Experimental Evidence *

Costless and non-binding pre-play communication (cheap talk) has been found to often be effective in achieving efficient outcomes in experimental games. However, in previous two-player experimental games each player was informed about both his payoff and the action of the other player in the pair. In the field, people may engage in cheap talk and subsequently learn their payoffs, but frequently only learn their own payoffs and not the actions of other people. We model this uncertainty in the framework of a 2x2 coordination game, in which one choice leads to the same payoff regardless of the action of the other player. We vary whether messages about intended play are permitted, and whether participants are informed about the other person's play. Cheap talk is found to be effective, as there is much more coordination in both Signal treatments than in either of the No Signal treatments. We also find that information about the other person’s play appears to increase coordination when messages are permitted. However, in the No Signal treatments, the round-to-round changes in choices induced by this additional information are unable to overcome the apparent pessimism about the feasibility of coordination without a signal.