Estimation. In this study, earnings manipula-tors were oversampled relative to their true propor-tion in the population. The econometric justification is that such a state-based sample is likely to generate a larger number of manipulators than a random sample would generate, which would make the identification of a model for classifying earnings manipulation difficult. However, because estima-tion of a dichotomous-state model that ignores the state-based sample procedures would yield asymp-totically biased coefficient estimates, I used weighted exogenous sample maximum likelihood (WESML) probit as well as unweighted probit. 6 The estimation sample spanned the 1982–88 period and consisted of 50 manipulators and 1,708 controls