A Model for Confidence Judgments in Choice Tasks

Abstract

<p>It is proposed that the basis of confidence judgments in choice reaction time tasks is the distance between the initial and terminating values of a random walk decision process. Several qualitative predictions and one important quantitative prediction about confidence are derived from the model and tested in a "same-different" line length discrimination reaction time task with two levels of response time deadline. In the task, S simultaneously indicates whether two sequentially presented line lengths are the same length or different lengths and a level of confidence associated with the choice response. The model predicts all the major qualitative features of the data. In addition, a predicted linear relationship between [P(correct)-P(error)] and the average "signed" confidence fits the data very well.</p> <p>A second test of the model, using two deadlines and an "accuracy" condition in which response time was unimportant, replicates both the results of earlier experimenters and the results of the first experiment. Despite a consensus in the literature to the contrary, confidence and response time are shown to be positively related.</p>