Betting on Own Knowledge: Experimental Test of Overconfidence

This paper presents a new incentive compatible method for measuring confidence in own knowledge. This method consists of two parts. First, an individual answers several general knowledge questions. Second, the individual chooses among three alternatives: 1) one question is selected at random and the individual receives a payoff if he or she has answered this question correctly; 2) the individual receives the same payoff with a probability equal to the percentage of correctly answered questions; 3) either the first or the second alternative is selected. The choice of the first (second) alternative reveals overconfidence (underconfidence). The individual is well calibrated if he or she chooses the third alternative. Experimental results show that subjects, on average, exhibit underconfidence about their own knowledge when the incentive compatible mechanism is used. Their confidence in own knowledge does not depend on their attitude towards risk/ambiguity.