A Bayesian approach to mixed group validation of performance validity tests

Mental health professionals often use structured assessment tools to help detect individuals who are feigning or exaggerating symptoms. Yet estimating the accuracy of these tools is problematic because no "gold standard" establishes whether someone is malingering or not. Several investigators have recommended using mixed group validation (MGV) to estimate the accuracy of malingering measures, but simulation studies show that typical implementations of MGV may yield vague, biased, or logically impossible results. In this article we describe a Bayesian approach to MGV that addresses and avoids these limitations. After explaining the concepts that underlie our approach, we use previously published data on the Test of Memory Malingering (TOMM; Tombaugh, 1996) to illustrate how our method works. Our findings concerning the TOMM's accuracy, which include insights about covariates such as study population and litigation status, are consistent with results that appear in previous publications. Unlike most investigations of the TOMM's accuracy, our findings neither rely on possibly flawed assumptions about subjects' intentions nor assume that experimental simulators can duplicate the behavior of real-world examinees. Our conceptual approach may prove helpful in evaluating the accuracy of many assessment tools used in clinical contexts and psycholegal determinations.