University of Wisconsin - Madison J " 74 ' Institute for Research on Poverty

Data from the National Longitudinal Survey of Youth is used to empirically estimate the impact on the earnings capacities of young female dropouts if they were to return to complete either a regular high school education or a General Educational Development (GED) degree. To reduce the potential upward bias on these estimated rates of return, dropouts are allowed to have lower levels of innate ability and lower rates of human capital accumulation in school. After controlling for the sample selection bias associated with the observation of wage rates among only employed women, the rates of return for the average dropout are estimated to be 10.2 percent for a high school diploma and 6.2 percent for a GED degree. After allowing for self-selection in the decision of whether or not to complete a secondary education, these estimated rates of return actually rise slightly to 10.9 percent and 6.5 percent, respectively. Because of differences in other productivity characteristics, this education would only eliminate up to one-third of the substantial gap that already exists between the earnings capacities of dropouts and secondary school completers. ESTIMATING THE RETURNS TO A SECONDARY EDUCATION FOR FEMALE DROPOUTS There is renewed optimism in this country over the efficacy of general educational attainment in alleviating poverty.' Perhaps this is a by-product of recent concerns that the government has neglected basic education and that income and in-kind transfer programs have failed to substantially reduce poverty rates. The idea is that economic self-sufficiency can best be promoted by encouraging or requiring individuals who are at risk of welfare recipiency to increase their educational attainment. The Family Support Act of 1988 revised the national Aid to Families with Dependent Children (AFDC) program. To reduce long-term welfare dependence, states are now required to set up a Job Opportunities and Basic Skills (JOBS) program. Although states have a great deal of latitude in choosing the mix of services that will be provided to welfare recipients under their JOBS programs, some states have decided to include basic and remedial education in these packages. For example, a recent welfare reform proposal in Missouri would require AFDC recipients who have not completed their high school education and who are not exempt because of home responsibilities to work toward their high school equivalency or General Educational Development (GED) degree (Ashcroft, 1987). To make clear the potential benefits of this legislation to the majority of Missouri's AFDC recipients who are dropouts, proponents cited the substantially higher wage rates and family incomes of high school graduates in the general population. Unfortunately, there is little research to date that either confirms or refutes such claim^.^ We simply do not know how the inherent self-selection among female dropouts would affect their expected returns to a secondary education, or how these returns might vary by the type of high school credential obtained--a regular high school diploma versus a GED degree. The preferred approach would be to observe the change in potential market wage rates or earnings capacities as a result of the completion of a secondary education among randomly assigned AFDC recipients. Since no data are 2 currently available from such a controlled experiment, the next best approach is to econometrically model the process that leads to differences in earnings capacities among women with different levels of educational attainment who are at risk of welfare recipiency. For this reason, a sample of young women with no more than a high school education is taken from the National Longitudinal Survey of Youth (NLSY). Regression results are then used to estimate the potential rates of return to the average female dropout if she were either to complete a regular high school education or GED degree. Section I develops the general empirical approach used in this study. Section I1 describes the particular advantages of the NLSY data for the purposes of this analysis. Sections 111 through V formulate econometric procedures for estimating these rates of return and evaluate the empirical results. Section VI summarizes these findings and suggests future extensions to this study. I. A CONCEPTUAL FRAMEWORK A system of four general equations forms the basis for all of the empirical work in this paper.3 Years of schooling (S) completed by a woman are assumed to be a linear function of the exogenous variables contained in the vector (Z) and a disturbance term ( u ) . ~ These regressors include observable personal and family background characteristics. This reduced-form expression represents 3 the educational investment decision, where S is chosen so that the marginal rate of return is at least as great as the opportunity cost for her last year of schooling. Cognitive achievement or human capital (H) accumulated by the woman at the end of her schooling is a linear function of the same vector of regressors used in equation (I), educational attainment, and a disturbance term (v). The coefficient P, could be interpreted as the "value added" to her human capital from each year of schooling. The error term captures innate reasoning abilities that predate or are independent of educational attainment. Earnings capacity (W) is the wage rate facing the woman in the labor market after she has completed her schooling. It is written as a log-linear function of other proxies for her productivity (X), human capital, schooling, and a disturbance term ( E ) . Since differences in innate abilities may influence educational attainment, including the human capital variable in this equation should reduce any omitted-variable bias in estimating the returns to s~hooling.~ This equation also captures the two paths by which education ultimately affects the market wage: the indirect effect that occurs through the acquisition of human capital in school and its subsequent impact on earnings capacity dB,*?,); and the direct effect that comes from the potential signaling value of education (y,), because certain abilities may not be directly observable by potential employers (Spence, 1973).'j Since only those women who are employed and reporting a market wage rate can be included in the estimation of this wage equation, the possibility of sample selection bias must be considered. For this reason, we specify a simple linear function for the employment outcome. The latent employment propensity (E? depends on the observed determinants of her market wage and reservation wage (Q) (e-g., her marital status, the number and ages of children in her household). The problem is that unobserved factors that affect her earnings capacity may also affect her employment status. To produce unbiased coefficient estimates in the market wage equation, we must allow for the truncation of this error term.