A Bayesian Sequential Sampling Model of Choice Reaction Time Incorporating Stimulus Onset/Duration Uncertainty

Jordan MeyerUniversity of Michigan
Jun ZhangUniversity of Michigan

Abstract

We propose a Bayesian sequential sampling model of choice reaction time (RT) which incorporates uncertainties about stimulus identity, onset, and duration. The model is the now-standard random-walk/drift-diffusion model, with a threshold-based response mechanism. The "substance" of the drift, however, is the posterior probability (belief) that a participant updates on a moment-to-moment basis during a trial — the update is done by combining the likelihood function on the evidence (modeling trial-dependent perception) with prior probability about stimulus identity, onset time, and duration (modeling trial-independent task knowledge). Response threshold, which equals the probability of correct response in choosing each alternative conditioned on prior knowledge and accumulated evidence, modulates speed-accuracy tradeoff. While sequential Bayesian updating without temporal uncertainty (regarding stimulus onset/offset) is trivial, we overcome the hurdle of incorporating the temporal prior into the dynamics of belief updating to derive an analytic expression for Bayesian belief. The advantage of the Bayesian formulation is to allow full control of where and how many free parameters appear: in likelihood functions, priors, or response threshold. Comparison of computer simulation of our model with human performance data (Smith, 1995) will be reported.

Files

A Bayesian Sequential Sampling Model of Choice Reaction Time Incorporating Stimulus Onset/Duration Uncertainty (314 KB)



Back to Table of Contents