Overconfidence: Feedback and item difficulty effects.

Overconfident subjects were given immediate feedback of results in a general knowledge test in an attempt to de-bias them. In a 2 x 3 x 4 mixed factorial design (Feedback  Question Difficulty  Trial Blocks), the accuracy, confidence, and overconfidence of judgements of 150 subjects (48 male and 102 female) were measured. Hard questions produced significantly higher levels of overconfidence than medium-difficulty and easy questions, which in turn resulted in underconfidence. Combining all levels of difficulty, females were significantly less overconfident than males. No significant effect of external feedback was found, although better calibration in latter trial blocks for hard-level questions suggests that intrinsic feedback through self-monitoring occurred but was effective in reducing the bias only for hard questions.