Married women's labor supply: a comparison of alternative estimation procedures

A methodological and empirical comparison of alternative methods of estimating married women's labor supply parameters. Four procedures are examined in methodological comparisons: ordinary least squares estimated on sub-sample of workers; Tobit analysis estimated on sample of workers and nonworkers using OLS predicted wage as right-hand-side variable; grouped-data procedure suggested by Gronau and maximum likelihood approach developed by Heckman. The first three procedures are characterized by implicit restrictions placed on the covariance matrix of disturbances in wage and hours worked equations. Expected biases in estimated parameters if restrictions fail to hold in the data are explored. Empirical comparisons of estimated wage and hours worked parameters between three of the four empirical procedures (the exception being Gronau's) are made using data from 1967 National Longitudinal Survey of Mature Women. Comparisons reveal large and systematic differences between estimates of three procedures. These differences are in accordance with expectations of biases which would result from improper imposition of disturbance term restrictions.