Seite - (000370) - in Knowledge and Networks

368 Our methodology has limitations and drawbacks that one must consider when interpreting the final results. First, the degree to which the number of linkages observable in our data matches that in the real world heavily depends on the patent- ing practices among actors (e.g., cross-patenting or cases in which a central institu- tion may administrate the patenting process and is therefore the only applicant). Including only those collaborations that are defined by coapplication might under- estimate the number of actual linkages. Yet if we were also to take account of the connections realized through shared inventors, we might overestimate the number of linkages (Ter Wal & Boschma, 2009). In addition, we expect the number of dis- regarded cases to be rather small because inventor mobility is rare in Europe (Ter Wal & Boschma, 2009). Crescenzi, Gagliardi, and Percoco (2013) estimated that barely 5 % of inventors change their employer. A closely related drawback to our methodology is the underrepresentation of informal ties, for we considered only formal collaboration agreements. Prior studies have emphasized the importance that informal ties have for innovative outcomes (e.g., Powell & Grodal, 2006), but it has been found that formal ties, especially in the life sciences, are generally preceded by informal ties (Powell et al., 1996). On this basis we argue that preceding informal ties are manifest in formal ties and are therefore captured in the study of the latter. Second, by focusing on the research of the dynamics in bilateral R&D collabora- tion, we set aside the study of the effects of the micromechanism on the overall network structure. We thereby also opted to forgo explicit consideration of the feed- back effects that an actor’s position in the overall network has on partner choices at the microlevel. We tried to control for this limitation by incorporating information on whether an actor was highly connected (central) or rather peripheral and by adapting the standard errors accordingly. However, recent research on networks has made advances regarding the explicit modeling of endogenous structural mecha- nisms such as triadic closure and preferential attachment (Broekel et al., 2014). Our analysis could be extended by elaborating the overall network evolution as a result of partner choice at the microlevel, a selection that is itself determined by similarity and diversity aspects. Stochastic actor-oriented models, for instance, allow for examination of the relationship between the individual partner choice and overall network dynamics (Balland et al., 2013). In this context, however, it is debatable to what extent firms can directly influence and are aware of the network beyond their ego network (direct connections) (Gilsing et al., 2008). The third concern about studies that focus on analyzing a certain pattern in a specific industry is the generalizability of their results. Application of our results is limited, for example, by the appearance of patterns that might be caused by industry specificities. However, some of the factors that our analysis identifies (e.g., positive effects of overlap, the reciprocal cumulative advantage, and reciprocal general col- laboration experience) have also been observed in other environments and at other levels of observations (Cantner & Meder, 2007; Dahlander & McFarland, 2013; Gulati, 1999). In view of our results and the type of analysis suggested with this study, we have taken a further step in the effort to disentangle the coevolution of the proximity of collaboration partners and the formation and repetition of cooperative ties. In doing U. Cantner et al.