Measuring the Completeness of Theories ⇤
Measuring the Completeness of Theories ⇤
复制标题
衡量理论的完整性⇤
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
S. Mullainathan
中科院分区：
文献类型：
--
作者：
D. Fudenberg;J. Kleinberg;Annie Liang;S. Mullainathan
我们使用机器学习为理论捕获的数据中可预测的变化量提供一种易于处理的度量，我们称之为“完整性”。我们将此方法应用于三个问题：分配特定的彩票等效物，游戏中的初始玩法，以及人类生成随机序列。我们发现在现有模型的完整性方面存在相当大的差异，这就说明是应该专注于开发具有相同特征的更好的模型，还是应该寻找能够改进预测的新特征。我们还说明了完整性如何以及为什么随所考虑的实验而变化，这突出了在选择要运行的实验中所起的作用。假设我们有一个劳动力市场理论，认为一个人的工资取决于他的知识和能力。我们可以通过观察更多的教育是否确实预示着劳动力数据中的更高工资来检验这一理论。如果确实如此，这将为该理论提供支持的证据，但它不会告诉我们是否有另一种理论可能更具预测性。是否有更多的预测理论，如果有的话，它们的预测能力会提高多少，这是“理论是预测的，但它是完整的吗？”的更新版本。我们感谢Amy Finkelstein和Johan Ugander的有益评论。我们也感谢Adrian Bruhin、Helga Fehr-Duda、Thomas Epper、Kevin Leyton-Brown和James Wright与我们分享数据。†麻省理工学院经济系‡康奈尔大学计算机系§宾夕法尼亚大学经济系¶芝加哥大学经济系
We use machine learning to provide a tractable measure of the amount of predictable variation in the data that a theory captures, which we call its “completeness.” We apply this measure to three problems: assigning certain equivalents to lotteries, initial play in games, and human generation of random sequences. We discover considerable variation in the completeness of existing models, which sheds light on whether to focus on developing better models with the same features or instead to look for new features that will improve predictions. We also illustrate how and why completeness varies with the experiments considered, which highlights the role played in choosing which experiments to run. Suppose we have a theory of the labor market that says that a person’s wages depend on their knowledge and capabilities. We can test this theory by looking at whether more education indeed predicts higher wages in labor data. If it does, this would provide evidence in support of the theory, but it would not tell us whether an alternative theory might be even more predictive. The question of whether there are more predictive theories, and if so how much more predictive they might be, raises ⇤This is an updated version of “The Theory is Predictive, but is it Complete?” We thank Amy Finkelstein and Johan Ugander for helpful comments. We are also grateful to Adrian Bruhin, Helga Fehr-Duda, Thomas Epper, Kevin Leyton-Brown, and James Wright for sharing data with us. †Department of Economics, MIT ‡Department of Computer Science, Cornell University §Department of Economics, University of Pennsylvania ¶Department of Economics, University of Chicago