Calculating estimates of sensitivity from group data: Pooled versus averaged estimators

Abstract

Two methods of estimating the value of population d′ from a group of subjects were investigated to determine the conditions under which each would be the less biased estimator. The first method, calculating d′ for each subject and then averaging the estimates, yields an estimate of population d′ called average d′. The second method requires that data from all subjects be pooled and then an estimate of population d′ calculated. This estimator is called pooled d′. Monte Carlo simulations indicated that pooled d′ was the less biased estimator when the number of trials per subject was fairly low, whereas average d′ was the less biased estimator when the number of trials per subject was fairly high. The crossover point at which average d′ becomes less biased than pooled d′ is shown to depend on the value of population d′, the average bias of subjects in the population, and also on the level of criterion variability within the population. Copyright 1997 Psychonomic Society, Inc.