Three essays on information transmission games and beliefs in perfect information games

This dissertation consists of three chapters. The first two chapters study strategic information transmission. The third chapter studies beliefs in perfect information games.;The first chapter examines strategic information transmission with interacting decision-makers. I analyze a cheap-talk game between an informed agent and two uninformed decision-makers who coordinate their actions. I compare public communication with private communication. I find that the agent responds to the decision-makers' coordination by providing less precise information. Conditions that support a full information revelation equilibrium in private communication also support the same type of equilibrium in public communication; but the reverse is not true.;The second chapter investigates the role of persuasion mechanisms in collective decision-making. A persuasion mechanism consists of a family of conditional distributions over the underlying state space. A biased, perfectly informed sender adopts a persuasion mechanism to provide a group of uninformed receivers with signals about the unknown state of the world. I compare public persuasion with private persuasion. I find that the sender can always reach the convex upper bound of the set of expected payoffs under public persuasion, regardless of the number of signals or the signals' correlation structure. The sender is worse off under private persuasion. Moreover, I show that private persuasion is always more informative than its public counterpart. As a result, the receivers make better decisions under private persuasion.;The third chapter experimentally explores people's beliefs behind the failure of backward induction in the centipede games. I elicit players' beliefs about opponents' strategies and 1st-order beliefs. I find that subjects maximize their monetary payoffs according to their stated beliefs less frequently in the Baseline Centipede treatment where an efficient non-equilibrium outcome exists; they do so more frequently in the Constant Sum treatment where the efficiency property is removed. Moreover, subjects believe their opponents' maximizing behavior and expect their opponents to hold the same belief less frequently in the Baseline Centipede treatment and more frequently in the Constant Sum treatment.