Post by Erik Hovenkamp
In the most recent Private Law Workshop, Yun-Chien Chang discussed his ongoing and very interesting effort (with Nuno Garoupa and Martin Wells) to use machine learning techniques to classify legal systems into families—groupings whose members are similar to each other, but relatively distinct from those in other groups. The paper, entitled Redrawing the Legal Family Tree: Empirical Comparative Law Using Data on Property Doctrines, is the first attempt to classify legal regimes into families using advanced statistical methodology.
As the authors describe it, research on international legal families has reached an impasse. Prior efforts to delimit legal families have been challenged as “largely subjective.” And, although legal families have grown popular within economic research, particularly in development economics, these studies tend to take for granted these prior genealogies. Because of this, according to the authors, the “economic literature has failed to influence scholarly debate in comparative law.”
In keeping with prior work in the field, Chang and his coauthors were interested in delimiting families based on substantive private law. They ultimately chose to focus exclusively on property law, both for methodological convenience and because, in the authors’ words, it is “neither too local nor too global.” This ostensibly helps to ensure that there are sufficient international similarities to formulate meaningful groupings, but not so much so that all countries are lumped into just one or two silos.
The authors hand-coded substantive property law attributes of 154 jurisdictions. Specifically, each data point describes how a nation’s regime of property law comes out on 73 different issues—for example, whether it recognizes a lease as a property form, or whether it affords a right to abandon real property. They then used a machine learning clustering algorithm to agglomerate legal systems into families. And as a robustness check, they applied a technique called “sparse linear discriminant analysis,” which is designed to enable effective grouping when the number of variables is high in relation to the number of observations.
The authors’ results comport with some of the prominent distinctions drawn in prior genealogies, such as that between civil and common law systems. But, at least within these broad categories, the clustering algorithm identifies different legal families. For example, unlike prior accounts, it does not identify the legal systems of former-Soviet bloc as an independent family; the same is true of East Asia. With respect to civil law jurisdictions, the authors identify four major families, which they associate with French, German, Spanish, and Scandinavian law.
The paper was a fun read. As the authors proceed with this project, it would be helpful to receive a more thorough discussion of how their results relate to the historiographical arguments underpinning prior efforts to classify legal families. In any case, it will be interesting to see how the authors and other comparative-law scholars employ these results in the future.