Identifying the Precedential Value of Case Law - A Machine Learning Approach.

Studying legal change is a central task for both political science and law. While political science has focused mainly on legislative processes, a significant – and increasing – subset of legal change occurs in the precedent setting by high courts and international courts. By contrast, lawyers have long analyzed and normatively evaluated precedent setting through doctrinal research. This typically involves identifying what the law “is” and how cases should be resolved, and benchmarking these evaluations against actual court judgments. However, both disciplines lack empirical approaches to identifying court driven legal change over a large body of case law. As a consequence, general theories aiming at explaining such change have been hard to test. This paper reports progress from a project that aims to improve on this condition, focusing on the Court of Justice of the European Union (CJEU). Like other precedent setting courts, the judgments of the CJEU inlude holdings, which specify rules and principles relevant to the case at hand. The core challenge that we aim to address is how to identify changes to the relevance of such holdings without having to engage in resource intensive qualitative legal research. This is especially challenging in the EU context, where the CJEU rarely ackknowledges when it departs from previous precedent. The solution we propose is to study how the Court cites itself. Legal change occurs when the Court decides to no longer rely on a certain precedent in cases where it previously would have done so. We develop a method for capturing when such shifts occur based on the Court’s previous citation behavior and the characteristics of new incoming cases. The key innovation in our approach is that we consider the extent to which a holding is expected to be cited in new cases that arrive at the Court, given the relevance of its holding to those cases. We use machine learning techniques to predict expected citations and operationalize precedential value as the ratio of actual citations to expected citations. Precedential value decreases for cases that are expected to be cited — given the legal issues presented by new cases and the Court’s previous citation patterns — but where the expectation is not matched by actual citations. These unfulfilled expectations provide insight into the extent to which the Court departs from or stays on its doctrinal course. Besides contributing to academic research on court driven legal change our project aims at developing better tools for judges and practicing lawyers when deciding which cases to rely on.