A Taxonomy of Litigation II: Eight Typical Clusters of Causes of Action
As I explored in a previous post, some terrific co-authors and I have written a paper which taxonomizes federal complaints- that is, we find patterns in the kinds of causes of action that attorneys plead. In this post, I’m going to explore those patterns in some more detail.
In our data, spectral clustering revealed eight clusters of causes of action. Each grouping organizes together causes of action that are more likely to be pled together than they are to be pled with others. (This eight-cluster finding is probably not generalizable to all litigation – the paper goes into some detail about the kinds of cases that we included and excluded from our dataset.) When you think about it, that there will be some patterns from this kind of exercise is obvious — there are only a limited number of legally cognizable fact patterns that can cause injury, and attorneys often follow form books/precedent when pleading. Still, we didn’t know what the patterns would be before completing the analysis.
The Figure below provides the most common two or three causes of action per cluster:
This illustrates how, for example, intellectual property claims (like trademark infringement) often travel together with consumer protection claims; civil rights claims (like 1983 allegations) accompany state law torts; and tort claims often fit with contract and fraud claims. This should be old news to anyone who has ever practiced law. Moreover, the Figure doesn’t give us a good handle on how alike or unlike each pattern is from another. Follow me after the jump for the Figure that tries to accomplish just that.
The Figure below results from a force directed graph layout of each case (~2500) in our data. The case’s color represents the cluster it is assigned to; the distances between the cases represent how similar (close) or dissimilar (far) they are from one another.
As the Figure illustrates, there is significant overlap between clusters 3 (civil rights/constitutional), 7 (tort/contract/fraud), and 8 (contract/equity), which you can readily observe from the huge colorful blot the middle of the figure. The remaining clusters are spread further from each other. Indeed, the cluster which is least like the others is one in which federal securities law claims dominate (cluster 6 – the black stars) and which the resulting combination of legal theories is very unlike all others in the data. From this, we can conclude that federal securities law cases have less in common – legally – than do cases based in ordinary commercial torts and contract claims. They rest on a set of facts and doctrine that is consequently more remote.
What does this mean? I’ll leave for another post the practical implications of this kind of clustering work, though the advantages in precision over NOS codes should be obvious. Here, I’d just say that thinking about complaints as defining and advancing clusters of causes of action made me actually reconsider what public adjudication processes were actually doing. I used to think that litigation was performing a winnowing function on plaintiffs, with “more important” cases, or those where the truth was murkier, reaching the
dead klieg-lights of trial. But now, my frame is that of litigation as a tournament – not for plaintiffs, but for causes of action. Each case begins with a cloud of possible legal theories competing for attorney and judicial attention. Some survive (by chance or by law or by facts) later into cases, and they turn up in judicial opinions, which are then read, acontextually, by students, who in turn believe that there is such a thing as a “contract” case, or a “torts” case. But at least as an original matter, there was no such thing. Why do some causes of action survive longer than others? What does pleading practice tell us about attorney strategy and competence? Read (on) and find out.