March 18, 2009
Bracketologists plotting this year's March Madness should take note. To win the office pool, you'd better ignore any team's seeding because, in later rounds, its rank is a statistically insignificant predictor of its chances of winning the NCAA basketball tournament.
The basketball tournament to decide a national champion features 65 teams, chosen from all U.S. Division I college teams and ranked in order from first to sixty-fifth by the NCAA Selection Committee. Bracketologists try to predict the field (brackets) before the committee makes its decisions. Once the committee ranks the teams and assigns them to the various brackets for play in the single-elimination tournament, basketball fans try to predict the winners step by step to the final pairing.
"The deeper you get into the tournament, the less effective seeding is in predicting winners," computer scientist Sheldon H. Jacobson of the University of Illinois at Urbana-Champaign said, after studying the results of past tournaments.
At the same time, seedings do matter in earlier rounds, making those results easier to predict. "In the first round, the No. 1 seed has beaten the No. 16 seed 100 percent of the time," Jacobson noted. But after the Sweet Sixteen, a "team's seeding can be thrown out the window. They really do not give you a good indication of who is going to win."
"From the Elite Eight round and onward, you might as well pick names out of a hat," he said. "That is when you are going to see teams you do not expect to win, ending up winning games at a higher statistical rate than would be expected." Intangibles, such as player match-ups, a team's style of play, and a team's "hotness" or "coldness," have a greater impact on outcomes in later rounds of the tournament than a team's seeding.
For seasoned tournament watchers, Jacobson's insights are probably not news. But Jacobson verified their prediction strategies through statistical analysis of data from the modern era of the tournament, beginning in 1985 when the field expanded to 64 teams. Jacobson, along graduate student Douglas M. King, gathered the findings for the article "Seeding in the NCAA Men's Basketball Tournament: When Is a Higher Seed Better?" (forthcoming, in the Journal of Gambling Business and Economics).
Despite its weakness as a prediction model, Jacobson praised the NCAA's seed-based ranking system. "The committee has a very challenging job seeding the teams, and the tournament format by design is exciting," he said.
And that doesn't stop researchers from creating mathematical models to try to get into the prediction game. Operations researcher Joel Sokol and his colleagues at Georgia Tech used a logistic regression Markov chain model to come up with an alternative set of rankings that puts North Carolina, Pittsburgh, Memphis, and Louisville in the Final Four, with North Carolina emerging as the victor.
Source: University of Illinois at Urbana-Champaign, March 16, 2009; Georgia Tech, March 16, 2009.