Breaking the Percent Memory Retention Ceiling using Bayesian Statistics.

Abstract

Neuropsychological tests of episodic memory often include a measure of memory retention to facilitate the diagnosis of memory disorders. However, the traditional percent retention (PR) score has limited interpretability when smaller amounts of information are both initially learned and later recalled, creating a pseudo-ceiling effect. To improve psychometrics of PR, we investigated a scoring procedure that incorporates levels of certainty into estimates of memory retention based on learning level. Word-list recall data from adults with traumatic brain injury were modeled using a uniform prior in the Bayesian framework. From the resultant posterior probability distributions, we derived a measure referred to as retention probability (RPr), which distinguishes the retention of relatively good and poor learners. PR and RPr scores were compared on their distributional properties and associations with theoretically related memory measures. Significant distributional differences between PR and RPr were observed. RPr removed the conspicuous ceiling of PR, resulting in stronger correlational and predictive relationships with other memory measures. A Bayesian procedure for quantifying memory retention has psychometric advantages and potentially widespread applicability for measuring the change in behavioral features over time. Future directions are briefly discussed. A sample RPr calculator is provided for interactive exploration of the method.