A Comparison of Two-Stage Estimators of Censored Regression Models

This paper presents a Monte Carlo comparison of the small-sample performance of subsample ordinary least squares, the Heckman-Lee two-stage estimator, and the robust estimator of Lee. Each estimator is considered under bivariate normal, t, and chi-square error structures. The estimates indicate that the Heckman-Lee and Lee estimators do not provide an unequivocal mean square error improvement upon subsample ordinary least squares in small samples. While effectively controlling for selectivity bias, the two-stage estimators suffer a substantial loss of small-sample precision relative to subsample ordinary least squares. Copyright 1991 by MIT Press.