Learning in Games and the Interpretation of Natural Experiments

(joint with Drew Fudenberg)

Link to paper

Abstract:

We examine natural experiments where the variable of interest is the effort of the agents, the treatment and control correspond to success or failure, and there is unobserved heterogeneity in the agents' efforts. We show that in such experiments the treatment effect estimated by standard methods such as regression discontinuity analysis or difference-in-differences may contain a transient learning effect that is entangled with the long-term preference eect of the treatment. This learning effect occurs when agents are uncertain of the effectiveness of their effort: Success or failure gives agents information about how much their eort matters to success, and consequently changes the amount of effort they provide after treatment. We examine how the learning eect changes the estimated treatment effect, and when its impact is likely to be substantial. We illustrate our findings with applications taken from the literature, and show how under some circumstances the presence of learning can alter policy conclusions.