Number preference, precision and implicity confidence

Abstract

In elicitation tasks, people are asked to make estimates under conditions of uncertainty but elicitors then interpret these estimates as if the estimator were certain of them. An analysis of people’s patterns of responding during the elicitation of uncertainty, indicates that there are markers of confidence incorporated into these estimates that can be used to predict the person’s true level of confidence. One such marker is the precision (number of significant figures) of the estimate. Analyses of elicited data show the expected positive relationships between accuracy, precision and explicit confidence and, further, that precision offers information beyond that of explicit confidence ratings. We then demonstrate the importance of incorporating this information on an overconfidence task, showing that it can account for a 9% difference in calibration.