Belief elicitation when more than money matters: controlling for "control"

Incentive compatible mechanisms for eliciting beliefs typically presume that the utilty of money is state independent or that money is the only argument in utility functions. However, subjects may have non-monetary objectives that confound these mechanisms. In particular, psychologists have argued that people favour bets where their ability is involved over equivalent random bets - a so-called preference for control. We propose a new belief elicitation method that mitigates the control preference. Using this method, we determine that under the ostensibly incentive compatible matching probabilities method, subjects report self-beliefs 18% higher than their true beliefs in order to increase control. Non-monetary objectives account for at least 68% of what would normally be measured as overconfidence. We also find that control manifests itself only as a desire for betting on doing well; betting on doing badly is perceived as a negative. Our mechanism can be used to yield better measurements of beliefs in contexts beyond the study of overconfidence