Dynamic overconfidence: a growth curve and cross lagged analysis of accuracy, confidence, overestimation and their relations

Abstract

Research has paid little attention to how overconfidence evolves over time. We examined how task experience (experience within a task using a sequence of items) and outcome feedback affected accuracy, confidence and overconfidence in experiments over several trials. We conducted five studies involving 614 participants and used growth curve modelling and cross-lagged analyses. Findings revealed that mere task experience (without feedback) reduced overestimation linearly. Task experience coupled with feedback reduced overconfidence quadratically; the decreasing rate was initially strong but faded away over time. The decrease in overestimation was explained due to accuracy increasing at a faster rate than confidence did. Accuracy had lagged effects on confidence; a correct estimate led to more confidence in a subsequent estimate. We also found some evidence indicating that confidence had a negative lagged influence on accuracy. This dynamic influence between accuracy and confidence is a unique finding in the overconfidence literature.