Abstract

We employ machine learning techniques to identify common characteristics and features from cases in the US courts of appeals that contribute in determining dissent. Our models were able to predict vote alignment with an average F1 score of 73%, and our results show that the length of the opinion, the number of citations in the opinion, and voting valence, are all key factors in determining dissent. These results indicate that certain high level characteristics of a case can be used to predict dissent. We also explore the influence of dissent using seating patterns of judges, and our results show that raw counts of how often two judges sit together plays a role in dissent. In addition to the dissents, we analyze the notion of memetic phrases occurring in opinions - phrases that see a small spark of popularity but eventually die out in usage - and try to correlate them to dissent.

Replaces