Semiparametric Estimation Of Selection Models: Some Empiric

In order to provide a practical application of a semiparametric estimator of selectivity bias, semiparametric methods are used to reanalyze data on the labor supply of married women first studied by Mroz (1987). The data set Mroz analyzed consists of measurements on characteristics of 753 married women, of whom 428 were working at the time of the study. Annual hours of work is the dependent variable, with regressors including a logarithm of the wage rate, family income less the wife's labor income, indicators for young and older children in the family, and the wife's age and education. Application of the semiparametric estimation methods to this data set give results similar to parametric methods assuming normal errors, but provide somewhat weaker evidence of the presence of the selectivity bias. These results support the position that specification of the regression function and set of instrumental variables appear to be more important than specification of the error distribution for these data.