Modeling uncertainty in performance validity testing with Bayesian logistic regression

Le résumé fourni par la source

Objective: The aggregation of multiple performance validity tests (PVTs) into a summary validity determination is an established challenge in neuropsychology. The standard dichotomous accumulation approach relies upon binary recoding of continuous PVT scores and yields ambiguous outcomes in a sizable proportion of cases. This study evaluated Bayesian logistic regression as a framework for estimating the posterior probability of an invalid profile from continuous embedded PVT scores. Method: A retrospective sample of 233 military veterans was classified into Valid (n = 132), Ambiguous (n = 55), and Invalid (n = 46) groups based on three standalone PVTs. A Bayesian logistic regression model was trained on the Valid and Invalid groups using six embedded PVTs as continuous standardized predictors. The Ambiguous group was withheld from training and subsequently evaluated using model-derived posterior probability estimates. Results: The full six-predictor model yielded the highest posterior probability. PVTs varied in inclusion evidence. Probability distributions were consistently low in the Valid group, consistently high in the Invalid group, and markedly heterogeneous in the Ambiguous group. Conclusions: Bayesian logistic regression provides a viable framework for probabilistic PVT aggregation that preserves continuous score information, accounts for intercorrelation among predictors, and models genuine uncertainty when present. Validation in more diverse samples using sophisticated outcome measures is needed.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.