An empirical investigation of parametric and semiparametric estimation methods in sample selection models

In this paper we analyze empirically different specifications of a sample selection model. We are interested in how the estimates vary across alternative assumptions concerning the joint conditional distribution of the sample selection equation errors, such us the specification of error distribution, the functional relationship of the index function and heteroskedasticity. To do this, we estimate a wage equation for the Spanish labor market using two different approaches: Maximum Likelihood and Two-Step Methods. For the latter, three alternative semiparametric procedures are used to compute the sample selection mechanism, and thus three alternative two-step estimators of the parameters of the wage equation are obtained. We compare theses estimates with Heckman's approach.