Abstract
Researchers studying cognition often rely on behavioral measures to uncover the underlying process behind correct and incorrect categorizations. However, these behavioral measures do not usually take into account the correct responses that are simply due to chance, which occurs when subjects guess. In a two-alternative discrimination task with a 75% correct response rate, for example, as much as 25% of all responses (or a third of all correct responses) are falsely correct. Here, we present a simple response reclassification procedure that reduces noise caused by false correct responses using response times (RT). The procedure determines, from the observed correct and incorrect RT distributions, a RT cutoff above which correct responses are relabeled as incorrect responses. To illustrate the procedure, we used two published datasets (Faghel-Soubeyrand & Gosselin, 2019; Royer et al., 2015) that employed Bubbles, a method relying heavily on response accuracy to reveal the information used to resolve a visual task. The standard weighted-sum computation applied to the reclassified accuracies led to a 15-20% increase in signal-to-noise ratio—equivalent to running between 32-44% more subjects—compared to the same computation applied to the recorded accuracies. A Matlab implementation of this reclassification procedure is freely available.