 |
This paper proposes a new theoretical and experimental methodology that aims to address these two critical issues in order to assess whether overconfidence causes under- reaction to objective information. We decided to apply our method to the perceptual decision framework given that numerous empirical studies in this field have revealed that people do not sufficiently adjust their decision to objective information. We have developed a model based on a Signal Detection Theory approach that has enabled us to: quantify under-reaction, estimate the overestimation of the precision of subjective information, and predict the expected impact of overconfidence on under-reaction by using a Bayesian model. Moreover, our experimental design tries to address the methodological limitations previously mentioned. First, we used a measure of calibration- based overconfidence. Secondly, participants were required to combine subjective and objective information within the same task. Third, we obtained independent behavioral measures of overconfidence and under- reaction. Finally, we describe how our model can be implemented with experimental data and test its predictive power. In the perceptual task, subjects had to compare the number of dots contained in two circles. The two circles were only displayed for a short fraction of time, about one second, so that it was impossible to count the dots. Subjects had to tell which circle contains the higher number of dots. The observation of these two circles constitutes their subjective information. To measure their overconfidence, we asked them to give their confidence in the choice made. Furthermore, objective information which indicated the correct response with 75% validity was provided to participants before the display of the circles. To obtain two independent measures of overconfidence and under- reaction to objective information, participants came to two sessions: one for confidence and one for objective information.
Our results suggest that there is an empirical link between overconfidence and under- reaction to objective information. Overall, participants deviated from an optimal processing of objective information: they set decision criteria that are twice lower than the ideal one. This inefficient use of information results in a sub-optimal performance: they perform at 51% of what they could have ideally reached. Moreover, we found that overconfidence explains 46% of this loss in performance. These results provide evidence in favor of the two main theoretical hypotheses on which are based the model which predicts the link between overconfidence and under- reaction.
