Assessing working memory test performance with a cognitive psychometric race model

The present study tested the properties of a cognitive psychometric model for the joint analysis of responses and response times in binding-type working memory capacity (WMC) tasks. These models aim to decompose the task solving process into a set of informative process parameters and to dissociate person and item effects at the same time. Specifically, we fit an IRT-race model with two accumulators to data from three binding-type WMC tests (letter-colour: N = 2,385, word-number: N = 3,898, letter-position: N = 2,272). The accumulator for correct responses (knowledge accumulator) was interpreted as the efficiency of binding processes, the accumulator for incorrect responses (discontinuation accumulator) as a tendency to discontinue the task at hand. The fitted models displayed decent fit to the data. Reliability of the trait estimates across the three tests was satisfactory (ω = 0.86 and ω = 0.76 for the two accumulators). Validity of the trait estimates was investigated through correlations with performance in different working memory updating tests, a test of crystallized intelligence, and with the Big Five personality traits. Notably, the trait estimate for the knowledge accumulator was correlated more strongly with updating performance (r = 0.62) as compared to standard accuracy scores (r = 0.52). Furthermore, there was a moderate negative correlation of the trait estimate for the discontinuation accumulator with Openness (r = −0.26) and with crystallized intelligence (r = −0.26). This study demonstrates that individual differences in process parameters can be incrementally predictive when modelled reliably.