Résumé en anglais

Miscalibration can be defined as the fact that people think that their knowledge is more precise than it actually is. In a typical miscalibration experiment, subjects are asked to provide subjective confidence intervals. A very robust finding is that subjects provide too narrow intervals at the 90% level. As a result a lot less than 90% of correct answers fall inside the 90% intervals provided. As miscalibration is linked with bad results on an experimental financial market (Biais et al., 2005) and entrepreneurial success is positively correlated with good calibration (Regner et al., 2006), it appears interesting to look for a way to cure or at least reduce miscalibration. Previous attempts to remove the miscalibration bias relied on extremely long and tedious procedures. Here, we design an experimental setting that provides several different incentives, in particular strong monetary incentives i.e. that make miscalibration costly. Our main result is that a thirty-minute training session has an effect on men's calibration but no effect on women's.