A Monte Carlo Analysis of Alternative Estimators in Models Involving Selectivity

In a simultaneous-equation model involving selectivity, Monte Carlo and response-surface techniques are used to assess the performance of five estimators commonly applied to a behavioral equation conditioned on an endogenous binary selectivity decision. The estimators include least squares with an exogenous dummy variable for the selectivity decision, three two-stage estimators that employ the estimated probability of the selectivity decision, and full information maximum likelihood (FIML). Although formally inconsistent, least squares with dummy variables is found to perform nearly as well as FIML, based on mean squared error measures. All two-stage estimators are found to be seriously deficient in terms of robustness.