The NCAA Tournament committee announced its picks for the 2018 tournament yesterday, and all across the country fans are furiously filling out brackets based on skillful analysis, win-loss records and good old-fashioned guesswork. A University of Kansas professor who specializes in statistical modeling has just updated his model for predicting who will win each tournament game.
Jonathan Templin, professor of educational psychology, updated his online model for NCAA tournament game predictions this weekend. The model is based on a formula analyzing all 351 NCAA men’s basketball teams and how consistently they perform. The model considers how many points teams score on average, how many they allow defensively and where the game is played. Templin, who can discuss the model and its predictions with media, said the model does not work on a single-elimination, winner-take-all format like the “big dance.”
“The model works by factoring who is playing, where the game is being played and how consistent each of the teams has been,” Templin said. “Using the model, 10,000 simulated tournaments were drawn at random, and the results were tabulated to determine how far, on average, a team would progress through the tournament.”
Virginia Cavalier fans will be happy to hear the model predicts their team is the most likely to cut down the nets as Final Four champions, with a 22.8 percent chance. KU’s Jayhawks have a 4.1 percent chance of being crowned champions — eighth most likely of the 68 teams — but are also one of the more inconsistent teams in the field, meaning a trip to San Antonio would not be all that surprising. The chances of winning drop quickly after second slotted Villanova, meaning apart from the top two, no team has more than a 6.5 percent chance of winning the tournament, according to the model. The other teams in the top five include Villanova (16.6 percent), Purdue (6.5 percent), Michigan State (6.3 percent) and Michigan (6.2 percent).
Templin cautioned, however, that the model is not a silver bullet for winning office pools or making the unbeatable Vegas bet. The model and its algorithms should be viewed as a source of information, one that can be used in concert with gut feelings and other available analysis when trying to pick who will advance through each round of the tournament.
“I think the model is helpful at picking who may be the stronger team when looking at a pair of teams that you may not have followed closely,” Templin said.
A lifelong sports fan who specializes in statistics and analytics, Templin is a co-mentor of a KU club for students interested in sports analytics. He has also authored studies on variations of movement patterns in people with and without mild Alzheimer’s disease to see if the patterns could predict likelihood of the disease and co-authored a book examining how psychometrics and statistics can help teachers better understand math scores and teach more effectively.