ABSTRACT
The central tenet of the human capital explanation of sex differences in wages is that, because of specialization in household production, women intend to work in the labor market more intermittently than men, and therefore invest less. Because much human capital investment is unobserved, it is argued that this lower investment leads to lower wages, even after controlling for observable variables that affect wages (Mincer and Polachek, 1974; Polachek, 1975; Becker, 1985). Numerous studies show that women accumulate less actual experience and job tenure than men (e.g. O’Neill, 1985). Moreover, the evidence originally presented by Mincer and Polachek (1974) is consistent with the human capital explanation of sex differences in wages. In log wage regressions the coefficients on experience tend to be higher for never married women and women without children than for women with children. Other researchers have presented related supporting evidence (e.g. Sandell and Shapiro, 1980), including evidence partially consistent with the human capital model’s hypotheses regarding the effects of household specialization on men’s and women’s wages (Korenman and Neumark, 1991, 1992). On the other hand, empirical evidence against the human capital explanation has also been garnered, focusing on estimation of the returns to experience (Sandell and Shapiro, 1978; Blau and Ferber, 1992), and on the role of depreciation of human capital during labor market withdrawals (Corcoran, 1979). An alternative “feedback” hypothesis that is consistent with these findings is
that women experience labor market discrimination-partly reflected in lower wage levels and wage growth-and respond with career interruptions associated with specialization in household production. These interruptionsmay in turn lower subsequent wage growth. The competing explanations of sex differences in wages have been tested in recent work by Gronau (1988) and Blau and Ferber (1991), and in previous work surveyed in Blau and Ferber (1992, chapter 7). The papers by Gronau, and by Blau and Ferber focus explicitly on distinguishing between the human capital and feedback hypotheses, by addressing the joint causality running from women’s labor market intermittency to lower wages, and from lower wages (presumably attributable in part to discrimination) to intermittency. Both papers do this by invoking sufficient identifying assumptions to estimate
simultaneous equations models with wages and future labor market separations as the jointly dependent variables. Gronau’s results with respect to the human capital model are ambiguous. He finds evidence that future separations lower current wages, once account is taken of the effect of such separations on on-the-job training.1 But he does not find evidence that future separations reduce the skill requirements of women’s jobs. Furthermore, consistent with the feedback hypothesis, he finds that lower wages lead to labor market separations. Blau and Ferber tend to find no evidence supporting either the human capital model or the feedback hypothesis, with no effect of expected years of full-time work on expected earnings growth (conditional on continuous employment), nor of expected earnings growth on planned years of work. However, the validity of the findings in these papers hinges in large part on identifying assumptions in the form of exclusion restrictions, which are somewhat arbitrary (as Gronau acknowledges). The goal of this chapter is to pursue an alternative empirical approach to studying
the feedback hypothesis. Rather than attempting to untangle the joint determination of wages and labor market interruptions, this chapter looks at the relationship between women’s self-reports of sex discrimination on the job and future labor market outcomes, including labor market interruptions, employer changes, demographic changes, and wage growth. A variety of approaches to estimating these relationships are considered, to account for numerous possible biases arising from the use of self-reported discrimination data. Thus, this approach differs in two ways from the existing research. First, since
reported discrimination is presumably not a choice variable of the individual (in contrast to wage levels or growth rates, which, according to the human capital hypothesis, are partly choice variables), there is not an inherent simultaneity problem. That is, self-reported discrimination may provide a more exogenous measure of discrimination experienced by women than does the wage. Second, in contrast to previous research, this approach focuses on the effects of discrimination per se, in contrast to the effects of wages paid. Thus, it is potentially a more direct test of the hypothesis that sex discrimination contributes to women’s labor market interruptions. While these are advantages of the approach, the disadvantage is that self-reports of discrimination are subjective measures. We view it as a goal for future research to further assess the validity of such subjective measures, and to better integrate the use of objective and subjective measures in the study of discrimination.
