Brokerage evolution in innovation contexts: Formal structure, network neighborhoods and knowledge

• Inventors’ collaborations (co-patenting) within organizations can be seen as networks that evolve over time. • A major structural element of networks is brokerage (that occurs when one inventor is collaborating with two disconnected inventors or alters ) and has been associated with innovation and creativity. • But brokerage is a fragile structure and the patterns its evolution matter for innovation outcomes. • Brokerage structures toward either disintermediation of broker or triadic closure (all three inventors connect directly over time). • Greater the hierarchical rank similarity among inventors, more the brokers are disintermediated. • Greater the indirect connectedness of alters in their own network neighborhoods, more the brokers are disintermediated. • Complementarity from non-overlap of inventors’ (broker and alters) knowledge bases results in closure. • The implication for R&D management is that it may be better served by allowing the invisible hand of self-assembling collaborations to operate freely, while limiting itself to setting ground rules for facilitating, incentivizing, and evaluating innovation outcomes. Research in a number of fields has shown that brokerage is typically fragile while creating consequential outcomes. However, little work has examined the conditions under which brokerage ends, and furthermore, whether and when it terminates with closure in a closed triad that includes the broker, or in a dyad that connects the previously-disconnected alters but disintermediates the broker. We employ a comprehensive theoretical framework drawing on constrained agency to study these questions in a context of organizational innovation. Specifically, we investigate the role of hierarchy, inventors’ network neighborhoods and knowledge differences in shaping the evolution of brokerage. We test our ideas in the a setting of co-patenting in 41 large Chinese research-intensive organizations over the period 1996-2008, with a dataset of 36,338 patents applied for by these organizations. We first show that the type of brokerage ending matters for innovation outcomes by demonstrating that disintermediation creates more subsequent innovativeness than closure. Thereafter, we use a two-step model to first model the termination of brokerage and in the second step to predict the type of closing: disintermediation or closure. Our results show that the broker's and alters’ hierarchical rank similarity promotes disintermediation, as does alters’ connectedness in network neighborhoods, while knowledge differences among the broker and alters encourage the evolution of brokerage toward closure. We spell out the implications of our findings for organizational innovation and the management of R&D.