Detecting Social Interactions in Bivariate Probit Models with an Endogenous Dummy Variable

Some Simulation Results

Abstract

This paper analyzes the possibility of detecting observable and non-observable social interactions in a bivariate probit model with an endogenous dummy regressor via Monte Carlo simulation. In small samples, we find severe size distortions of the Wald test and only a low probability of detecting observable and non-observable social interactions. In large samples, however, we find this test to be very powerful.