Selection Bias Corrections Based on the Multinomial Logit Model : Monte-Carlo Comparisons

This survey presents the set of methods available in the literature on selection bias correction, when selection is specified as a multinomial logit model. It provides a theoretical exposition of underlying assumptions made by the methods and shows results from a set of Monte-Carlo experiments. We find that, in most cases, the approach initiated by Dubin and MacFadden (1984) is to be preferred to the more commonly used Lee (1984) method, as well as to the semi-parametric alternative method recently proposed by Dahl (2002), even in the presence of high non-linearity in the selection term. Monte-Carlo experiments also show that selection bias correction based on the multinomial logit model generally provides fairly good correction for the outcome equation, even when the IIA hypothesis is violated