Publications

Professors Create a Method to NCAA March Madness with SAS Predictive Analytics 32 of 34 at-large bids correctly predicted this year (94%); Professors also create new "Score Card" model for game predictions

Every March, the NCAA announces its selections for the annual men's and women's NCAA Basketball Tournaments, aka, "The Big Dance". And every March this creates a storm of controversy, as well as a flurry of office pools and marathon hours of basketball coverage on TV.

This year is no different. However, Jay Coleman, an operations management professor at the University of North Florida, and Allen Lynch, an economics professor at Mercer University, have come to the rescue and removed much of the guesswork by tapping into the power of SAS predictive analytics and business intelligence software to predict which teams will make the tournament. Over the last 11 years, they have a 94% accuracy rate, garnering the attention of the NCAA Selection Committee and the national sports media.

And again this year, Coleman and Lynch correctly picked 32 of the 34 at-large bids (or 63 of 65 total teams) for a 94% success rate, missing only on North Carolina State and Iowa State. However, when Coleman and Lynch plugged in the numbers for the old "RPI" (Ratings Percentage Index) formula used by the NCAA until this year, they picked 33 of 34 at-large teams correctly (minus Iowa State).

Check out the professors' Dance Card site for more information and to see how the new RPI formula adopted by the NCAA affects teams, compares with the old: www.dancecard.unf.edu.

Professors add Game Predictions for the First Time

This year, Coleman and Lynch have added a new twist using the same predictive analytic factors: predicting individual game winners and upsets.

Everyone has their own technique for filling out their NCAA tournament brackets. Sometimes people pour over statistics and team matchups, while others simply guess. But Coleman and Lynch have adopted a more precise approach, a SAS-powered model called the Score Card that predicts individual Tournament games. Using the results of 256 tournament games between the years 2001 and 2004, and nearly 50 pieces of information for each of the two teams that played in each game (e.g., RPI, conference records, out-of-conference records, etc.), they retroactively earned a 75% accuracy overall; 78% accuracy in the first rounds; 72% accuracy in rounds 2 through 6; and a 66% accuracy in "close" games (i.e., games in which the two teams are within 10 spots of each other in the RPI ranking).

How have they done with the 2005 NCAA Tournament? Check out their Score Card website to monitor the latest in game prediction preformance, accuracy, performance and procedures: www.unf.edu/~jcoleman/score.htm.

For more information (and for a SAS video clip to hear Coleman and Lynch talk about the Dance Card in their own voices), see www.sas.com/news/feature/01mar05/dancecard.html.