Disparate Treatment: Theories and Evidence

半个多世纪以来，种族和性别差异一直是美国社会和文化话语的前沿。这项研究有两个不同的项目，有助于更深入地了解不同治疗的来源以及如何补救它。第一个项目研究执法方面的不同待遇。以前关于执法的研究侧重于解释基于可观察到的特征的歧视，但这个项目表明，待遇上的差异不一定与可观察到的差异有关。这项研究开发了一个理论模型，在这个模型中，最大限度地减少犯罪可能需要在一组明显相同的个人中对一个子群体进行更严格的监管。分析表明，随着时间的推移，接受更严格警务的个人的身份可能会轮换，因此随着时间的推移，警务的总体影响是平等分配的。这一理论有助于更全面地理解公平与效率之间的权衡。这一分析也适用于其他环境，如税务审计，它提供了比较静态的结果。这一理论被用来分析高速公路超速数据，并计算警察资源的边际增加所带来的效益，即挽救的生命。作为该项目的一部分开发的建模方法和产生的估计将有助于分析减少不安全驾驶从而减少事故的政策的有效性。第二个项目涉及匹配互补性、指导和榜样效应。这些影响常常被用来为平权行动(本身就是一种不同的待遇)提供理由。虽然在教育学、社会学、经济学和心理学中有很多关于辅导和榜样效应的讨论，但事实证明，这些效应很难定量证明，部分原因是缺乏详细的数据来控制匹配过程的内生性。本项目将研究高等教育背景下的配对效应。构建了一个数据集，允许按种族、民族和性别估计配对学生和教师的好处。评估方法在将学生分配给教师的过程中利用了随机化的优势。通过收集关于学生和教师之间发生的互动类型的全面数据，该项目还提供了对指导和榜样效应发生的渠道的见解。第二个项目的目标是在坚实的实验基础上建立指导和榜样效应的概念。这项研究将更好地了解不同治疗方法的效果。

Racial and gender disparities have been at the forefront of social and cultural discourse in the US for over half a century. This research has two distinct projects that contribute to a deeper understanding of the sources of disparate treatment and how to remedy it. The first project studies disparate treatment in the context of law enforcement. Previous research on law enforcement focuses on explaining discrimination based on observable characteristics, but this project shows that disparities in treatment need not be related to observable differences. The research develops a theoretical model in which minimizing crime may require a subgroup be policed more intensely within a group of observably identical individuals. The analysis suggests that the identity of individuals subjected to more intense policing may be rotated over time, so that over time the overall impact of policing is distributed equally. This theory can help achieve a more comprehensive understanding of the trade-off between fairness and efficiency. The analysis is also applicable to other settings such as tax auditing, and it affords comparative-statics results. The theory is used to analyze data on highway speeding and to compute the benefit, in terms of lives saved, of a marginal increase in police resources. The modeling methods developed as part of this project and the estimates produce will be useful for analyzing the effectiveness of policies that reduce unsafe driving and therefore accidents.The second project concerns match complementarities, mentoring and role model effects. These effects are often used to provide a rationale for affirmative action (itself a form of disparate treatment). While there has been much discussion in education, sociology, economics and psychology about mentoring and role model effects, these effects have proved difficult to demonstrate quantitatively partly to lack of detailed data to control for the endogenous nature of the matching process. This project will study matching effects in the context of higher education. A data set is constructed that allows estimation of the benefit of pairing students and instructors by race, ethnicity and gender. The estimation approach takes advantage of randomization in the process by which students are assigned to instructors. By gathering comprehensive data on the types of interactions that take place between students and instructors, the project also provides insights into the channels through which mentoring and role model effects take place. The goal of this second project is to establish the notion of mentoring and role model effects on a firm experimental basis. This research will provide a better understanding of the effects of disparate treatment.