The encyclopedia · R&D & Science · Operational decision · 2012–2018
A math model matched the NCAA committee's bracket picks 37 of 38 teams.
Mathematical programming selects and seeds the 68-team March Madness field; on 2017 data it matched 37 of the committee's top 38 teams.
NCAA (National Collegiate Athletic Association)
The solution
Each March, a rotating 10-person committee picks 68 of 351 Division I teams for the NCAA men's basketball tournament watched by more than 80 million people, then seeds them. Selection and seeding are subjective and vulnerable to bias and human error.
Bruce Reinig and Ira Horowitz built a mathematical programming model that selects, ranks, and seeds the 68 teams from season data, independent of committee structures, tournament outcomes, and past seasons. Tested against the 2012–2016 tournaments, it consistently reproduced the committee's rankings.
Applied to the 2017 tournament, the algorithm identified 37 of the 38 teams the committee selected; 24 seeds were exact matches, 77.6% were within one seed, and 89.6% were within two seeds of the committee's rankings.
Why it worked
- It turns a subjective committee judgment into an objective, repeatable computation.
- Because the model is transparent, athletic directors can see how schedule changes would move their seed.
- It gives the committee a bias-free starting point while leaving room for human review.
What can be applied
When a small group makes high-stakes ranking decisions, an optimization model built from the same data exposes bias, speeds the work, and gives everyone a defensible baseline.
Aftermath
INFORMS published the study in Interfaces and made the full text freely available during March Madness 2018, putting algorithm-assisted bracket selection into the public debate.
Sources
- New study takes the guesswork out of selecting and seeding teams for 'March Madness'
- New study takes the guesswork out of selecting and seeding teams for 'March Madness'
- Using Mathematical Programming to Select and Seed Teams for the NCAA Tournament
spotted an error? The archive wants to know.