Sports

A Math Professor Turns Better Brackets Into Homework


The Louisville Cardinals celebrate after winning the 2013 NCAA Men's Final Four Championship on April 8, 2013 in Atlanta

Photograph by Kevin C. Cox/Getty Images

The Louisville Cardinals celebrate after winning the 2013 NCAA Men's Final Four Championship on April 8, 2013 in Atlanta

As a teacher, Tim Chartier’s main goal is to get his students at Davidson College and the broader public more engaged in math. One of the most effective tools in the professor’s arsenal in recent years has been his phenomenally accurate system for predicting the National College Athletic Association tournament. It started out as a homework assignment for his students, but the results were so good that he turned it into an area of research. His bracket predictions have reached the 99th percentile in pools with millions of people.

You can see some of Chartier’s work on the Davidson website. Click on the Massey Rating section and then click on the interval approach, which breaks up the college basketball season into smaller parts and lets you adjust the importance of each part. A win in March, for example, should probably matter more than a win in December. If you believe the end of the year matters three times more than the beginning of the year, you can put in that exact weighting. Davidson and his team of student researchers have found that the best predictive abilities came by breaking the season down into eight intervals.

The work is definitely not done yet, as the professor and his students search for ways to improve the system. One area of focus is integrating Dean Oliver’s Four Factors of Basketball Success: shooting, rebounding, turnovers, and free throws. By combining offense and defense, each team’s profile can be defined through eight numbers. When trying to predict who will win a future game, Chartier can see the results of past games against teams that have similar profiles as their upcoming opponent. Maybe your favorite team can win against good rebounding teams but not against low-turnover squads—Chartier wants to use that in his predictions. Another area of refinement will modify the analysis if a particular player is lost to injury.

Because the NCAA tournament features many close games by similar teams, one of Chartier’s students came up with the idea of focusing on how teams played against opponents closest to them in the rankings. Chartier likes this idea, because performance against an evenly matched foe is what matters most in the tournament. This measure of “clutch” performance is important when trying to predict those tricky 8-9 matchups in the first round or 4-5 matchups in the second.

If he can improve his prediction accuracy by just a single percentage point, it makes a much bigger difference than you might realize. The odds of getting a perfect bracket can bluntly be quantified as 1 in 9.2 quintillion, if we assume every game is a 50/50 coin flip. But we know it’s not 50/50: Sports statistician Ken Massey says 70 percent is the historical norm for how often favorites win, a number that Chartier agrees with. At 70 percent odds, getting a perfect bracket is much easier: 1 in 5.7 billion.

Chartier and his students work in the zone above 70 percent. All of a sudden, every additional percent matters: At 71 percent accuracy, the odds of perfection improve remarkably—now only 1 in 2.3 billion. And at 72 percent, they are 1 in 973 million. This is why he keeps pushing to find improvements, and for the guessing public, it could be thing that helps you win your pool.

Professional basketball teams are taking notice, too. One of his students, Miles Abbett, interned with the Golden State Warriors D-League affiliate. The Warriors, of course, employ Davidson-alum Steph Curry as their superstar guard. The future might bring even more Davidson grads into the pros, this time in the front office.

Eric-chemi
Chemi is head of research for Businessweek and Bloomberg TV.

Ebola Rising
LIMITED-TIME OFFER SUBSCRIBE NOW

Sponsored Links

Buy a link now!

 
blog comments powered by Disqus