Page - (000364) - in Knowledge and Networks

362 we found 198 realized combinations, of which most (138, or 70 %) were nonrecur- ring. Most (41) of the sustainable links were repeated only once, and the maximum number of link repetitions was 6. Unlike the findings reported by Gulati (1995) and Gulati and Gargiulo (1999), who found stability in link formation, our first findings suggest that firms are inclined to change partners regularly rather than repeat col- laboration with the same partner. Our findings complement the results by Wuyts et al. (2005) and Cantner and Graf (2006), which support the contention that the search for diversity of knowledge sources tends to lead firms to switch their R&D partners. Estimation Results Table 16.3 shows the bilateral correlations between the variables included in the estimations. With regard to correlations between the explanatory variables, we do not seem to have a severe problem of colinearity. With respect to the correlation between the explanatory variables and the dependent variable (Coop), we find that RelOverlap, TransKnowledge, CoopExp, DyadSingle-PAT5, DyadCoopPAT5, and DStatus have a weak positive correlation with cooperation, whereas ReciPOT, DCentrality, and DPatAge are negatively correlated. To deepen our understanding of the forces that determine the partner choice, we ran a random-effects logistic regression on our panel data. Table 16.4 shows the outcome of our estimations for seven model variations. The results for the base model, which comprises the two control variables, DStatus and DPatAge, are shown in the last column. We found that DStatus was highly significant and positively cor- related to the probability to cooperate (Coop), indicating that firms prefer to cooper- ate with partners that are of a different organizational form. Concerning the dynamics of cognitive proximity, we analyzed three dimensions: overlap (RelOverlap), reciprocal potential, and knowledge transfer. First, we found 1 2 3 4 5 Number of Partners 6 7 12 17 Fig. 16.3 Diversity of the partner portfolio among firms in the sample (Design by authors) U. Cantner et al.