Experiments of Communications in Games

Abstract

My dissertation includes three experiments to study communications in games. The first chapter uses a game theoretical model and experimental method to study communications. Absence of a shared language is an evident barrier to effective communication. The impact of lack of common knowledge of a shared language is potentially more insidious. We implement a sender-receiver game in a laboratory experiment. Agents agree on the optimal action in every state. Uncertainty about message availability rules out equilibria in which receivers interpret messages consistent with their focal meaning. We predict that when play has converged, receivers will ignore messages, resulting in ineffciency. The experimental results mostly support this hypothesis. The support is stronger when the pooling action is optimal for a larger set of receiver beliefs. The second chapter experimentally investigate a classic question, whether competition stimulates information revelation, by comparing two Bayesian persuasion models, one with one sender (Kamenica and Gentzkow, 2011) and the other with two competing senders who move sequentially (Wu, 2018). The first experiment provides strong support for Kamenica and Gentzkow (2011), where the sender uses a vague signaling device and the receiver complies with his suggestions. In the second experiment, we find that (1) more information is revealed than in the first experiment; (2) the first sender reduces the use of the vague device as compared to the single sender in the first experiment; (3) the second sender exhibits a “matching” behavior pattern; (4) the receiver can make use of information from both sides. However, our experiments also document deviation from the theory. Competition does not improve information revelation to the extent of full information, as predicted by theory. To rationalize the behavior, we use the Quantal Response Equilibrium model to explain the features of the empirical results in our experiments. The third chapter studies the donation to group recipients in response to requests and cooperation. Two treatments were used in the experiment: the recipient can/cannot ask from the donor; the recipients can/ cannot contribute to the group account. Recipients’ group contribution in the threshold public goods game was used to measure the cooperation within the group. The experiment result shows the donor gives more to the high request when deemed as reasonable, but gives less to the excessive request. The donor gives more to the group with high cooperation as a reward.