Abstract
The drift diffusion model (DDM) has been successful in capturing the joint dynamics of accuracy and latency data in various perceptual decision making tasks. We evaluated how well the DDM describes dynamics of perceptual decision when subjects were under a varying degree of time pressure. We collected choice and latency responses from human subjects, who discriminated the size of a thin ring stimulus with a varying degree of uncertainty. The degree of time pressure was manipulated both by giving subjects an explicit instruction of different time limits across sessions (0.7 ∼ 1.2 s) and by providing feedback to responses that were made later than those time limits. When fitted to the data of choice and latency, the three major variants of the DDM (with static bounds & gain, with time-varying bounds, and with time-varying gain) showed a systematic pattern of latency-dependent prediction errors. Here we propose a new variant of the DDM, which adopts a ‘boundary for time hazard’ on the time axis in addition to the choice boundary on the choice-evidence axis in decision space. Our model did not exhibit the biased pattern of errors and was superior than the other models in goodness of fit to the data.