Experimental Studies of Strategic Sophistication and Cognition in Normal-Form Games

范式博弈中策略复杂性和认知的实验研究

• 批准号: 9617358
• 负责人: Vincent Crawford
• 金额: $ 23.71万
• 依托单位: University of California-San Diego
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 1997
• 资助国家: 美国
• 起止时间: 1997-03-01 至 2001-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9617358&HistoricalAwards=false
• 关键词: Experimental; Studies; Strategic; Sophistication; Cognition

This project investigates the extent to which peoples' behavior in games reflects their attempts to predict others' decisions by taking their incentives into account. (These games are interactive decision problems in which each person's rewards may depend on others' decisions as well as his own, so that a person's best decision may depend on his predictions of others' decisions.) This notion, called `strategic sophistication`, is of primary importance in predicting the outcomes of economic, political, and social interactions. Not surprisingly, it is also a central issue in theoretical analyses of strategic behavior. The leading theoretical frameworks treat it very differently: Noncooperative and cooperative game theory take the degree of strategic sophistication to be unlimited, while evolutionary game theory and adaptive learning models take it to be nonexistent or severely limited. The persistence of such disparate theoretical treatments suggests that questions about strategic sophistication are unlikely to be resolved by theory alone. Empirical analysis requires tight control of preferences and information, which seldom occurs naturally in the field but which can be achieved in laboratory experiments. Experiments have an additional advantage for studying sophistication, which has not yet been fully exploited. Sophistication is an aspect of cognition, which usually must be studied indirectly, by drawing inferences from the model that best describes observed behavior. However, Camerer, Johnson, and their collaborators have developed methods to study cognition via subjects' information search patterns in `extensive-form` games--games in which peoples' decisions are represented in sequence, as in a decision tree. Their results suggest that appropriately designed experiments make it possible to study sophistication more directly. Camerer and Johnson's designs exploit special features of the extensive-form games they studied, but the Mouselab software they developed also makes it possible to present games in `normal form` -- as payoff matrices or tables in which subjects can look up their own and others' payoffs for the different possible combinations of decisions as frequently as desired, one at a time. The experiments in this project use Mouselab to study strategic sophistication in normal-form games. There are close connections between game-theoretic analyses of behavior in extensive- and normal-form games, but their cognitive foundations are very different. The different presentation and processing of payoff information in normal-form games make it possible to explore aspects of sophistication that do not come into play in extensive- form games. The main goals are to test noncooperative game theory's cognitive implications, and to learn whether deviations from them help to predict the occurrence and nature of deviations from the theory's behavioral implications. The experiments should yield cognitive insights complementary to those of Camerer and Johnson, and shed new light on strategic behavior. The proposal concerns experiments to be designed and conducted in collaboration with Miguel Costa Gomes, a graduate student at the University of California, San Diego, and Bruno Broseta of the University of Arizona.

该项目调查人们在游戏中的行为在多大程度上反映了他们通过考虑他人的动机来预测他人决策的尝试。 （这些游戏是交互式决策问题，其中每个人的奖励可能取决于他人和自己的决策，因此一个人的最佳决策可能取决于他对他人决策的预测。）这种称为“战略复杂性”的概念对于预测经济、政治和社会互动的结果至关重要。 毫不奇怪，这也是战略行为理论分析的中心问题。 主要的理论框架对此的处理方式截然不同：非合作博弈论和合作博弈论认为策略的复杂程度是无限的，而进化博弈论和适应性学习模型则认为策略复杂程度不存在或受到严重限制。这种不同的理论处理方式的持续存在表明，有关战略复杂性的问题不可能仅靠理论来解决。 实证分析需要严格控制偏好和信息，这在现场很少自然发生，但可以在实验室实验中实现。实验对于研究复杂性还有一个额外的优势，但这一优势尚未得到充分利用。 复杂性是认知的一个方面，通常必须通过从最能描述观察到的行为的模型中得出推论来间接研究。 然而，卡默勒、约翰逊和他们的合作者已经开发出了通过受试者在“扩展形式”游戏中的信息搜索模式来研究认知的方法，在这种游戏中，人们的决策是按顺序表示的，就像决策树一样。 他们的结果表明，适当设计的实验可以更直接地研究复杂性。 Camerer 和 Johnson 的设计利用了他们研究的扩展形式游戏的特殊功能，但他们开发的 Mouselab 软件还可以以“正常形式”呈现游戏——作为收益矩阵或表格，受试者可以在其中根据需要一次一个地频繁地查找自己和他人对不同可能的决策组合的收益。 该项目中的实验使用 Mouselab 研究正常形式游戏中的策略复杂性。对扩展型博弈和常态型博弈行为的博弈论分析之间存在密切联系，但它们的认知基础却截然不同。正常形式博弈中收益信息的不同呈现和处理使得探索在扩展形式博弈中不发挥作用的复杂性方面成为可能。 主要目标是测试非合作博弈论的认知含义，并了解偏离非合作博弈论是否有助于预测偏离理论行为含义的发生和性质。 这些实验应该会产生与卡默勒和约翰逊的认知见解相补充的认知见解，并为战略行为提供新的启示。 该提案涉及与加州大学圣地亚哥分校研究生米格尔·科斯塔·戈麦斯和亚利桑那大学的布鲁诺·布罗塞塔合作设计和进行的实验。