Non-normality as Distributional Misspecification in Single-Equation.

This paper provides a reevaluation of often-used limited dependent variable models. It shows that parameter estimators based on such models may be seriously biased and inconsistent if the true d istribution of the disturbance term deviates from normality. The degr ee of difference from normality is controlled in this paper by varyin g the skewness and kurtosis of the true distribution-these are distri butional characteristics that an empirical researcher can easily chec k for himself. This paper proposes to alert users of limited dependen t variable models to the potential problems of these models in the fa ce of nonnormality. Copyright 1987 by Blackwell Publishing Ltd