Cross-sectional regression analyses of wage gaps may be biased by omission of unobserved worker characteristics. Recent studies therefore have used longitudinal data to "difference out" the effects of such variables. This paper. however. shows that self-selection of job changers may cause...