Association for Public Policy Analysis & Management

Drawing upon multiple panel datasets from the U.S., we document a startling empirical pattern: the earnings premium from a four-year college degree (relative to a high school diploma) for persons from low-income backgrounds is only half as large as it is for those from more-fortunate backgrounds. For individuals whose family income in high school was above 1.85 times the poverty level,  we estimate that career earnings for bachelor’s graduates are 135 percent higher than earnings for those whose education stopped at high school. However, for individuals whose family income during high school was below 1.85 times the poverty level, the career earnings of bachelor’s graduates is only 69 percent higher than that of high school graduates. This reduced earnings premium amounts to a career loss of over $300,000 in present discounted value. Furthermore, this relative disadvantage for persons from low-income backgrounds increases over the career. For example, bachelor-degree holders from low-income backgrounds start their careers earning about two-thirds as much as bachelor degree holders from higher-income backgrounds, but this ratio declines to one-half by mid-career. In contrast, persons from low-income backgrounds who did not earn a post-secondary credential earn about 80 percent as much in early career as persons from a higher-income background, and this ratio declines more modestly to 70 percent by mid-career.

This stylized fact suggests a significant shortcoming in strategies that simply promote more education as a way to alleviate poverty or reduce inequality. It also suggests a possible mechanism for how persistent poverty and increased economic inequality can retard economic growth: they may reduce the returns to educational investments.

In this paper, we establish the prevalence and robustness of these differential returns to education in three, restricted-use panel datasets: the Panel Study of Income Dynamics, and the National Longitudinal Studies of Youth, 1979 and 1997 cohorts. We also examine more thoroughly and flexibly the relationship between family income during adolescence and the observed career earnings premia to higher education, and how this varies across demographic groups and birth cohorts. Taking advantage of the rich detail over the lifecycle in our data, we employ modern semiparametric decomposition techniques to explain the observed differences in education premia across three categories of factors: (a) family background (race, gender, region of country during childhood, type of neighborhood during adolescence, family structure and parental characteristics during adolescence, and both academic and socioemotional test scores at the end of high school); (b) higher education (the type and quality of college attended, college major, and postgraduate education); and (c) labor market (occupation and industry mix, geographic location, the likelihood of employment and hours worked, and residual hourly wage differences). These decompositions help inform us about the mechanisms behind growing wage inequality within levels of educational attainment, which has been an open research question for some time. Moreover, they can offer guidance to construct more-effective education policies.