Bayesian adaptive model estimation to solve the speed accuracy tradeoff problem in psychophysical experiments

Abstract

AbstractMost psychological experiments measure human cognitive function through the response time and accuracy of the response to a set of stimuli. Since response time and accuracy complement each other, it is often difficult to interpret cognitive performance based on only one dependent measurement and raises a speed-accuracy tradeoff (SAT) problem. In overcoming this problem, SAT experimental paradigms and models that integrate response time and accuracy have been proposed to understand information processing in human cognitive function. However, due to a lengthy SAT experiment for reliable model estimation, SAT experiments' practical limitations have been pointed out. Thus, these limitations call for an efficient technique to shorten the number of trials required to estimate the SAT function reliably. Instead of using a block's stimulus-onset asynchrony (SOA) accuracy with long block-based task trials, we introduced a Bayesian SAT function estimation using trial-by-trial response time and correctness, which makes SAT tasks flexible and easily extendable to multiple trials. We then proposed a Bayesian adaptive method to select optimal SOA by maximizing information gain to estimate model parameters. Simulation results showed that the proposed Bayesian adaptive estimation was highly efficient and robust for accuracy and precision of estimating SAT function by enabling "multiple-step ahead search."