SAIL hosts March MATHness event ahead of NCAA basketball tournament

UNC’s Sports Analysis Intelligence Lab and the Department of Statistics and Operations Research hosted Tim Chartier for March MATHness on March 4. 

Millions of people make predictions and brackets for the NCAA basketball tournament every March. Chartier talked to UNC students about the math of bracket making and how they can better their predictions using analytics and math. 

“Bringing in people to talk to not only SAIL, but just the UNC community as a whole about sports analytics, being such a popular topic, we thought was a worthwhile endeavor,” SAIL founder Conor Kerr said. 

Chartier is a  professor of mathematics and computer science at Davidson College and specializes in sports analytics. One subject he teaches students is bracketology — the process of creating mathematical formulas and models to predict winners of tournament events.

Using bracketology, Chartier and his students at Davidson have created a computer code that can help predict the outcome of games during the March Madness tournament. 

Chartier emphasized that the code does not yield a perfect bracket. Instead, he refers to the predictions as a “cloudy crystal ball” because the data cannot take into consideration any randomness that happens during the tournament. 

There are two methods Chartier uses: Massey's method and the Colley Matrix. 

The Massey method looks at game scores. It rewards blowouts and can be adapted to use only wins. The Colley method looks at wins and losses. Both methods look at the strength of the team’s schedule. Chartier encouraged students to give each method a try. 

Students and people anywhere can use these methods on the March MATHness website. Here, people can pick whether to rank the men's or the women’s tournament and choose either the Massey or the Colley method. Then, people can adjust the rankings by weighting certain factors. Users can upweight home, away or neutral court games and even select a time during the season they think will be the most predictive to their bracket. 

“It will often find certain first-round misseeded teams,” Chartier said. 

Chartier warns bracket makers to not take conference championships into too much consideration, as research has shown that they are not very predictive of tournament play, and to beware of overfitting — an issue in which a prediction model may be accurate to training data but not future inputs. 

“Bracketology itself is always evolving,” statistics and operations research doctoral student Kendall Thomas said. “I think there’s always ways that we can improve our algorithms and that they are going continue to need improvements as the years go on.” 

Chartier ended the event by encouraging students to find data that isn’t already being recorded and to try what comes to mind when analyzing statistics and methods. 

“Sports analytics and data analytics is a place where you can learn to take your thoughts seriously,” Chartier said. “You can create insight that we otherwise don’t have today but you created for tomorrow.”

@dthsports | sports@dailytarheel.com