Changes in the Returns to Skill and the Variance of Unobserved Ability

Changes in the variance of wages among groups of workers with common observable characteristics are often interpreted as reflecting changes in the return to unobservable skills. This interpretation relies on the crucial and highly restrictive assumption that the variance of these unobservable skills remains constant over time. We propose a new identification strategy which relaxes this assumption using longitudinal data, and requires only two observations per individual. Using data from the Current Population Survey's Merged Outgoing Rotation Group sample over the period 1982-2012, we find that relaxing the assumption of constant within-group skill variance is crucial. Contrary to the conclusion drawn when this assumption is imposed, we find that the return to skills has fallen over our sample period, and that increases in within-group wage inequality are driven by increases in the dispersion of unobserved skills, particularly among college graduates.