A Bayesian approach to abbreviated MMPI's accuracy.

Abstract

This study compared the performance of three abbreviated MMPIs, the Mini-Mult, the FAM, and the MMPI-168 as well as the MMPI, in their ability to predict diagnosis, in their ability to emulate the MMPI, and in their utility with the Gilberstadt and Duker prototype system. Most of the previous research had used the one, two or three highest profile scale matches from the short to long forms to assess code type compatibility instead of an actual typing system. Further, most of the prior studies had used traditional maximum likelihood statistical measures to assess performance. Posterior probabilities calculated from Bayes' rule were used as the primary measure here. Both the original norms and new prediction equations calculated from the sample data for the three short forms were used for all three tests, with a sample size of 578 male veteran inpatients for the Mini-Mult and FAM, and a sample size of 358 for the MMPI-168. The new equations generally outperformed the original ones by most measures, but the trends were similar in both conditions. There was little code type agreement between the short forms and the full MMPI, even though they usually had high scale-to-scale correlations and similar scale means. Part of the reason for this was the low rate of typing on any test, around 30%. Another reason was the low rate of high point correspondence between the short and full forms. A normal code with all clinical scales less than or equal to 70 T-scores performed best, although it is not part of the prototype system. All of the short forms had high valid and false negative rates for the code types with respect to the MMPI's code types. Diagnoses, as set by clinicians and recorded from each case's psychology service file, formed the external performance criteria. Across tests, few code types tended to be associated with a particular diagnosis and diagnoses weren't basically associated with code types. Trying to match diagnoses predicted by the code types with file diagnoses yielded only about a .10 overall probability of a match across tests, including the MMPI, under the most strict conditions. Liberalizing the conditions could bump the hit rates to about .70. The MMPI-168 with new equations was the best short form and outperformed the full MMPI in some conditions. Limitations of this study, some possible implications, and future research are discussed.