Improving the efficiency of crowdsourcing contests

While winner-take-all crowdsourcing contests are wide spread in practice, several researchers have observed that their social welfare can be poor due to effort exerted by contestants who are never rewarded. In this paper we study the problem of efficiency in winner-take-all crowdsourcing contests. Using a discrete choice model to capture contestants' production qualities, we introduce a mechanism which filters out low-expertise contestants, before they are asked to produce a solution. We show that under a set of natural assumptions, such a mechanism has desirable incentive properties, attracts high-quality contestants and can improve social welfare. We also provide insights into the problem of prize setting for such contests.