Learning-by-Doing and Project Choice: a Dynamic Structural Model of Crowdsourcing

This paper studies determinants of project choice in online crowdsourcing contests using a unique dataset from the world’s largest competitive software development portal. Particular attention is given to the strategic roles of learning and forward-looking behavior in influencing contestants’ decisions. We use a structural dynamic discrete programming (DDP) model to conduct our analysis and adopt a Bayesian approach to estimation. Our preliminary results provide evidence of learning-by-doing influencing propensities of users to choose projects of different types. The value of the parameter of intertemporal substitution that we identify suggests that while users are forward-looking, the aggregate behavior is far from fully rational. We attribute that result to mix of forward-looking and myopic users in the population.