Maximum likelihood estimation of social interaction effects with nonrandom group selection

This paper derives a maximum likelihood estimator for an econometric model of discrete choice with social interaction effects. Endogenous selection of reference group is addressed within the econometric model through the incorporation of a reduced form within-group correlation in both observed and unobserved characteristics. The estimator requires only standard survey data which provides information on a binary choice made by the respondent, a vector of the respondent’s background characteristics, and the average choice made by a large reference group (for example, school or census tract). Properties of the estimator are demonstrated analytically and through Monte Carlo experiments.