Modeling Evidence Accumulation With Partial Loss in Expanded Judgment

Abstract

Various sequential sampling models of discrimination were quantitatively tested in a task requiring binary classification of the means of distributions of line segments. Experiment 1 tested accumulator and random walk models, which were shown to be unable to account for either the speed-accuracy trade-off or the moments of the response time distributions. Results were confirmed by Experiment 2 in which distributions were fitted. Data were well described by a model that assumed magnitude-dependent information loss during the sampling period.