How does our simulator work?

Our bracket simulator is based on peer-reviewed research on the historical seed distributions reaching certain rounds of the tournament.

The original simulator starts with the first round, a.k.a. the Round of 64, and independently chooses a winner of each game at random, with the win probabilities determined by the power model (see the power model paper). Those winners then proceed to the second round, and we apply the power model again to choose the winners of those games. This process continues through the remaining rounds to simulate a complete bracket.

Filling out the bracket in this chronological manner is intuitive but not necessarily optimal, since you could make big mistakes early in the tournament that cost you later. Games in the later rounds are worth more points each in the ESPN Tournament Challenge, so it's essential to get most of the teams right for these rounds. To improve the simulator, we incorporated insights about the seed distributions reaching the Elite Eight and the Final Four in the men's tournament. Modeling these distributions allows us to start by randomly selecting the Elite Eight or Final Four teams, which determines the outcome of the games they must win to get there. You could think of this approach as anchoring the bracket in a round where we are more confident about our forecasts. We then use the power model to fill in the remaining games before and after the predetermined round. Our extensive experiments for the 2013-2019 tournaments show these simulator variants tend to produce better outcomes.

We encourage you to try using each of our three simulator variants and see which you like the best. Enjoy, and let the games begin!

For comments and feedback on this website, send an email to bracketodds@illinois.edu.

Media requests should be made by either emailing bracketodds@illinois.edu
 or calling 217-244-7275. Follow us on twitter: @shjanalytics

bracketodds.cs.illinois.edu ©  2020 - The Board of Trustees of the University of Illinois.
All Rights Reserved.