Accounting for speed–accuracy tradeoff in perceptual learning

In the perceptual learning (PL) literature, researchers typically focus on improvements in accuracy, such as d′. In contrast, researchers who investigate the practice of cognitive skills focus on improvements in response times (RT). Here, we argue for the importance of accounting for both accuracy and RT in PL experiments, due to the phenomenon of speed–accuracy tradeoff (SAT): at a given level of discriminability, faster responses tend to produce more errors. A formal model of the decision process, such as the diffusion model, can explain the SAT. In this model, a parameter known as the drift rate represents the perceptual strength of the stimulus, where higher drift rates lead to more accurate and faster responses. We applied the diffusion model to analyze responses from a yes–no coherent motion detection task. The results indicate that observers do not use a fixed threshold for evidence accumulation, so changes in the observed accuracy may not provide the most appropriate estimate of learning. Instead, our results suggest that SAT can be accounted for by a modeling approach, and that drift rates offer a promising index of PL.

► Both accuracy and response times are important because of speed–accuracy tradeoff. ► We applied the diffusion model in a perceptual learning (motion detection) task. ► Improvements in accuracy can be due to changes in boundary separation. ► Participants required less evidence to respond by end of training than at beginning.