Language and Coordination: An Experimental Study

We report on an experiment exploring whether and how players may learn to use a random device to coordinate on a correlated equilibrium that Pareto dominates the Nash equilibria of a two-player Battle of the Sexes game. By contrast with other studies exploring recommendations and correlated equilibria, the mapping from the random device to the action space of the game is not necessarily known by subjects a priori, which serves to highlight the additional coordination problem that is introduced by the use of such a random device. We find that subjects have an easier time coordinating on the efficient correlated equilibrium of the game when there is a common language mapping from the realizations of the random device to the action space of the game. However, we also find that it is possible for subjects to learn to develop a language mapping from realizations of the random device to the action space of the game when that mapping is not common to begin with. We further find that a random device is more useful as a coordination mechanism when subjects are randomly matched. When subjects are in fixed matches, other strategies that ignore the random device and condition instead on the players` joint history, e.g., `alteration` or `turn-taking` can achieve the same efficient outcome that is made possible by introduction of the random device.