Semiparametric Estimation of a Sample Selection Model: A Simulation Study
Published March 1997
Standard approaches to the estimation of sample selection models are known to be inconsistent under non-normality. In particular, this paper considers the two-step Heckman (1976, 1979) estimator of the interecept of the outcome equation. This estimator is compared with a consistent asymptotically normal semiparametric estimator suggested by Andrews and Schafgans (1996). Using a root mean squared error criterion, the semiparametric estimator performs better for a range of bandwidth parameter choice for a variety of distributions of the errors and regressors. For error distributions that are close to the normal, however, the two-step parametric estimator performs better.