Doctoral Dissertation Research: An Experimental Analysis of Adpative Learning

博士论文研究：适应性学习的实验分析

• 批准号: 9732854
• 负责人: Edi Karni
• 金额: $ 0.17万
• 依托单位: Johns Hopkins University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1998
• 资助国家: 美国
• 起止时间: 1998-05-01 至 1999-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9732854&HistoricalAwards=false
• 关键词: Doctoral; Dissertation; Research; Experimental; Analysis

How do people play games? More formally, how do people make decisions in strategic situations? This is a question that has concerned game theorists for decades. Do they use the analytic techniques employed by game theorists or do they use other approaches? Evidence from experiments conducted over the past 20 years has shown that in many situations people do not engage in the sort of analysis that a game theorist would. In the last five years, there have been a number of studies aimed at trying to determine if the decision making behavior used by experimental subjects can be described by a few simple learning heuristics and algorithms. The current study first identifies a recurring problem in the methods used to identify player types in these studies and then proposes an alternative experimental design to eliminate this problem. This research is the basis for my dissertation and the funding from the grant will be used in performing the experimental portion of the work. The basic problem with previous studies is that the econometric and statistical methods they use are not capable of discriminating between learning rules. The two basic classes of learning rules differ according to the type of information players are thought to base their decisions on: their own payoff history or past plays by their opponent. Rules that are contained in the same class differ on the particular amount of and manner in which this information is used. Previous studies observed only strategy choices in the experiments and then attempted to use econometric tests to determine simultaneously which type of information the subjects used and how they used it. A simulations based study has shown that these econometric tests fail to be able to succeed in both tasks. Based on the reasons these methods fail, the proposed study includes an extra component in the experiment intended to monitor the use of information by subjects. This is intended to allow the experimenter to determine which type of information the subjects are using in order to aid in the process of modeling their learning or decision making behavior. The result should be a more accurate and reliable means of identifying the use of learning rules among the subjects and possibly even identifying behavior not described by current models.

人们如何玩游戏？更正式地说，人们如何在战略形势下做出决策？这是博弈论学家关注了数十年的问题。他们使用博弈论学家采用的分析技术还是使用其他方法？过去 20 年进行的实验证据表明，在许多情况下，人们不会像博弈论学家那样进行分析。在过去的五年中，已经有许多研究旨在尝试确定实验对象的决策行为是否可以通过一些简单的学习启发法和算法来描述。当前的研究首先确定了这些研究中用于识别玩家类型的方法中反复出现的问题，然后提出了替代的实验设计来消除该问题。这项研究是我论文的基础，赠款的资金将用于执行工作的实验部分。 先前研究的基本问题是他们使用的计量经济学和统计方法无法区分学习规则。学习规则的两类基本类别根据玩家做出决策所依据的信息类型而有所不同：他们自己的支付历史或对手过去的表现。同一类中包含的规则在使用此信息的特定数量和方式方面有所不同。以前的研究仅观察实验中的策略选择，然后尝试使用计量经济学测试来同时确定受试者使用哪种类型的信息以及他们如何使用这些信息。基于模拟的研究表明，这些计量经济学测试未能成功完成这两项任务。基于这些方法失败的原因，拟议的研究在实验中加入了一个额外的组成部分，旨在监测受试者对信息的使用情况。这样做的目的是让实验者确定受试者正在使用哪种类型的信息，以帮助建模他们的学习或决策行为。结果应该是一种更准确、更可靠的方法来识别受试者之间学习规则的使用，甚至可能识别当前模型未描述的行为。