Overconfidence and performance: Evidence from a simple real-effort task

Abstract

Using a simple real-effort counting task and frequency-based prediction elicitation, we document significant absolute and relative overconfidence for a diverse subject pool. Consistent with the Dunning–Kruger effect, an inverse relationship exists between task performance and overconfidence such that low (high) performing individuals exhibit significantly more (less) overconfidence. This relationship holds for absolute overconfidence even after accounting for better-than-average effect and regression-to-the mean and can potentially explain the lack of absolute overconfidence reported in some economic studies. Further, we find negligible correlation between our task-based measures and survey-based overconfidence measures commonly used in psychology studies, indicating these two methodologies may capture different behavioral phenomena.