Taking Justices' Legal Preferences Seriously

• In Event: Reconsidering Justice Preferences and Goals on the US Supreme Court

Abstract

Theorists tend to concentrate on the policy or doctrine output of the Court, while empiricists focus on the disposition of cases. The problem this presents is not that different camps are engaged in different enterprises, but rather, that we often fail to distinguish the two modes of work at all. This paper makes three key contributions. First, we present a novel model of judicial decisionmaking at the Supreme Court that centers justices' preferences over legal policy. Second, we trained a deep learning Transformers classifier to extract paragraphs containing legal opinions from all 2010 Supreme Court decisions. Our fine-tuned RoBERTa model is able to distinguish the law, as presented in legal text, from facts, applications, orders, and other types of content within opinions with a very high degree of accuracy. This is itself is a major contribution to the field: our model can separate wheat from chaff in any legal text. Third, we bring together these first two elements and show that our model of justices' observed votes, along with the legal content of written opinions, can yield new--and better--estimates of justices' legal policy preferences. Specifically, we adapt Kim, Londregran, and Ratkovic's sparse factor analysis (SFA) approach to estimate justices' ideal points from a combination of legal text and their decisions to join opinions.

Authors

• Ben Johnson, Penn State Law

• Logan Strother, Purdue University

• Joan C. Timoneda, Purdue University