Statistician weighs the NCAA

by Peter Kauffner

As employees place their office pool bets on the “March Madness” NCAA men’s basketball tournament, some are seeking the advice of Bradley Carlin, an associate professor of biostatistics at the University.
Carlin has created a computer program that simulates the tournament based on the past performances of the teams involved. An article he wrote on the subject was published recently in Chance magazine.
“If your pool is only rewarding you for picking the right answer, you probably want to pick a lot of favorites because favorites historically do well,” Carlin said.
Carlin is betting on Kansas, the number one seed in the Southeast region. The tournament’s 64 teams are divided into four regions of 16 teams each. The teams in each region are ranked based on performance in the regular season.
“If the pool gives you some incentive to take chances (by rewarding upset winners), you do start looking for some (middle) seeds maybe to get upset. I have (Miami of Ohio) making the Sweet 16 even though they are only a 13th seed. Then they get wiped out by the Gophers right away,” Carlin said.
The Sweet Sixteen are the 16 winners of the tournament’s semifinal round, which this year will be played March 21 in San Antonio, Texas, and Syracuse, New York.
Carlin gives the Gophers a 40 percent chance of making it to the Final Four, which will be played on March 29 in Indianapolis. He gives them a 10 percent chance of winning the championship, to be played on March 31.
“I haven’t actually won any money with this system, but then I haven’t been horsing around with it for very long either,” Carlin said.
Carlin uses the average points per game scored by a team during the regular season as a measure of that team’s probability of winning. These probabilities are used by the program to create simulated tournament outcomes.
“For example, in the first round, the probability of the Gophers actually getting beat by Southwest Texas is about 1 percent,” Carlin said.
Many newspapers publish the point spreads for upcoming games. The spread is an estimate of how many points the winning team is likely to win by.
Point spreads can be converted into probabilities through the use of a simple formula. A point spread of 30 corresponds to a 100 percent chance of victory. Thursday’s Kansas-Jackson State game was predicted to be the tournament’s biggest mismatch with a spread of 35 points. In fact, Kansas won by only 14 points, beating Jackson State 78-64.
“These spreads are remarkably accurate,” Carlin said. “Half the time the favorite will cover the spread. No matter which side of the spread you’re on, you have a 50 percent chance of winning the bet.”
“I’m not sure how I got interested in this,” he said. “As a statistician, you think to yourself, there’s got to be an angle, a better way to play these office pools than just picking based on the color of their shorts.”