Page - (000306) - in Knowledge and Networks

303 sourced knowledge; it has to be expected to have a negative effect on the recombi- natorial capacity of a city (Uzzi & Spiro, 2005). All the explanatory variables (except the number of internal patents) are 1-year lagged over a 5-year moving window (i.e., computed over the period [ ,t t− −1 5 ]) with respect to the dependent variable to mitigate endogeneity concerns. The defini- tion of variables is summarized in Table 14.1. Table 14.2 reports summary statistics, while Table 14.3 reports the correlation matrix.12 The empirical model estimated is based on a conditional negative binomial framework with fixed-effects by controlling for annual effects, which allows accounting for the integer and over-dispersed nature of the dependent variable, as was done in similar studies (Fleming, King, & Juda, 2007; Fleming et al., 2007 Schilling & Phelps, 2007). 12 The correlation matrix reported some relatively high correlation coefficients. To test for risks of multicollinearity, we did run the collin Stata command on all variables included in the regression (with the exclusion of SHDIR_GK and SHINDIRGK, because the sum of the two was simply equal to SHREACH_GK). The variance inflation ratio (VIF) tended to exclude serious risks of multicol- linearity because it was below 2.5 for most of the variables with the exception of the (log) of internal patents (6.95), the number of gatekeepers (5.96), and the Krugman index (3.73). However, even for these variables, the VIF was well below 10, which is the rule of thumb value usually con- sidered in order to detect serious problems of multicollinearity. The average VIF was slightly above 2.5 (2.59). Table 14.1 Definition of dependent and independent variables Variables Definition Number of new technological combinations in a city at time t (dependent variable) INTPAT Number of internal patents made in a city at time t (in logs) HFIRMS Concentration (Herfindahl) at the level of patent assignees in a city HPATENTS Concentration (Herfindahl) at the level of IPC 4-digits in a city NPLCIT Average number of citations to NPL of a city KI Krugman index (at the level of IPC groups) of a city LARGE Percentage of inventors in the largest component of the co-invention network within a city CLUST Clustering coefficient of the co-invention network within a city ADWR Average distance-weighted external reach of a city NUMGK Number of gatekeepers in a city SHREACH GK_ Share of external reach mediated by gatekeepers in a city SHINDIR GK_ Share of external reach indirectly mediated by gatekeepers in a city SHDIR GK_ Share of external reach directly mediated by gatekeepers in a city GKREMOV Share of the external reach mediated by gatekeepers accounted for by the top 10 % gatekeepers in a city CONCGK Share of the external reach mediated by gatekeepers accounted for by the top 4 firms in a city Note. All independent variables are computed in the period ( t t− −1 5, ), with the exception of INTPAT which is measured at time t. Source: Elaboration by authors 14 Are Gatekeepers Important for the Renewal of the Local Knowledge Base?…