Page - (000363) - in Knowledge and Networks

361 that of gathering the empirical distribution by repeated random permutation of the complete adjacency matrix—an approach known as multiple regression quadratic assignment procedures (MRQAP). Although this method has proven to be appropri- ate for linear models with a continuous dependent variable, it is still unclear how it performs when employed to analyze binary models (Broekel, Balland, Burger, & van Oort, 2014; Dekker, Krackhardt, & Snijders, 2007). Besides, MRQAP has not been tested much in panel settings. Results Descriptives Diversity in Partner Portfolio For an initial overview of the diversity of the firms’ partner portfolios, we consid- ered the number of different partners firms cooperated with in the years from 1978 to 2010. Table 16.1 contains summary statistics about the number of partners and the continuity of links. As shown by the distribution of actors across the different partners (see Fig. 16.3), most firms cooperated with two different partners, the median being 2. Only a few firms cooperated with a larger variety of actors. The maximum number of different partners in one portfolio was 17. In other words, one firm cooperated with 17 different actors during the period under study. For the firms in our sample, the implication was that repeated collaboration with only one partner was not a dominant behavior. Dynamics of Link Formation Concerning the recurrence of links, we found that 138 of the 293 realized links came about just once (nonrecurring), whereas 60 links were repeated at least once (the sum of repetitive links was 155). Without double-counting the repeated links, Fig. 16.2 The model of the cooperation decision that is estimated to explain cooperation by the presented explanatory variables 16 Coevolution of Innovative Ties, Proximity, and Competencies