T HE stereotypical view of female employees is that they have relatively weak job attachment and that, in particular, they are especially prone to voluntary job separations. Although this notion is borne out by overall sex differences in aggregative quit rates, this evidence is at best only suggestive since it does not distinguish sex-specific differences in quit behavior from other factors, such as differences in job characteristics and wage rates.' The principal study to date of sex differences in worker quitting is that of Barnes and Jones (1974), who analyzed differences in aggregative quit rates by sex. Although their findings were consistent with the view that females are more prone to quitting, the analysis was restricted to observations for only 19 two-digit industries for each sex so that there was not sufficient information in the sample to analyze many important patterns of interest.2 Quit rate studies that do not focus specifically on female quit behavior typically have included a variable reflecting the percentage of workers of a particular sex in the industry. While industries with larger percentages of female employees generally have been associated with higher levels of quitting,3 these findings for samples of 47-52 two-digit industries are somewhat different from those found in other samples. Indeed, analysis of 95 3-digit industries by Viscusi (1979) reveals no significant sex effect on aggregative quit behavior. In this paper, I will utilize data for a large sample of individuals in an attempt to resolve the ambiguities in these earlier findings. The most familiar economic motivation underlying potential male-female quit differences is that women often leave the labor market to bear and raise children. Moreover, since wives typically earn lower wage rates than do their spouses, they may serve as secondary earners, entering the labor force during periods of temporary economic needs and exiting thereafter. In addition, family migration decisions, such as those analyzed by Mincer (1978), may lead to quits by wives whose husbands have been transferred to new locales. There also may be important differences in the lifetime employment choice pattern related to the role of quitting as part of an adaptive choice process.4 To the extent that women have less precise notions of their prospects for advancement and their working conditions, such as the presence of co-worker discrimination, they will be more likely to use the initial period of employment as a period of experimentation and then quit if their experiences are sufficiently unfavorable. An offsetting influence is the fact that males have a greater expected future period of work so that learninginduced quit behavior may offer greater potential gains even though the informational content of the on-the-job experiences may be less.5 Finally, in situations in which workers are unable to voice their complaints effectively and have them settled through grievance procedures, they will adopt an alternative economic response of exiting from the undesirable job context.6 CoReceived for publication October 23, 1978. Revision accepted for publication July 25, 1979. * Northwestern University and Council on Wage and Price Stability. Helpful comments were provided by Gregory M. Duncan, an anonymous referee, and members of the Northwestern Labor Seminar. John Link performed his usual excellent job as programmer for this research. I The importance of quit behavior to analyses of sex differences in employment and the inconclusive nature of existing studies is discussed by Reynolds (1978), especially on page 167, and by Pigors and Myers (1973). 2 Their principal regressions included only two age variables and an industry wage variable. Inclusion of a worker education variable knocked out the wage effect for males. See footnote 16 on page 447 of Barnes and Jones (1974). 3 See, for example, Burton (1969), Burton and Parker (1969), Parsons (1972), Pencavel (1970), and Stoikov and Raimon (1968) for aggregative results of this type. The signs for the worker sex variable are sometimes mixed or statistically insignificant.

H UMAN capital theory posits that individuals invest in the acquisition of productive skills in order to derive higher future earnings. While the dichotomy between investments in and skills has long been recognised,t empirical tests of the theory do not usually distinguish between the two earnings components.2 In this paper the conventional earnings model is expanded to incorporate worker-financed specific training. The inclusion of tenure allows a segregation of estimates of returns to general and specific human capital. This disaggregation permits an examination of the relationship between worker-financing of specific training and interindustry wage differentials. Cross-section investigations typically report the importance of industry of employment as a wage determinant, even controlling for a myriad of individual characteristics.3 This result has been taken, perhaps prematurely, as evidence of pervasive labor market imperfections. If interindustry differences in skill specificity are important, we should expect systematic wage differentials. Increased acquisition of worker-financed specific training will be associated with lower initial wages and subsequent higher rates of wage increase. In section I we consider the implications of using alternative specifications of the earnings model. The distinction is drawn between general and specific human capital, allowing a relaxation of the assumption constraining the returns from these mix of skills to be equal across individuals. In the process of this exposition the specific training prediction of a negative relationship between industry standing wages and subsequent rates of wage growth is derived. This hypothesis is subjected to test using individual data from the 1971 National Longitudinal Survey of Young Men, and the results are reported in section II. The complication is raised that workers remain longer in jobs paying higher wages independently of specific training. A simultaneous equations model treating tenure and wage as endogenous variables is tested revealing little bias in the single equation specification. A concluding section summarizes the main findings and offers suggestions for future research in this area.