Fun With Data: March Madness Edition

Got Gonzaga winning it all? Think again.

It’s that time of year again. The wildest and most exciting event in all of sports (in my opinion, though extreme ironing is a close second) has kicked off: the NCAA College Basketball Tournament, or, simply, March Madness.

It's a tournament that is always ripe with storylines, upsets, disappointments, nail-biting, and so much more. One of the traditional aspects of March Madness from a fan perspective is filling out a tournament bracket and joining a pool of friends, co-workers, or strangers on the internet to see if you can predict the bracket the most accurately.

Every year there is either a clear favorite or a group of teams deemed the most capable of winning it all. This year, it’s overwhelmingly the Gonzaga Bulldogs. The metrics would suggest they are the best team in the country, and most experts and fans agree.

So of course, when you’re filling out your bracket, you should pick the clear favorite to win, right?

Think again.

Enter game theory. If you’re looking to win your bracket pool, it’s very likely that you’re looking at the whole problem incorrectly. You’re probably thinking: if I choose the most games correctly, I win. In actuality, the real thing you should be thinking is: if I choose the most games correctly that other people aren’t choosing, then I am most likely to win.

So what does that mean? Let’s take a look at the information available on ESPN. ESPN tracks and records all brackets entered into their system. According to them, fans are picking Gonzaga to win approximately 27% of the time. Rightfully so, given the metrics suggesting that Gonzaga is indeed the likely victor.

But if you end up choosing Gonzaga and you’re in a bracket pool of 100 people, then you also need to have a better bracket than 26 other people. Even if you correctly guessed Gonzaga to win, you’d still only have about a 3.7% chance of actually winning your bracket pool.

So does that mean you should eliminate Gonzaga in the first round of the tournament to the 16-seed Georgia State? It was only a few years ago that the tiny University of Maryland – Baltimore County Retrievers knocked off the top-ranked Virginia Cavaliers (by 20 points, no less) for the first 16-over-1 upset in tournament history.

Not so fast. The name of the game isn’t exclusively picking the teams nobody else is picking. If that were the case, your Georgia State versus Bryant University final would probably net you some cash. The teams actually still need to win the games for you to win the pool.

You need to pick games other people aren’t picking and you need to be right. You’re looking for a percentage advantage that is roughly the difference between the likelihood of a team to win and the percentage of people who are picking them to win that specific game. Websites like kenpom.com can offer some insight into which teams are “better” and therefore more likely to win. There’s plenty of other sites out there (some free, some not) that will crunch numbers even further and offer up match-up specific probabilities. Get your hands on those and couple with publicly available fan-selected data like the aforementioned ESPN metrics, and you’ve got yourself the recipe to make picks that will have you stand out from the rest of your bracket pool.

But, as we all know, March Madness is March Madness for a reason. Humans are not static data points; they’re more like variable nodes. Anything can happen.

“Conference tournament winners carry momentum.” “You’ve always got to pick at least one 12 over 5 and 11 over 6.” “Guards win in March.” “Teams with NBA talent win.” “Veterans coaches and players have an advantage.” “Don’t pick the team that just fired their coach for potential rules violations.” “If you’re in a pool with a bunch of Tennessee fans, consider bouncing them early.” All of these can be backed by data, and some data is better than none at all.

But there’s a reason why even if you know what you’re doing, your chances of getting everything perfect is absurdly low, and in the ballpark of 1 in 9 quintillion if you’re flipping a coin.

Again, anything can happen.

Ask UMBC or Virginia.