Measuring Jurisprudence

Abstract Legal jurisprudence is widely debated but rarely measured. We present the first comprehensive measure of jurisprudence in U.S. Supreme Court opinions from 1870 to 2024. Building on qualitative studies of legal reasoning, we classify court opinions into two contrasting types: “formal” reasoning and anti-formal or “grand” reasoning. The foundation of this measurement dataset is a smaller, hand-annotated dataset created by a team of domain experts. Using this annotated dataset, we fine-tune and evaluate a foundational large language model, which is then employed to predict legal reasoning across all opinions in the full dataset. We demonstrate the potential of this new measure for applications in empirical research, enabling analyses of shifts in jurisprudence over time, the reasoning styles of individual justices, and the relationship between legal reasoning and other judicial features, such as ideology. To support further research, we release the annotated dataset, the fine-tuned model, and the final measures, offering a resource for both studying legal reasoning and judicial behavior and evaluating language models in the legal domain.