Measuring psychometric functions with the diffusion model.

The diffusion decision model (Ratcliff, 1978) was used to examine discrimination for a range of perceptual tasks: numerosity discrimination, number discrimination, brightness discrimination, motion discrimination, speed discrimination, and length discrimination. The model produces a measure of the quality of the information that drives decision processes, a measure termed drift rate in the model. As drift rate varies across experimental conditions that differ in difficulty, a psychometric function that plots drift rate against difficulty can be constructed. Psychometric functions for the tasks in this article usually plot accuracy against difficulty, but for some levels of difficulty, accuracy can be at ceiling. The diffusion model extends the range of difficulty that can be evaluated because drift rates depend on response times (RTs) as well as accuracy, and when RTs decrease across conditions that are all at ceiling in accuracy, then drift rates will distinguish among the conditions. Signal detection theory assumes that the variable driving performance is the z-transform of the accuracy value, and, somewhat surprisingly, this closely matches drift rate extracted from the diffusion model when accuracy is not at ceiling, but not sometimes when accuracy is high. Even though the functions are similar in the middle of the range, the interpretations of the variability in the models (e.g., perceptual variability, decision process variability) are incompatible.