LAWRENCE — Jonathan Templin, associate professor of psychology and research in education and associate scientist in the Achievement & Assessment Institute at the University of Kansas, is available to speak with media about using statistical modeling to predict winners of the remaining NCAA Men’s Basketball Tournament games.
The Sweet 16 games tip off tonight, and the Final Four will crown a champion April 4-6 in Indianapolis. Templin’s model shows Kentucky and Wisconsin with the highest probability of winning their next matchups, while Xavier and UCLA have the lowest.
Templin developed a model based on a modified version of the Bradley-Terry model to estimate a relative strength for each NCAA Division I team. The model predicts game outcomes using a logistic regression based on teams played, team strength and other co-variates, including location of the game. Like many people filling out brackets across the country, Templin’s model favors Kentucky to win it all.
“Perhaps the easiest way to summarize the table is to say it is Kentucky versus the field. Kentucky has an estimated .422 probability of winning the tournament. Only five teams (Kentucky, Virginia, Wisconsin, Villanova and Gonzaga) have win probabilities of greater than .05,” Templin wrote on his website prior to the tournament.
And two of those teams, Virginia and Villanova, have been eliminated from the tournament. Through the first full round, the model had correctly identified winners about 73 percent of the time.
Templin’s research primarily focuses on diagnostic classification models and models that seek to provide highly reliable scores from psychological tests. His work has been supported by the National Science Foundation, and he has published in leading academic journals and is the co-author of the book “Diagnostic Measurement: Theory, Methods and Applications.”
The model may be more effective in picking winners than thousands of people who filled out brackets this year the old-fashioned way, but neither method is perfect, and both have their benefits.
“It depends on how good the model is and how good your gut happens to be,” Templin said of using statistical analysis versus “trusting your gut” in making picks. “Probably the best approach would be to blend model results with gut feelings. There is a saying in statistical sciences, commonly attributed to George Box, that ‘all models are wrong, but some are useful.’ Blending model results with other evidence would help in most cases as it would be very difficult to get a model to incorporate everything that would matter in an outcome.”