Most of this year’s NCAA bracket contests follow the same tired format: Millions of people fill out a 64-team bracket and get points for correct picks — more if those picks come in later rounds. But some contests have their own quirks and present their own incentives. Some pools award bonus points for correctly predicting upsets; others use an auction format to draft teams. FiveThirtyEight’s office pool will reward participants for choosing results no one else or few people in the pool do.
Mark Glickman, a statistician and research professor of health policy and management at Boston University, has created a very different kind of March Madness competition. Participants won’t merely predict which team will win each game; they’ll specify how sure they are. Entrants will offer a probability for every game that could happen: 2,278 possible matchups.
Glickman’s contest scores entrants by the product of the average probability they gave for each matchup’s winner. The setup is designed to push people to examine their assumptions and balance confidence with caution. If you ever thought you were 100 percent sure about a game — say, between a No. 1 seed and a No. 16 seed — think again: If you’re wrong and a team you gave no chance somehow wins, you will lose a lot of points. Call every game 50-50, and you’ll know your score before the tournament starts. But it won’t be a good one.
The format also changes entrants’ incentives. A great upset pick helps, like in any pool, but a bad one could help sink a bracket. Participants can recover more easily from early misfires than in other contests, because they’re assessed on their probability estimates for every matchup — even ones they didn’t expect. On the other hand, in many other contests, including ESPN’s Tournament Challenge, points per game double in each round, so a correct title pick is worth 32 times an accurate round-of-64 pick. But in Glickman’s contest, later games don’t count more than earlier ones.
“By forcing people to make probability predictions, we will help distinguish people who have pretty close predictions but not exactly the same,” Glickman said.
His novel setup is one of many bracket contests to go against the grain. What kinds of contests have you participated in, and how would you design an office pool for quants that offers more interest than the typical format?