Strategic Reasoning in Persuasion Games: An Experiment

Abstract

We study experimentally persuasion games in which a sender (e.g., a seller) with private information provides verifiable but potentially vague information (e.g., about the quality of a product) to a receiver (e.g., a buyer). Various theoretical solution concepts such as sequential equilibrium or iterated admissibility predict unraveling of information. Iterative admissibility also provides predictions for every finite level of reasoning about rationality. Overall we observe behavior consistent with relatively high levels of reasoning. While iterative admissibility implies that the level of reasoning required for unraveling is increasing in the number of quality levels, we find only insignificantly more unraveling in a game with two quality levels compared to a game with four quality levels. There is weak evidence for learning higher-level reasoning in later rounds of the experiments. Participants display difficulties in transferring learning to unravel in a game with two quality levels to a game with four quality levels. Finally, participants who score higher on cognitive abilities in Raven's progressive matrices test also display significantly higher levels of reasoning in our persuasion games although the effect-size is small.