The Heckman Correction for Sample Selection and Its Critique

This paper gives a short overview of Monte Carlo studies on the usefulness of Heckman’s (1976, 1979) two‐step estimator for estimating selection models. Such models occur frequently in empirical work, especially in microeconometrics when estimating wage equations or consumer expenditures. It is shown that exploratory work to check for collinearity problems is strongly recommended before deciding on which estimator to apply. In the absence of collinearity problems, the full‐information maximum likelihood estimator is preferable to the limited‐information two‐step method of Heckman, although the latter also gives reasonable results. If, however, collinearity problems prevail, subsample OLS (or the Two‐Part Model) is the most robust amongst the simple‐to‐calculate estimators.