Collaborative Research: The Dimensions of Supreme Court Decision Making, 1946-2000

什么因素最能解释美国最高法院法官的法律决定？在判决案件时，法官的政策偏好在多大程度上超过了纯粹的形式和法律考虑？有多少政策维度构成了战后时期法官的偏好？当前的最高法院是否比文森和早期的沃伦法院更保守？随着时间的推移，下级法院的判决是否变得更加自由？随着时间的推移，法院的政策产出发生了怎样的变化？这些只是研究人员将根据美国最高法院的特点定制的统计模型来回答的少数实质性问题。更具体地说，使用贝叶斯推理方法，主要研究者开发了项目反应模型的变体：1)适用于少数受试者的多维选择情境；2)在理想点和情况参数下显式建立动力学模型；3)可以用协变量（案件事实、原审法院、提出的论点、考虑的问题等）来解释投票行为，同时控制和衡量政策偏好。为了模拟理想点和案例参数的动态，主要研究人员在项目反应建模的背景下使用动态线性模型（DLM）机制。该建议的目标之一是集成这些建模策略，以便多维模型能够以计算高效的方式适合纵向数据。本提案的第二个方法学目标是建立和拟合模型，共同测量政策偏好，并解释测量协变量对投票决策的影响。该项目位于社会统计中几个最新发展领域的交叉点：二元时间序列和时间序列横断面数据分析、混合建模、层次贝叶斯建模和潜在变量建模。虽然提议的模型类别将针对涉及美国最高法院决策的工作进行定制，但许多提议的方法将与政治学的其他应用（如下级法院的行为，委员会决策和立法行为）和社会统计的其他领域（如教育统计，心理测量学和计量经济学）相关。

What factors best explain the legal decisions of U.S. Supreme Court justices? To what extent do the policy preferences of justices outweigh purely formal, legal concerns when deciding cases? How many policy dimensions structure the preferences of justices in the post-war era? Is the current Court more conservative than the Vinson and early Warren courts? Have the decisions of lower courts become more liberal over time? In what manner have the policy outputs of the Court changed over time? These are but a handful of the substantive questions the researchers will answer using statistical models customized to the peculiarities of the U.S. Supreme Court. More specifically, using a Bayesian inferential approach, the principal investigators develop variants of item response models that: 1) are suitable for multidimensional choice situations with small numbers of subjects; 2) explicitly model dynamics in ideal points and case-parameters; and 3) can be used to explain voting behavior with covariates (case facts, the court of origin, the arguments raised, the issues considered, etc.) while simultaneously controlling for and measuring policy preferences. To model the dynamics of ideal points and case parameters, the principal investigators use the dynamic linear model (DLM) machinery within the context of item response modeling. One of the goals of this proposal is to integrate these modeling strategies so that multidimensional models can be fit to longitudinal data in a computationally efficient manner. The second methodological goal of this proposal is to build and fit models that jointly measure policy preferences and account for the effects of measured covariates on voting decisions. This project lies at the intersection of several recent growth areas in social statistics: binary time-series and time-series cross-sectional data analysis, mixture modeling, hierarchical Bayesian modeling, and latent variable modeling. While the proposed class of models will be customized for work involving decision making on the U.S. Supreme Court, many of the proposed methods will be relevant to other applications in political science (such as behavior on lower courts, committee decision making, and legislative behavior) and other areas of social statistics (such as educational statistics, psychometrics, and econometrics).