends our understanding of drivers of collaborations and tie formation. However, Casciaro and Lobo (2008) found that interpersonal affect trumps task-related competence within a company's boundaries. We found that within an established knowledge transfer network, the reverse holds true. Future research could provide a better understanding of the exact dynamics of preferences, for instance by looking both within and outside a company for collaboration preferences. We have extended research on why commercial organizations collaborate with universities (Geisler, 1995). Research typically focuses on why firms collaborate with universities (e.g. Hanel and St-Pierre, 2006) or why universities collaborate with firms (Santoro and Gopalakrishnan, 2001) or even why academic scientists collaborate with each other (Bozeman and Corley, 2004), but very little research has investigated why and under what circumstances, firms prefer to ally with other firms or with universities and vice versa. Also, most research in the university-industry intersection focuses on large firms, with some exceptions (Fontana et al., 2006; Hadjimanolis, 2006). Our research extends this field in three ways. Firstly, by not only focusing on large organizations we increase understanding of the drivers of SME collaboration preferences. Secondly, by introducing matching theory to university-industry collaborations we extend the theoretical framing of the field beyond the more common resourcebased and institutional perspectives (Boardman, 2009). Thirdly, our approach laid bare some differences between drivers of collaborative R&D projects. As our research allowed firms and universities to choose both within and between themselves, it provided some insights into whether firms prefer to work with firms or with industries and vice versa. Despite the common
29_ _

_

belief that firms join knowledge transfer networks to collaborate with universities, we found that institutional similarity is a significant predictor of preferences for firms but not for universities who seems to be indifferent between collaborating with either a firm or another university. Preference for institutional similarity was especially strong for SMEs as they drove the similarity effect completely (this specific result is not reported due to space limitations). This adds to research that compares the relative preference of firms over universities as collaboration partners. Bercovitz and Feldman (2007) for instance found that firms prefer to work with universities rather than other firms when conflicts over intellectual property are likely. Also, the finding confirms that SMEs are indeed less interested in collaborating with universities (Woolgar et al., 1998). Further research should delve deeper and investigate for which specific collaboration goals firms prefer firms over universities and vice versa. 5.2 Conjoint analysis We introduced conjoint analysis to innovation studies. This method allows for the weighing of complex match characteristics, embedded in different scenarios (Green, 1984; Green et al., 1981; Green et al., 2001; Green and Srinivasan, 1978, 1990; Green and Wind, 1975). In doing so, it explicitly incorporates important interactions between different attributes in a not necessarily linear way. Unlike the individual attributions of characteristics' importance, the conjoint approach compares and weighs more realistic scenarios to derive actual preferences. This methodology thus provides a novel way of investigating preferences, perceptions and judgements of companies that could prove fruitful in different managerial fields. The results of the conjoint analysis demonstrate a more complex interdependency among preferences than has previously been identified, suggesting that it is not only important to take various attributes into account in predicting network members' preferences, but also to consider the weighting of these attributes in the preference function. However, the method has some drawbacks as well. It is for instance important to note that our findings are dominated by research funding. Implicit in the methodology of conjoint analysis, is the belief that people are able to weigh different attributes of a focal object (Green et al., 2001). When one of the attributes appears to be non-compensatory (Riquelme and Rickards, 1992), this has a severe impact on the appreciation of the other characteristics. Future research that uses conjoint analysis could improve granularity of findings
30_ _

_

by developing scenarios in which specific attributes that are expected to be non-compensatory, are excluded. This would allow for greater variance amongst the compensatory attributes. 5.3 Implications for policy For governments, the objective of establishing knowledge transfer networks in which a governmental organization serves as a mediator and/or financial resource provider between industry and university is typically to stimulate R&D activity that would not otherwise take place by improving the returns for each party. Our findings suggest that the value of such arrangements to participants derives chiefly from research funding. Collaborating in government-funded projects is perceived as worthwhile, and more so than trying to access partners' knowledge through informal cooperation, because commitment to shared objectives within formally structured projects provides a basis for reducing uncertainty about the outcome of the cooperation. Our findings suggest that research funding is responsible for 65% of the formation of matching preferences. This has an impact on national innovation systems (Sharif, 2006). Motohashi (2005) found that SMEs in Japan achieve higher productivity thanks to collaborations with universities and that in China government funding effectively induces more matches between universities and firms (Motohashi and Yun, 2007). Additionally, Dickson (1983) found that government funding can indeed influence the direction of innovation in non-trivial ways. In combination with our findings, this suggests that constrained government resources are well spent through KTNs. Nonetheless, projects funded by a partner were generally preferred over governmentfunded project, which could be explained by relative absence of bureaucratic requirements associated with partner funding, easier coordination because no government stakeholder is involved, and the heightened involvement of the partner who funds. Future research could investigate which of these possibilities carries most weight. Besides the importance of funding, we also found that partner characteristics are important drivers of collaboration preferences. Mutual ties, prior familiarity and partner reputation all contribute to the desire to establish new matches. Government networks thus have an important social function that enables the building

31_ _

_

of trust and familiarity that co-shape collaboration preferences. However, the attenuating influence of seeking technological capabilities 5.4 Limitations The research has some limitations as well that potentially constrain the applicability of the findings. First, the size of the sample is relatively small. Although we investigated the preferences of 201 respondents for 1546 scenarios containing 6 focal characteristics, a larger sample would bolster external validity of our findings. Also, the focus on a single KTN could lead to some bias, though we find no reason to believe that the findings are not largely valid for other established networks as well. While we could have opted for conducting this research in multiple KTNs, our theoretical framing is contingent on within-network dynamics. As these dynamics might be very different in other networks due to differences in common ground (Bechky, 2003, 2006; Fleming, 2004), we chose to limit ourselves to one KTN within which we were able to get a high response rate. Besides, investigating multiple KTNs would have caused problems of construct validity as we would have needed tailored survey designs for each KTN. Finally, common method variance might be a problem because all but one variable stem from the same survey. Although various methods exist to account for common method bias (Lindell and Whitney, 2001; Podsakoff and Organ, 1986; Podsakoff et al., 2003), a recent simulation study has shown that all these methods actually deliver less accurate results than those without corrections (Richardson et al., 2009). Importantly, Siemsen et al. (2010) reported that including more independent variables from the same survey reduces common method variance and that interaction effects are never artefacts of method variance because they are always attenuated. Therefore, we followed the advice of Conway and Lance (2010) and refrained from using any statistical method to diminish common method variance. 6. Conclusion We researched collaboration preferences in a UK knowledge transfer network both from the firm-perspective and the university-perspective. By using a novel methodology, we were able to rank different task and partner characteristics according to their relative salience in explaining preference formation. Our data suggest that overall, companies and universities are economically
32_ _

_

rational and prefer collaborations in which they are not the financial resource provider. However, for collaborations that are crafted to develop technological capabilities, this effect was significantly less pronounced, which indicates that the focus of collaboration moderates the importance of research funding. This implies that specific combinations of collaboration characteristics are less preferred because of the heightened behavioural and outcome uncertainty associated with them, especially in a context of an emerging industry like plastic electronics. We also found that experience with prior collaborations has a small and positive moderation effect on the importance of research funding on collaboration preferences. Organizations that developed relational capabilities experience a lower need for control while organizations engaged in technology development collaborations attempt to control behavioural and outcome uncertainty through funding.

33_ _

_

Table 1: Match Characteristics and Levels used in Conjoint Analysis

* denotes the reference category used in the GLM regressions.

34_ _



doc_129792529.docx