Clinical and statistical issues related to predicting therapeutic outcome

Attempted to determine whether the therapeutic outcome of an alcoholism treatment program could be predicted (N = 81). Data were analyzed both by a mathematically oriented linear regression approach and by a clinically oriented retrospective parametric approach. When the two types of analyses were applied to the same data base each analysis identified different predictors of therapeutic outcome. This created an apparent dilemma as to how prediction of therapeutic outcome ought to be done. It was suggested that the correlational regression approach be used to determine what to expect from therapy and the parametric clinical approach be used to explain why therapy succeeded or failed.