Learning dynamics in research alliances: A panel data analysis

Abstract

The aim of this paper is to empirically test the determinants of Research Joint Ventures’ (RJVs) group dynamics. We develop a model based on learning and transaction cost theories, which represent the benefits and costs of RJV participation, respectively. According to our framework, firms at each period in time weigh the benefits against the costs of being an RJV member. RJV dynamics can then be interpreted as a consequence of this evolving trade-off over time. We look at entry, turbulence and exit in RJVs that have been set up under the US National Cooperative Research Act, which allows for certain antitrust exemptions in order to stimulate firms to co-operate in R&D. Accounting for unobserved project characteristics and controlling for inter-RJV interactions and industry effects, the Tobit panel regressions show the importance of group and time features for an RJVs evolution. We further identify an average RJVs long-term equilibrium size and assess its determining factors. Ours is a first attempt to produce robust stylized facts about co-operational short- and long-term dynamics, a neglected dimension in research co-operations, but an important element in understanding how collaborative learning works.

Introduction

Though being competitors in the product market, firms often co-operate in research and development (R&D). The US National Cooperative Research Act of 1984 (NCRA) was created precisely for this purpose. The NCRA stimulates firms to co-operate in R&D in non-equity ventures on a large scale, thereby aiming to provide a solution to perceived competitive threats to US high-tech industries (Link et al., 2005). As a natural consequence of the goal of promoting such broad and loose co-operations, one characteristic of the NCRA collaborations stands out: firms frequently enter and exit the NCRA research consortia after their initial formation.³ The aim of this study is to explore the drivers of these in-and-out movements, as a first step in deepening our understanding of the dynamics of research collaborations.

There is growing recognition that instabilities are a central feature of not only the NCRA research consortia, but of inter-firm co-operations in general. Indeed, several empirical studies have provided evidence that collaborative agreements are inherently unstable organizational forms (Barkema et al., 1997, Beamish, 1985, Dussauge et al., 2000, Franko, 1971, Gomes-Casseres, 1987, Killing, 1983, Kogut, 1989, Li, 1995, Park and Russo, 1996, Pennings et al., 1994). These movements, as Ariño and de la Torre (1998) and Doz and Hamel (1998) state, can be an important indicator of the learning processes inside alliances and of the net benefits that firms obtain from participating. According to this line of reasoning, firms enter research collaborations with the expectation of learning, but adapt over time and alter their commitments, which may ultimately lead to exit (Balakrishnan and Koza, 1993, Kogut, 1991, Hamel, 1991, Khanna et al., 1998, Koza and Lewin, 1998, Reuer and Zollo, 2005). Therefore, given the importance of collaborative learning for the NCRA program – and, of course, for research alliances in general – it seems key to identify the drivers of these dynamics, with the idea of giving initial insights on how firms co-operate and learn in research alliances.

Following Reuer and Zollo (2005), we draw upon two different streams of the institutional economics literature for our framework. On the one hand, based on evolutionary economics we argue that learning represents the benefits of alliance membership.⁴ On the other hand, through transaction cost theory and industrial economics reasoning, we identify the costs of its participation. Alliance dynamics can then be interpreted as an evolving interplay of these benefits and costs.

First, according to evolutionary economics, organisms evolve through the formation and marginal adjustment of behavioral patterns (Nelson and Winter, 1982). Adapted to organizations, evolutionary economics highlights the role of the tacit accumulation of knowledge through learning (Teece et al., 1997). Research co-operation is then considered a mechanism to facilitate the transfer of certain types of knowledge and to enhance a firm's learning processes (Teece, 1986, Hagedoorn, 1995). Therefore, this reasoning focuses on the benefits of alliance participation.

Second, however, co-operative learning in an alliance is complex. The transaction cost view of Williamson (1992) applied to R&D collaborations highlights the particular characteristics of these ventures that give rise to various exchange hazards, such as free-riding or coordination costs among partners (Oxley, 1997, Veugelers, 1998). Thus, transaction cost theory focuses on the costs of the involvement in alliances.⁵

We combine these two dimensions – evolutionary economics and transaction cost theory – to analyze drivers of alliance dynamics. We focus on factors that influence the post-formation entry and exit-movements of firms into and out of an NCRA research alliance. The underlying rationale of our framework is that firms at each period of time weigh the (expected) benefits of learning against the (expected) costs of free-riding and coordination in these research collaborations. In particular, we use the following reasoning. If a firm considers the benefits to be higher than the costs, then it enters an alliance.⁶ Thereafter, given that firms alter their commitments, change investments and learn from other firms, they re-evaluate the costs and benefits of being in the research collaboration. Thus, if at one point in time the costs of participation are higher than the benefits, the firm exits. Our reasoning is thus reminiscent to Osborn and Hagedoorn's (1997, p273) proposal to study “evolutionary dynamics [of alliances] as a trade-off between transaction cost economies and technological development”.

The NCRA research consortia, being non-equity alliances, are a more effective environment for learning than equity forms such as research joint ventures (RJVs), which are more likely to stress control issues (Hagedoorn and Narula, 1996). The negative side of consortia, vis-à-vis equity ventures, however, is facing a higher uncertainty and having more diffuse goals for undertaking research (Aldrich and Sasaki, 1995). R&D consortia, therefore, have both potentially higher and more volatile costs/benefits than other types of R&D collaborations, which makes them particularly interesting to investigate alliance dynamics. As an aside, it must be noted that, though R&D consortia are different from RJVs because their members pool research resources into loose and relatively long-term projects, R&D consortia are normally labeled RJVs. From now on, we will stick to this terminology.

We focus on the two drivers of dynamics that we believe to be most relevant in our setup. First, evolutionary economics highlights the role of knowledge accumulation and development of routines over time. Therefore, we focus on how the age of the RJV influences its forces. Second, transaction cost analysis focuses on the organizational characteristics that give rise to various exchange hazards. In particular, several scholars have identified these costs as increasing with the number of agents that interact, due to suffering larger coordination costs or to opportunistic behavior becoming more severe (e.g. Holmström, 1982, Milgrom and Roberts, 1992). On the other hand, being a member in a larger RJV gives access to a larger pool of knowledge (Bloch, 1995, Veugelers, 1998), which may increase the benefits of learning. Therefore, we focus on how an alliance's group size, i.e. the number of participants (“insiders”), affects this trade-off and consequently influences its dynamics. We further investigate whether industry characteristics and interactions between different RJVs have an impact on an alliance's evolution.

We identify RJV movements on two levels. First, we analyze how group and time characteristics determine its short-run dynamics; i.e. we investigate how the number of insiders and the age of an RJV influence its entry and exit patterns. We find that the group-variables are robust drivers in a non-linear way. Especially the found U-shaped impact of the number of insiders on entry into an RJV may be of interest. This result indicates that the perception of transaction cost problems at first goes up with the number of insiders, i.e. entry initially decreases with size. This is indirect evidence for the classic theory of teams where moral hazard increases with the number of agents (Holmström, 1982), yet in a dynamic rather than a static context. However, those RJVs that are very large, experience more entry when they further increase in size. This hints at some RJVs having overcome the typical problems of large groups through an optimal design of their organizational variables. Firms perceive, therefore, entrance into these larger RJVs to be beneficial, given the larger pool of knowledge available. Further, the age of an RJV has a negative effect on both entry and exit, suggesting that RJVs become more stable over their lifespan, which may be due to divergent learning (Nakamura et al., 1996) – and hence firms become more complementary over time – or due to RJVs becoming more effective over time in dealing with transaction cost problems, because of the development of trust and more efficient routines (Chiles and McMackin, 1996, Parkhe, 1993).

Second, once short-run dynamics are analyzed, one can go a step further and investigate whether a long-term equilibrium size exists – i.e. if entry and exit evolve around a long-term stable group size. We find that this is indeed the case, which confirms – again in a dynamic rather than static context – RJV-group formation models such as Bloch (1995) that predict an optimal RJV size due to the trade-off between costs and benefits of participation. Moreover, some of the industry characteristics in which the RJV is embedded are found to have a determinant impact on this group size. In particular, factors which both increase the benefits of learning and the possibility of controlling other RJV members – as, for example, the concentration of an industry – lead to larger RJVs in the long-run.⁷

Our study is novel in several aspects. First, although recent theoretical studies have indicated the importance of organizational changes in existing partnerships, very little empirical research has been conducted on these ideas.⁸ This small empirical literature takes into account how alliance termination is influenced by either host country characteristics (Contractor, 1990, Franko, 1989, Barkema et al., 1997), degrees of partner rivalry (Kogut, 1989, Kogut, 1991, Park and Russo, 1996) or partners’ previous experiences (Barkema et al., 1997, Reuer and Zollo, 2005).⁹ But, although relatively varied, these studies exclusively focus on the ending of collaboration.¹⁰ Our study adds to the empirical literature on collaborative dynamics by studying the development of still operating research ventures. This approach offers two main advantages. First, it allows a far more complete picture of an alliance's evolution over time and hence of its learning processes. Second and related, from a statistical point of view our analysis adds true time-variation to the econometric analysis by using panel data methodologies (Wooldridge, 2002). Given the focus of previous work on termination, dissolution has only been analyzed in either cross-sectional studies (e.g. Reuer et al., 2002, Reuer and Zollo, 2005) or hazard rate models (e.g. Barkema et al., 1997, Park and Russo, 1996).

Further, the size and heterogeneity of the NCRA-RJVs allow us to add a novel element to empirical studies that link transaction costs theory to alliance instabilities.¹¹ Existing empirical works focus mainly on cultural differences (e.g. Contractor, 1990), firms’ previous alliance experience (e.g. Reuer and Zollo, 2005) or firms being direct competitors in the product market (e.g. Kogut, 1989) as a source of potential problems. However, seminal papers on transaction costs, such as Holmström (1982) and Milgrom and Roberts (1992), have shown that a key factor in explaining organizational difficulties is the number of cooperating actors. Indeed, problems of coordination and opportunistic behavior typically become more problematic with more interacting agents, as e.g. Holmström and Tirole (1989) and Oxley (1997) argue. At the same time, learning benefits may be higher when the “pool of knowledge” is larger, which makes bigger alliances more attractive (Veugelers, 1998). To our knowledge, this study is the first attempt to link alliance dynamics to the size of RJVs, which should help in understanding how the number of participants impacts the learning versus transaction cost trade-off.

Finally, our focus on group dynamics allows us to determine the long-term stable RJV size, and its dependence on the elements of the industry in which it operates. This serves as a reality check for theory models on RJV formation, such as Bloch (1995) and Cassiman and Greenlee (1999). More importantly, it gives us further insights in exactly how the costs and the benefits of collaborating balance each other out at a certain alliance size. Our approach allows us to identify an “ideal” long-term alliance size in function of these costs/benefits via the characteristics of the industry in which the co-operation is established.

The structure of the paper is as follows. The second section, building on our general architecture, explains our hypotheses. The third section presents the data, while the fourth explains the estimation strategy and the chosen variables. The fifth section shows the econometric results and some robustness checks. The sixth section offers a discussion of the results and possible implications for firms and policy-makers. The seventh section concludes.