An Experimental Study of Rationality and Learning in Noncooperative Games

经济学中的大多数博弈论应用都假设参与者是理性的、相互理性的、纯粹的自利的。在游戏的实验研究中，其中一个或多个假设似乎经常被违反，因为人们不玩纳什或更精细的均衡(学习有时会减少违规，但并不总是)。通常很难单从策略选择就判断出受试者违反了哪些假设。由于博弈论解决方案概念可以被解释为用于选择策略的算法，因此我们可以使用计算机系统来直接测试算法，该计算机系统记录受试者正在查看游戏中的什么信息(例如，哪些收益)以及持续多长时间，以及他们的选择。信息处理数据揭示了人们对信息的偏好，这可以用来推断一种看不见的思维过程。我们还可以测量学习对信息处理的影响，以了解受试者是如何学习的，以及他们是否学习了最佳反应或一般原则(例如，解决方案概念)。本项目将研究的游戏和想法包括：排序讨价还价、正向归纳、纳什对抗子游戏完美，以及“几乎是常识”的游戏。

Most game-theoretic applications in economics assume players are rational, mutually rational, and purely self-interested. In experimental studies of games one or more of these assumptions often appear to be violated because people do not play Nash or more refined equilibria (learning sometimes reduces violations but not always). It is usually difficult to tell from strategy choices alone which assumptions a subject is violating. Since game- theoretic solution concepts can be interpreted as algorithms for choosing a strategy, we can test the algorithms directly using a computer system that records what information in a game (e.g., which payoffs) a subject is looking at, and for how long, along with their choices. The information processing data are revealed preferences for information, which can be used to infer an unobservable thinking process. We can also measure the effect of learning on information processing to see how subjects learn, and whether they learn optimal responses or general principles (e.g., solution concepts). The games and ideas that will be studied in this project include: sequencing bargaining, forward induction, Nash vs. subgame perfection, and "almost common knowledge" games.