Forming a Family, Leaving School Early, and Earning a GED: A Racial and Cohort Comparison

Abstract

One of the most enduring controversies in family demography concerns the effect of early family formation--becoming a parent or getting married--on educational attainment. A positive association between age at family formation and educational attainment is well documented (Bartz & Nye, 1970; Bloom & Trussell, 1984; Call & Otto, 1977; Kerckhoff & Parrow, 1979; Nye & Bernardo, 1973; Rindfuss, Morgan, & Swicegood, 1988; Wilkie, 1981). It is undisputed that young women delay family formation in order to achieve high levels of education (Teachman, Polonko, & Scanzoni, 1987). What scholars cannot agree about is the existence of a reciprocal effect--that is, whether or not women who form their families early earn less education than they otherwise might (Alexander & Reilly, 1981; Hofferth & Moore, 1979; Marini, 1978, 1984; Rindfuss, Bumpass, & St. John, 1980). Previous tests of the hypothesis that early family formation causes educational deficits are plagued by three problems of specification. SPECIFICATION PROBLEMS IN PAST RESEARCH Years of Completed Schooling First, scholars have often looked at the influence of age at first family formation on years of completed schooling, rather than on reaching educational milestones either positive (e.g., postsecondary school attendance) or negative (e.g., leaving school prematurely). The latter approach is superior for two reasons. First, it accurately reflects the way educational attainment is organized and rewarded in America--by the acquisition of particular credentials like a high school equivalency degree (GED) or college degree, rather than incrementally by years of schooling. Second, American schools are highly age graded (Hogan & Astone, 1985). Thus, a focus on one educational outcome at a time will allow us to test whether family formation interferes with educational attainment at one point in life, but not another. The possibility of such variation across the life course has been proposed as an explanation of conflicting empirical results (Hofferth, 1984). Two-Stage Least Squares A second problem of past research is the use of two-stage least squares estimation techniques to test for reciprocal causality. These techniques were developed for use with variables such as attitudes, which can change over time for an individual and thus be mutually adjusting. They are not suitable when the variables in question are events--like leaving school or getting married--that can only happen once and are invariant for an individual (Alexander & Reilly, 1981). A better approach is to look at whether family formation comes before or after the exit from school (Upchurch & McCarthy, 1990). The assumption underlying this strategy is that an event that occurs after another cannot be said to have caused it. This assumption is not without problems, since people may adjust their behavior in anticipation of events (Marini, 1984). When the focus is on secondary school attainment the assumption is probably justified, since family formation during adolescence is often sudden and unplanned (O'Connell & Rogers, 1984; Zabin, Astone, & Emerson, 1993). Marriage or Motherhood? Third, past studies usually focused on the effect of age at first birth, or the effect of age at first marriage, on educational attainment, but not on both together. The difficulties of separating the effects of variables that are as highly correlated as these two makes such a choice understandable. Even in theoretical formulations, however, the two are usually treated separately. In studies of educational attainment where the dependent outcome is years of schooling completed over the life course, the ordering of the two events is often assumed to be first marriage, then motherhood. In one study, for example, age at first marriage constituted the lower bound on age at first birth (Marini, 1978). By contrast, other studies restrict attention to the effects of adolescent childbearing on educational attainment, ignoring the effects of marital status (see Astone, 1993, for a discussion of this issue). …