Procedural Complexity and Economic Behavior

摘要经济学家传统上认为，人类能够使用非常复杂的决策规则在市场和组织中做出最优选择。经济学家还传统上认为，人类有能力遵守政府和组织官僚机构制定的非常复杂的规则，比如复杂的税法和拍卖规则。几十年来，无论是在对照实验室实验中，还是在使用现场数据的研究中，这些传统的假设都受到了质疑。经济学家明白，人类能够执行的任务的复杂性是有限度的，但事实证明，很难正式定义和衡量复杂性，并将这一概念用于预测和制定政策。这项建议使用经济学、心理学和计算机科学的概念和经验工具，仔细地通过实验测量规则的哪些特征使它们在人类执行的难度和个人负担方面变得复杂。这一数据语料库将提高我们在面对复杂情况时预测人类行为的能力，并使我们能够制定有效、简化的政策，现实地考虑到人类应对复杂的能力。内部决策程序和外部规则都可以建模为算法，不是由机器实现的，而是由人类实现的。例如，重复博弈中的策略，以及许多决策问题中的过程，可以描述为理论计算机科学中简单的形式化算法模型。这项研究由一系列实验组成，旨在评估各种将程序和规则背后的算法归类为对人类来说或多或少复杂的方法，并了解哪些复杂性衡量标准准确地预测和描述了人的错误和主观成本。该提案考虑了对算法复杂性进行建模的不同方法，并检查了它们对受试者实施决策规则的难度的预测能力。这项研究的目标是开发一种基于经验的人类程序复杂性决定因素的特征，这将有助于建立人类行为的预测性模型，并有助于设计更有效的政策。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

AbstractEconomists traditionally assume that humans are capable of using very complex decision rules to make optimal choices in markets and organizations. Economists also traditionally assume that humans are capable of complying with very complex rules set by governments and organizational bureaucracies, such as complex tax codes and auction rules. Several decades of research both in controlled laboratory experiments and using field data call these traditional assumptions into question. Economists understand that there exist bounds on the complexity of tasks humans are able to perform, but it has proved difficult to formally define and measure complexity and put the notion to use in making predictions and setting policy. This proposal uses conceptual and empirical tools from economics, psychology and computer science to carefully experimentally measure what features of rules make them complex in terms of difficulty and personal burden for humans to perform. This corpus of data will improve our ability to predict human behavior when faced with complexity, and allow us to develop effective, streamlined policies that realistically account for human capacities to cope with complexity.Both internal decision procedures and external rules can be modeled as algorithms, implemented not by machines but by humans. For instance, strategies in repeated games, and procedures in many decision problems, can be described as simple formal models of algorithms from theoretical computer science. This research consists of a series of experiments designed to evaluate various ways of classifying the algorithms underlying procedures and rules as more or less complex for humans, and to understand which measures of complexity accurately predict and describe human mistakes and subjective costs. The proposal considers distinct ways of modeling the complexity of algorithms and examines how well they predict subjects’ difficulty with implementing decision rules. The goal of the research is to develop an empirically grounded characterization of the determinants of procedural complexity for humans, which will be useful for building predictive models of human behavior, and for designing more effective policies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.