We analyze knowledge spillovers in a search-theoretic spatial equilibrium framework with workers who are heterogeneous in knowledge type. Knowledge spillovers result from random face-to-face interactions between workers in the city. The outcome of those interactions crucially depends on the combination of the interacting individuals' knowledge types. In contrast to previous work, we explicitly model knowledge spillovers as the interplay of two channels: knowledge transmission (imitation) and knowledge creation (innovation).

Our results show that if the role of innovation is sufficiently important, individuals choose an excessively narrow range of partners to interact with, leading to lower than socially optimal creation of new ideas, which results in socially inefficient city sizes.