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...