Dynamic models of the labor force behavior of married women which can be estimated using limited amounts of past information

This paper examines the extent of the inadequacy of standard cross-sectional models of US labor force behavior and considers the abilities of alternative models to capture the observed continuity in the hours of work and earnings of individuals as well as in their employment histories. Both of the alternatives to the standard cross-sectional model considered in this study incorporate limited amounts of information about past work behavior that could easily be collected as part of a national population census. Using a population of 21 to 64 year old married working women taken from a 1969 through 1978 Michigan Panel Study of Income Dynamics, the variables included in the Z vector age: 1) age of the wife; 2) education of the wife; 3) state average hourly wage in manufacturing measured in 1967 dollars; and 4) unemployment rate for the state in which the wife lives. Results show that by using information about a women's hours of work and wage rate in the previous year, it may be feasible to improve on forecasts of a woman's employment and earnings behavior. For each model a separate estimate is made for wives aged 21 through 46, and for those aged 47 through 64. The dummy and difference models perform much better than the standard model, with the dummy model having the higher pseudo-chi-square statistic. These models show that systematic errors made in determining which individuals work, what they earn per hour, or how many hours they work, should result in prediction errors of the same sort year after year in the computation of annual earnings. These findings with respect to years of work and nonwork, years of part time versus full time work, and cumulative earnings over a 10 year period, confirm and extend Heckman's findings; thus, forecasting models of the work behavior of individuals should not be estimated using pure cross-sectional data. It would be important if researchers could identify what observable factors, if any, increase the likelihood that wives will alter their work behavior from what it has been in the immediate past, even if they are not able to fully understand or explain this previous work behavior.