Page - (000304) - in Knowledge and Networks

301 Third, the construction of the dependent variable is based on patents in which only inventors located in the focal MSA are reported in the document. The exclu- sion of new combinations that are the outcome of cross-city collaborative patents provides a more restrictive measure of a city’s autonomous recombinatorial and inventive capabilities and should mitigate possible endogeneity concerns with refer- ence to the main independent variables. In addition to the variables described in the previous section, the empirical model includes other factors that may potentially affect the renewal and expansion of a city’s knowledge base.8 In order to account for the importance of agglomeration economies (see Duranton and Puga [2004] for a review), it includes two variables. First, the number of internal patents (INTPAT) in the city at time t (i.e., by excluding patents with inventors external to the city) captures both the scale effect associated with the agglomeration of inventive activities at the city level, as suggested by Bettencourt et al. (2007) and Lobo and Strumsky (2008), and the potential for tech- nological recombination. Second, the degree of concentration of inventive activities among firms, computed as the Herfindahl index at the level of patent assignees (HFIRMS), accounts for (possible) effects of local market structure and competition (Beaudry & Schiffauerova, 2009). A further set of variables controls for the nature of the local knowledge base. First, an index of absolute specialization, namely the Herfindahl index of the share of patents made in IPC four-digit (i.e., subclass) technology fields (HPATENTS), captures to what extent a city is specialized in a narrow set of fields, thereby control- ling for the presence of externalities arising from technological specialization (Feldman & Audretsch, 1999).9 Second, the average number of citations of the non- patent literature made by a city’s internal patents (NPLCIT) measures the scientific orientation and generality of the local knowledge base. It is not a simple matter to anticipate the sign of this variable. More science-oriented knowledge can have more difficulties in finding its immediate technological application, leading to a negative effect of this variable on the recombinatorial capabilities of a city. On the other hand, more universal notions are more likely to find multiple applications at the junction of different technological domains (Fleming & Sorenson, 2004), meaning that a more general and science-based knowledge base could open greater possibili- ties of technological recombination. Third, because the Herfindahl index captures the absolute technological specialization of a city, the model also includes an index of relative technological specialization measuring the dissimilarity between the technological profile of city c and that of all the other cities with which inventors of city c have collaborative linkages, namely the so-called Krugman index (KI), at the level of technology groups in which a city’s patents are classified. For each city c a time t, KI is defined as follows: 8 The general logic of the empirical model is based on and expands Breschi and Lenzi (2015). 9 This variable is computed at the technology subclass (i.e., 4-digit IPC) level and not at the tech- nology group (i.e., the lowest technological aggregation level) level, because the use of groups would disproportionately and artificially inflate its value. 14 Are Gatekeepers Important for the Renewal of the Local Knowledge Base?…