Over-education and wages revisited: A two cohort comparison and random coefficients approach

We examine the effect of over-education on wages using both the 1979 and 1997 National Longitudinal Survey of Youth panels and expand the education mismatch literature in two notable ways. First, we determine whether the effect of education mismatch on earnings has changed over time by comparing the two NLSY cohorts of similarly aged individuals that earned wages roughly two decades apart. We find that the incidence of over-education has increased over time in the US, but its positive effect on earnings has declined dramatically. Over-education is insignificantly correlated with wages in the 1997 cohort, but is typically twice the size and statistically significant in the 1979 cohort. After accounting for the proportional increase over time in part-time workers in the labor market, we find that over-education’s positive effect on wages is evident only among full-time workers and in both cohorts. Yet, for full-time workers we document an even greater decline in the returns to over-education over time, as compared to our full sample estimates. We present these findings as a cautionary tale for students deciding whether to invest in education levels above what is typical for a particular occupation.

Second, we use an econometric technique novel to the literature that allows over-education to have a disparate impact on wages across individuals. Accordingly, we test whether these random slopes models yield significantly different and improved estimates compared to the baseline ordinary least squares (OLS) and fixed effects models generally found in the literature. Our tests suggest that correlated random slopes models significantly outperform the literature’s standard models. Moreover, in the 1979 cohort we find that the correlated random slopes coefficient on over-education recovers in significance, and to some extent magnitude, from its diminished size and at times insignificant value found in the literature’s fixed effects estimates. We propose that more studies should investigate changes in the effects of education mismatch over time, as well as how education mismatch affects wages differently across individuals.