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

Abstract

We address the bias from using potential vs. actual experience in earnings models. Statistical tests reject the classical errors-in-variable framework. The nature of the measurement error is best viewed as a model misspecification problem. We correct for this by modeling actual experience as a stochastic regressor and predicting experience using the NLSY79 and the PSID. Predicted experience measures are applied to the IPUMS. 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.