Endogenous Social Interactions with Unobserved Networks

We present a model of endogenous network formation to recover unobserved
    social networks using only observable outcomes. We propose a novel
    equilibrium concept that allows for a sharp characterization of equilibrium
    behavior and that yields a unique prediction under testable conditions.
    While the equilibrium is characterized by a large number of nonlinear
    equations, we show that it can be efficiently employed to recover
    the networks by an appropriately designed approximate Bayesian computation
    method. We apply the model to recover the network of social links
    between lawmakers in the U.S. Congress using data from the 109th to
    113th legislatures. We show that social connections are important
    for legislators' productivities and we identify some of the key determinants
    of network centralities in the U.S. Congress.