Modeling Law: Theoretical Implications of Empirical Methods

We examine a long-standing research program in empirical Political Science, fact-pattern analysis (FPA). We connect FPA to definitions of legal rules in jurisprudence and positive political theory. Foundationally, theoretical treatments view rules as functions partitioning case spaces into equivalence classes. Connecting FPA to formal theory has two advantages: first, many elements of the traditional understanding of legal rules become interpretable in terms of FPA, and vice versa. Second, the methodological issues in empirical FPA become much clearer. In particular, the similarity between FPA and related work in artifical intelligence, expert systems, and machine learning becomes obvious. On this basis, following Kastellec 2005, we critique logit, probit, and disciminant analysis-based FPA as less likely to uncover interpretable legal rules than other techniques developed to induce decision trees from data. As noted by Kastellec, classification and regression trees (CART) appear particularly promising in the legal context. We note the possibility of applying CART-based FPA to many areas of the law, with potentially significant applications in legal education and legal practice. We then examine recent attempts to use FPA to uncover changes in “legal regimes.” We suggest that CART could be employed more successfully in this task. We then examine attempts to use FPA in the long-standing “law vs. preferences” debate in Political Science. In our view, this debate as presently framed is unlikely to be productive. We illustrate our points with estimations using simulated data.