Experience-Weighted Attraction Learning in Sender-Receiver Signaling Games

Recent experiments have indicated that it is possible to systematically lead subjects to less refined equilibria in signaling games. In this paper, we seek to understand the process by which this occurs using Camerer and Ho's Experience Weighted Attraction (EWA) model of learning in games. We first adapt the model to extensive-form signaling games by specifying that senders update the chosen message for both the realized and unrealized type, but do not update the unchosen message. We test this model against the choice reinforcement and belief-based special cases of EWA; the latter is of particular interest because it formalizes the story about convergence to less refined equilibria offered by Brandts and Holt. We also test a variety of models which update unchosen messages. We find that while the Brandts-Holt story captures the direction of switching from one strategy to another, it does not do a good job at capturing the rate at which the switching occurs. EWA does quite well at predicting the rate of switching, and is slightly bettered by the unchosen message models, which all perform equally well.