Authorship Verification of Supreme Court Opinions

Abstract

In modern American jurisprudence, the decisions of the Supreme Court of the United States (SCOTUS) are treated as de facto law. However, little is officially disclosed about how justices develop their opinions; internal memos from the bench take decades to be released. This paper presents two contributions to understanding this process. First, we develop a dataset of text tagged with their authors for use in analyzing the style of justices and clerks. Second, we apply the natural language processing task of author verification to this dataset to assess the degree to which statistical machine learning can distinguish the writing style of justices from that of their clerks. We find that random forest classification is particularly adept at this task, with validation F1 scores as high as 0.96.