By Jason Rantanen
In her article Predicting Patent Litigation, recently published in the Texas Law Review (and available here.), Professor Colleen Chien proposed a model for predicting which patents are most likely to be litigated based on a combination of intrinsic and acquired patent file characteristics. The article has since spawned two short responses, one by Professor Lee Petherbridge of Loyola Los Angeles and the other by Professors Jay Kesan (Illinois), David Schwartz (Chicago-Kent), and Ted Sichelman (San Diego), both published by the Texas Law Review. Both acknowledge Chien's substantial contribution to the literature, but express caution about the extent to which it is possible to perform practically useful data-driven patent litigation prediction given the current state of knowledge.
In On Predicting Patent Litigation, Lee Petherbridge argues that data-driven patent litigation prediction presents difficult modeling problems, especially when models are constructed from the types of data identified in Predicting Patent Litigation. He points out that models used to identify which patents are likely to be litigated need to be both specific and sensitive in order for them to be useful to innovators and their advisors. He also explains that models that rely on causal relationships between acquired patent characteristics (such as reexamination) and patent litigation can be adversely affected by problems such as omitted variable bias and simultaneity. These concerns, he argues, limit such models’ predictive usefulness because they can suppress the specificity and sensitivity necessary to make models practically useful. The response is available here.
Kesan et. al. also express reservations about Professor Chien's model in Paving the Path to Accurately Predicting Legal Outcomes, but do so from the angle of methodological concerns. In particular, Kesan et. al. identify and discuss the merits and the limitations Chien's dataset and empirical methodology, also raising the endogeneity concerns discussed by Petherbridge. In addition, they question the connection between Chien’s observations and her policy recommendations regarding patent record-keeping. The response is available here.
Taken together, these three works provide both a great starting point and a useful guide for anyone considering a data-driven approach to predicting patent litigation.