A Comparison of Non-Adjusted and Bootstrapped Methods: Bootstrapped Diagnosis Might be Worth the Trouble

This study compared the diagnostic accuracy of non-adjusted (Rorer, Hoffman, LaForge, and Hsieh, 1966) and bootstrapped (Rorer and Dawes, 1982) diagnosis using a sample of 1,455 psychiatric inpatients who had completed the Millon Clinical Multiaxial Inventory. The usefulness of bootstrapping depended on the criteria for accuracy. Bootstrapping showed no benefit with regard to overall accuracy. Sensitivity increased for five scales upon bootstrapping, although specificity decreased on two scales. The strongest support for bootstrapped diagnosis was in terms of positive and negative predictive power. Positive predictive power increased for 10 scales upon bootstrapping, with no loss in negative predictive power. The discussion examines the conditions under which bootstrapping might increase diagnostic accuracy.