FIML estimation of an endogenous switching model for count data

This paper presents code for fitting a FIML endogenous switching Poisson count model for cross-sectional data in Stata 7: the espoisson command. The Poisson process depends on an unobserved heterogeneity term, xi; a set of explanatory variables, x; and an endogenous dummy, d. The endogenous dummy depends on an unobserved random term, nu. Correlation between xi and nu is allowed. If a model with exogenous d is fitted instead, correlation between xi and nu will result in simultaneous equation bias. The endogenous switching model corrects this problem. After describing the underlying econometric theory behind the command, an example is discussed.