Method to their March Madness: Insights from Mining a Novel Large-Scale Dataset of Pool Brackets

Each March, the NCAA Men’s Basketball Tournament attracts tens of millions of viewers, a billion dollars in revenue, and millions of pool brackets from fans attempting to predict the tournament outcome. Previous studies have examined March Madness pools, but no prior work gives an in-depth exploratory analysis of a large-scale bracket dataset. We present a novel dataset of over 200,000 brackets collected from an online pool platform, evenly split between the 2015 and 2016 tournaments. An exploratory analysis of the data reveals insights about the strategies of online pool entrants, which range from rational to absurd. The pool bracket distribution is shown to have surprising quirks that are stable from year to year, such as a large number of all-upsets brackets, and a tendency to over-back the top-ranked team. This exploratory work is a first step toward a generative model that can be used to predict the empirical bracket distribution, given tournament features such as estimated pairwise win probabilities.