Summary The common estimator for relative adjusted wage differentials based on dummy variables in loglinear regressions is inconsistent in the presence of group-specific heteroskedasticity. The direction and magnitude of the bias are derived. At the same time, a heteroskedasticity-consistent...