Page - (000358) - in Knowledge and Networks

356 the portfolios of both partners, using the relative overlap as one measure of cogni- tive proximity (RelOverlap). We also included this measure as a quadratic term to capture the trade-off between minimum levels of knowledge overlap (as a warrant for mutual understanding) and maximum levels of overlap (as a hurdle that knowl- edge redundancy poses to innovation) (RelOverlap2). Reciprocal Potential Following Cantner and Meder (2007), we tested hypothesis 1b by operationalizing the potential knowledge benefits from a potential collaboration as the relation between partner A’s and partner B’s new knowledge that is brought to the collabora- tion. However, we extended the approach of that earlier study by differentiating the individual classes that were new to the partner rather than solely considering the absolute number of patents. We counted the number of nonoverlapping IPC classes for each actor and took the ratio between the minimum number and the maximum number of new knowledge classes. This measure is named ReciPot. It is a continu- ous variable that ranges between 0 and 1, taking a 1 when the amount of new knowl- edge that the one partner offers is equal to that of the other (perfect reciprocity). The greater the divergence between the amount of partner A’s and partner B’s nonover- lapping knowledge (i.e., the less reciprocal the gain is between the partners), the more the measure of potential benefit approaches zero. Knowledge Transfer To test hypothesis 1c, we needed to measure the knowledge transfer between col- laborators. Citations of previous documents (patents and publications) pertaining to the patent have become a favored instrument with which scientific authors detect knowledge spillovers (e.g., Griliches, 1990; Hall, Jaffe, & Trajtenberg, 2001; Jaffe, Trajtenberg, & Henderson, 1993; Mowery et al., 1996; Nelson, 2009; Nomaler & Verspagen, 2008; Schmoch, 1993; Singh, 2005). A frequent criticism, however, has been that patent citations may not imply real knowledge flows, for many citations are added by the patent examiner rather than the inventor or applicant. We took a different avenue and measured knowledge transfer between partners. To do so, we defined the vector of a firm’s patented technological classes as its cumulated knowledge stock and compared pre- and postcollaboration knowledge stocks. We defined knowledge transfer as the appearance of a new patent class in the firm’s patent portfolio after the collaboration had taken place (i.e., after the copatent had been filed).6 To attribute the portfolio changes to the cooperation, the newly added class had to have been part of the partner’s precollaboration knowledge base. This measure enabled us to differentiate pure knowledge-sharing (as the pure access to knowledge) from knowledge exchange (the integration of new knowledge into the firm’s own knowledge base). We assumed that if a class was subsequently 6 New in this context meant that the patent class did not appear in the firm’s precooperation portfo- lio before the application for the copatent. U. Cantner et al.