We live in a time where data help illuminate our routines, habits and other information about our lives. But can it actually predict something as important as a Supreme Court decision?
Michigan State University law professor Daniel Martin Katz says yes.
Katz and his colleagues have created an algorithm that has accurately predicted 70 percent of the Supreme Court’s overall decisions, and 71 percent of the votes of individual justices -- more robust results than any other predictive study done to date. Applying various techniques from machine learning, the algorithm takes into account dozens of variables before it makes a prediction.
Katz believes the results represent a major advance in how science and data can be used to help predict legal outcomes.
“Most theories designed to forecast the court’s behavior have not been rigorously tested in a manner that demonstrates their ability to make future predictions,” Katz said. “So we decided to back test our approach against cases already decided.”
The team analyzed more than 60 years of Supreme Court data between 1953 and 2013, a total of 7,700 cases and more than 68,000 justice votes. The model predicts the behavior of 30 justices appointed by 13 presidents through six decades.
Why create the model?
Katz wants to make law more accessible and transparent to the people it is meant to serve. He recognizes that a predictive model has social, political and economic ramifications just as SCOTUS decisions do.
But Katz also believes the best approach to predictive models is a combination of his analytics-based prediction and human expertise.
“Many lawyers have years of expertise and knowledge that a computer simply cannot replicate,” he said. “However, there are actually three ways to forecast something – experts, crowds and algorithms. A combination of these three methods is usually the most powerful and accurate.”
Katz focuses on analytics to improve the accuracy and efficiency of the law, but is using the upcoming 2014 Supreme Court term to test against human prediction. In a tournament aptly named “FantasySCOTUS”, sponsored by Thomson Reuters, legal scholars, predictive experts and everyone in between will be pitted against the algorithm to predict outcomes before they occur.
“It’s like Fantasy Football for law geeks,” Katz said. “We expect that the combined predictions of the computer and the crowd will produce the most accurate outcomes.”
The paper can be found on arXiv and the Social Science Research Network.
Co-authors are Michael J. Bommarito, MSU College of Law, and Josh Blackman, South Texas College of Law.