Math professors unveil latest model designed to predict NCAA tourney upsets
Since they began working with ESPN senior writer Peter Keating six years ago, Furman math professors Kevin Hutson, Liz Bouzarth and John Harris ’91 have gotten pretty good at predicting upsets in the NCAA Division I men’s basketball tournament.
In 2018, for example, their top upset picks – defined at a team seeded at least five spots lower than its opponent – were No. 11 Loyola over No. 6 Miami and No. 9 Florida State over No. 1 Xavier. Loyola and FSU both won. In 2017, their models actually had No. 11 Xavier favored over Maryland, and, sure enough, the Musketeers were victorious.
The year before, their top four most likely first-round upsets all ended up being just that.
So when the trio says that this time around they’ve combined all of the successful features of their past models into one, you may definitely want to pay attention before filling out your office bracket.
“This year we made an ensemble model, incorporating all of our past models together, weighting each model in terms of its usefulness in predicting past games,” Hutson said. “We calculate a probability for each game rather than giving an outright prediction. Historically, upsets have occurred at a 22 percent rate, so any upset probability that we get above this is one that we think has a better than average chance of occurring.”
The team with the most to fear according to their calculations is No. 6 Iowa State, which has a 50-percent chance of being beaten by No. 11 Ohio State. Another 11 seed, St. Mary’s, is given almost as good of a shot (46 percent) to knock off Villanova.
Other games to watch are the Belmont/Temple winner against Maryland, Murray State and Marquette, and Liberty versus Mississippi State.
Wofford, which barreled through the Southern Conference undefeated, has a healthy 32-percent chance to defeat Kentucky in the second round, assuming the Terriers make it that far.
“Overall we created these eight models, and each model is trying to pick out something important about the game and then give its own probability,” Hutson said. “We looked at all the past tournaments and we let each model pick the games, and those models that turned out to have a higher percentage of picking the games were weighted more heavily.”
The professors gave Maryland-Baltimore County a less than 4 percent chance to defeat Virginia last year in the first round, but of course the Retrievers pulled off the greatest upset in tournament history when they became the first 16 to beat a 1 (prior to that No. 1 seeds were 132-0 against 16s). Virginia is seeded first again in 2019, and Bouzarth says the Cavaliers have by far the best chance of all of the No. 1s to be upset, with their models giving No. 16 Gardner-Webb a nearly 10-percent chance of victory.
“Virginia does look like the most vulnerable 1. Ten percent is huge for a 16-1,” she said. “Virginia is going to try to slow the game down, which is exactly the opposite of what a giant should do. A giant should not slow the game down.”
Keating relied heavily on their models when he was writing his “Giant Killers” column for ESPN, but this year, while they’re still collaborating, there is no formal relationship between the network and the Furman math professors. Which could end up being a good thing for the university.
“We have more flexibility now I think to follow this through from the academic side and see where it goes, because before we hadn’t been able to publish any papers,” Bouzarth said. “It was ESPN’s data.”
They were all hoping to be predicting Furman’s chances at an upset in this year’s tournament, but a historic season wasn’t enough to land the Paladins a spot in the 68-team field for the first time since 1980. Their computer rankings had Furman 66th.
They do, however, give the Paladins a good chance Wednesday night when they host Wichita State in the first round of the NIT.
“We predict Furman to win this game. We would give them a four-point edge,” Hutson said. “You can quote us on that.”
Director of News and Media Strategy