Performance in Contests: The Role of Risk and Confidence

We use a mixed-method approach to investigate individual performance in contests depending on the ability of the other contestants. We use field data (1,677 unique coders competing in 38 software algorithm competitions) and data from a laboratory experiment (69 participants who solve a task under an incentivized contest scheme). The field data qualify for a regression discontinuity design. Results show that individuals characterized by a medium ability competing against higher-ability contestants perform significantly lower than individuals characterized by a medium ability competing against lowerability contestants. Using the data from the laboratory experiment, we obtain a significant effect that goes in the opposite direction, i.e. contestants in high-ability groups outperform contestants in lowability groups. We do not find evidence that the effort put in the task or the willingness to take risks can fully explain our results. Once we split the sample by the level of self-confidence of the individuals, we find that medium-ability contestants with a high self-confidence perform better in high-ability groups than in low-ability groups. Medium-ability contestants with a low self-confidence, on the contrary, perform worse in high-ability groups than in low-ability groups. Our paper contributes to the contest literature by providing new and causal evidence of the mechanisms causing performance differentials in tournaments.