B - 139 A Bayesian Approach to Performance Validity in Neuropsychological Testing

Abstract Objective Using the Test of Memory Malingering (TOMM), this study seeks to explore the feasibility of a continuous approach to performance validity by simulating experiments using a Bayesian statistical algorithm. Methodology A literature review was carried out to conduct statistical analyses from data of two known groups (those putting forth credible effort and those not). TOMM scores were modeled using various statistical distributions, with the inflated Conway-Maxwell-Poisson (ICMP) distribution providing the best fit. Simulation experiments were conducted to test the Bayesian statistics model. This involved generating artificial data from known distributions and then seeing how well the algorithm could recover the probability of a score belonging to the credible effort group. We tested the algorithm’s robustness by intentionally starting with incorrect assumptions and seeing if it could still converge on the correct probabilities. Finally, the Bayesian model was applied to real TOMM data to estimate the probability of credible effort for each possible score. Results The simulation studies showed that the Bayesian approach was generally robust, even when starting with incorrect assumptions. When applied to real data, TOMM scores of 40, 45, and 50 were essential, with approximately 5% change occurring with each score. A score of 40 covers approximately 50% of scores, 45 covers approximately 75%, and, finally, 50 covers 95% of scores. A perfect score of 50 encompassed ~95% of credible performances but not 100%. Conclusion This study provides initial statistically derived evidence for the feasibility and potential value of a continuous, probabilistic approach to performance validity.