On the Use of Distributional Mis-specification Checks in Limited Dependent Variable Models.

A unified approach to testing for distributional misspecification in parametric limited dependent variable models is described and discussed. The approach involves examining the significance of certain moment-type expressions. A form of the test statistic may be obtained as nR(superscript)2 from a simple least squares regression. The procedure is sufficiently general to subsume several tests already described in the literature for particular limited dependent variable models, but to have a wider applicability. Two practical examples, binary choice and duration analysis, are provided both to motivate the procedure and to illustrate the implementation of the methodology. Copyright 1989 by Royal Economic Society.