Sample Selection Bias As a Specification Error (with an Application to the Estimation of Labor Supply Functions)

Abstract

Moreover, in each case, if it were possible to obtain the missing wage data for either the treatment or nontreatment population, it would be possible to utilize simple regression techniques to estimate the parameters of population functions.Simple comparisons between pre and post treatment wages would yield unbiased estimates of the economic benefits of the treatment.Sample selection bias may also arise as a direct consequence of actions taken by the analyst.In studies of panel data, it is common to require that "intact" observations be employed.For example, in analyses of the time series of the labor supply of married women, stability of the family unit is often required.foran observation sequence to be analyzed.The effect of such criteria operates in precisely the same fashion as self selection: fitted functions confound behavioral functions with sample selection functions.It is fair to say that most competent analysts have been aware of the possibility of both sources of selection bias.It is also fair to say that the accepted econometric practice has been to ignore the problem in making parameter estimates but to verbally qualify the estimates in light of possible selection biases.Recent work in econometrics has attempted to improve on previous work by making specific assumptions about the source of selection bias.In particular, this work assumes that both the missing data and the available data are drawn from a common probability distribution, typically assumed to be a normal law.Except for work by Amemiya (1973) and Gronau (1974), the authors of /these studies rely on maximum likelihood estimators to produce parameter estimates free of selection bias.In this paper, I present a simple characterization of the sample also selection bias problem that is/applicable to the conceptually distinct