Designing contests for data science competitions: Number of stages and the prize structures

Firms have been proactively holding data science competitions via online contest platforms to look for innovative solutions from the crowd. When firms are designing such competitions, a key question is “What should be a better contest design to motivate contestants to exert more effort?” We model two commonly observed contest structures (one stage and two stage) and two widely adopted prize structures (high spread and low spread). We employ economic experiments to examine how contest design affects contestants’ effort level. The results reject the base model with rationality assumption. We find that contestants exert significantly more effort in both the first stage and the second stage of the two‐stage contest. Moreover, it is better to assign most prizes to the winner in the two‐stage contest while it does not matter in one stage. To explain the empirical regularities, we develop a behavioral economics model that captures contestants’ psychological aversion to falling behind and continuous exertion of effort. Our findings demonstrate that it is important for contest organizers to account for the nonpecuniary factors that can influence contestants’ behavior in designing a competition.