Seite - (000311) - in Knowledge and Networks

308 Regarding the variables capturing the structure of the within-city co-invention network, the coefficients of the fraction of inventors in the largest component together with that of the clustering coefficient (albeit not significant) seem to sug- gest that in cities characterized by dense cliques of collaborators, knowledge may flow relatively quickly within cliques, but it may also be highly redundant, which hampers recombination. Coming to the main variables of interest, external links prove to be an important mechanism to support the expansion and renewal of a city knowledge base (model 2). The IRR of ADWR is larger than 1 and statistically significant, even though the magnitude of the effect is not extremely large: Keeping all other variables constant, a standard deviation increase in the average external reach brings around 72 00068883 1010 1 100. % exp . .= ×( )−    ×( ) more new combinations in a city. By decomposing the impact of external reach, it turns out that the sheer number of gatekeepers (NUMGK ) is detrimental to recombination (model 3), even though the corresponding IRR is very close to 1. Nevertheless, this result is somewhat counterintuitive at a first glance, because gatekeepers are generally thought to be creative and imaginative actors. A possible interpretation, already anticipated in the previous sections of the paper, is that the control power these actors exert on the bridging ties and knowledge exchanges they govern can weaken indigenous capac- ity of recombination because they are in the position to grant, restrict, or even block the access to external knowledge flows. A further possible interpretation is that per- forming the gatekeeping function effectively requires skills and abilities that are not widely distributed in the population. Both interpretations find further support in model (14.4). The share of the overall external reach mediated by gatekeepers (SHREACH_GK) shows a negative and sta- tistically significant effect. All else equal, a standard deviation increase in this share is associated to a reduction of around 6 % in the number of new combinations intro- duced in a city. To probe further into the impact of gatekeepers, model (14.5) splits the overall external reach mediated by gatekeepers into its direct and indirect components. Results show that the negative effect associated with gatekeepers is due precisely to the indirect external reach mediated by gatekeepers (i.e., SHINDIR_GK), rather than to the gatekeepers’ direct external ties (i.e., SHDIR_GK). Overall, these results suggest that, all else equal, direct links to external sources of knowledge outperform indirect ones, namely those mediated by gatekeepers, in sustaining expansion and renewal of a city’s knowledge base. In other words, the key message stemming from these results is that direct links, regardless whether developed by gatekeepers or other actors in the network, are far more reliable and fruitful in spurring technologi- cal recombination in cities. Model (14.6) introduces into the regression the variable that captures the fraction of the indirect external reach accounted for by the top 10 % of gatekeepers (i.e., GKREMOV). The rather surprising result is that the IRR of this variable is larger than 1 and statistically significant, thereby indicating that a greater concentration of S. Breschi and C. Lenzi