Performance Validity Classification Accuracy of Single-, Pairwise-, and Triple-Failure Models Using the Halstead-Reitan Neuropsychological Battery for Adults

Embedded performance validity tests (PVTs) have been criticized for their poor specificity and sensitivity. Aggregated models of embedded PVTs have been proposed to improve their classification accuracy; however, limitations to aggregation-based improvement of PVTs have yet to be explored. The current study evaluated the classification accuracy of 3 types of models of embedded PVTs in the Halstead-Reitan Neuropsychological Battery for Adults (HRNB): a single-, a pairwise-, and a triple-failure model. In addition, this study evaluated the impact of aggregating between 1 and 6 embedded PVTs in each of these 3 types of models. Analyzing only the 2, 4, and 6 most discriminating embedded PVTs in the single-, pairwise-, and triple-failure models maximized classification accuracy, respectively. Comparisons across these models indicated that the single-failure model including only the two most discriminating embedded PVTs had the best classification accuracy; however, classification accuracy was only minimally improved in this model relative to analyzing just Reliable Digit Span. These results suggest that aggregation of embedded PVTs from the HRNB does not substantially improve their classification accuracy and that the benefits of aggregating PVTs may only emerge when the PVTs entered into the aggregated models have sufficient classification accuracy on their own.