Work experience as a source of specification error in earnings models: implications for gender wage decompositions

This paper models the bias from using potential vs actual experience in log wage models. The nature of the problem is best viewed as specification error as opposed to classical errors-in-variables. We correct for the discrepancy between potential and actual work experience and create a predicted measure of work experience. We use the 1979 National Longitudinal Survey of Youth and the Panel Study of Income Dynamics and extend our findings to the Integrated Public Use Microdata Sample. Our results suggest that potential experience biases the effects of schooling and the rates of return to labor market experience. Using such a measure in earnings models underestimates the explained portion of the male–female wage gap. We are able to separately identify the decomposition biases associated with incorrect experience measures and biased parameter estimates.