March Madness. There’s a reason they call it that. Unpredictability and a Cinderella team in the making hoping the glass slipper fits. But is there a better way to predict the winner at each round? With nearly $2.5B wagered each year, there’s motivation to answer this question. Most March Madness prediction algorithms neglect how two teams actually match up against one another. What is the right approach? How can you pick a bracket based on this framework? What is the correct framework and analysis to predict March Madness outcomes based on match-ups?
The Approach is Essential
Many statistical predictions are advanced ranking methodologies. If one team is ranked ahead of another team, a higher probability of advancement is assigned to the higher rated team. In order to evaluate match-ups, the margin of victory for a team must be modeled against their opponents’ match-up attributes, to their opponents’ is critical to understand. These attributes fall into one of three categories:
1) Generating points on a given possession
2) Preventing opponents’ from scoring on a given possession
3) Creating more possessions for themselves
Assessing these categories, provides insight on pace of play, post-scoring, three point defending, and other elements that determine the outcome of the game.
Making Your Picks
For any match-up, running a model based on match-ups is critical. The models would agree on a winner, but selecting a winner for the model that has the statistical confidence is essential. The bracket below illustrates the outcome of our approach.
Can anybody beat Kentucky?
According to ESPN and Vegas futures odds most brackets currently have Kentucky as the champion. Though our model predicts Kentucky as the most likely champion, the analytics illustrates, there are some teams that have a chance at beating them. Though they didn’t lose a game prior to the tournament, sorting their margin of victory, common traits are revealed among the teams that can challenge Kentucky for the win. West Virginia, Georgetown, North Carolina, and Iowa all have a chance at the championship. The table below illustrates these four teams meeting the criteria.
Our model predicts a handful of upsets – including two double-digit seeds advancing to the Sweet 16 in Buffalo and Stephen F Austin State University. The table below illustrates Buffalo’s first round match-up against West Virginia.
West Virginia lost to teams that were strong at both offense and defense. However, teams who could defend in the close to mid-range and attempted 3-point shots were a challenge. Buffalo is one of nine teams in Division I basketball (351 total teams) that fits this profile. Looking at Buffalo’s next match-up against Maryland, a similar story can be told.
March Madness is exciting, but applying a statistical and analytical approach to match-ups can help you beat the odds if you’re looking to win.
Many thanks to both Dave Caughman, Strategist and Drew Yao, Senior Operations Research Consultant for assisting with conducting the analysis.