Urban social structure, social capital and spatial proximity

We develop a theoretical model where the existence and intensity of dyadic contacts depend on location. We show that agents tend to interact more with agents that are highly central in the network of social contacts and that are geographically closer. Using a unique geo-coded dataset of friendship networks in the United States, we find evidence consistent with this model. The main empirical challenge, which is the possible endogenous network formation, is tackled by employing a Bayesian methodology that allows to estimate simultaneously network formation and intensity of network contacts.