On the Specification of Mincerian Wage Regressions with Heterogeneity, Non-Linearity, Non-Separability, and Heteroskedasticity

Using panel data taken from the NLSY, I perform the joint estimation of i) a reduced-form dynamic model of the transition from one grade level to the next with observed and unobserved heterogeneity, and ii) a flexible version of the celebrated Mincerian wage regression with skill heterogeneity, non-linearity in schooling, non-separability between the effects of schooling and experience and heteroskedasticity (after conditioning on unobserved skills). The model rejects all simplifying assumptions common in the empirical literature. In particular, the log wage regression is highly convex, even after conditioning on unobserved and observed skills. Skill heterogeneity is also found to be over-estimated when non-linearity is ignored. After conditioning on skill heterogeneity, schooling has a causal effect on wage growth. I find that estimates obtained in a standard framework (assuming separability) may underestimate the returns to schooling upon labor market entrance by as much as 15%. Finally, I find that the variance of the stochastic wage shock decreases with accumulated experience but is more or less independent of schooling.