Why Predicting March Madness Upsets Is So Challenging

Trying to get that elusive perfect bracket prediction depends on being able to spot an upset coming up on the horizon. That's the fun of the tournament in many ways: A high-seeded team could theoretically progress much further than expected and could even have a Cinderella-story victory over the whole field. But picking that specific team or those individual games that will end in an upset is incredibly challenging.

What makes it so difficult? The answer provides an interesting insight into the nature of probability (not to mention how to maximize your chances in your bracket).

_ICYMI: Check out Sciencing's guide to 2019 March Madness, complete with statistics to help you fill out a winning bracket._

The first problem when you're talking about upsets is what actually constitutes an upset. The NCAA defines an upset as any team beating another that is two seed places above them. However, this isn't an "official" definition by any means. Others have opted for a difference of four or more in the seeds, but for our analysis, we're defining an upset as a gap of five seed spaces or more. So a No. 8 vs. No.9 game can't result in an upset but an No. 6 vs. No. 11 game can.

Understanding why predicting upsets is so challenging means understanding why it's so easy to predict that the favorite will win. Most upsets happen in the first round, so using this as a guide and based on games since 1985, the win percentages speak for themselves:

\(\def\arraystretch{1.5}
\begin{array}{c:c}
\text{Matchup} & \text{Games Won by High Seed (\%)}\ \hline
\text{\#1 vs. \#16} & 99 \
\hdashline
\text{\#2 vs. \#15} & 94 \
\hdashline
\text{\#3 vs. \#14} & 85 \
\hdashline
\text{\#4 vs. \#13} & 79
\end{array}\)

The first matchup in particular shows why it's really easy to correctly pick a favorite to win. Out of every 100 No. 1 vs. No. 16 games, only one will result in an upset. You can pretty much pick the first-seeded teams and be near-guaranteed a correct choice. The No. 2 and No. 3 seeds are in a very similar position. For the No. 4 seed, it's not quite as clear-cut, but there will still only be an upset in one out of every five games.

Despite these quite damning statistics, there is still an average of 8.1 upsets per year in the tournament. So if you're going to go for a perfect bracket, you're going to have to include some upsets.

But by the very nature of an "upset," you're trying to choose an unlikely result. Who would really choose the No. 16 seed in the first matchup, after seeing the statistics? Well, in 2018, 1.9 percent of people who completed the March Madness Bracket Challenge did, and at that point there had never been a victory in such a game. Then UMBC beat Virginia by an impressive 20 points (pictured above). The upset happened.

The nature of using probabilities is that you're never really certain. There is pretty much always some potential for the result to go the other way. The challenge of picking upsets is that not only do you need to beat the overwhelming odds, you also have to say where exactly the upset will occur.

It's like saying, if you roll two dice 50 times, which roll number will have a result of 12? The 1/36 chance of any roll being a 12 suggests it will happen over the whole 50, but choosing the individual roll is incredibly difficult. Probability doesn't tell you when; it's just as unlikely on every single roll. Even worse, there might never be a 12 roll, or there could even be 50 of them. This, more than anything, is why definite predictions against the odds are always challenging.

So you're always going to be fighting the odds, but if you want a perfect bracket, you'll have to include some upsets. Luckily, you can still use the data to help you decide where to put them.

Most of your picks should be in the first round. The data suggests that of the 4.6 average upsets per year, the No. 11 vs. No. 6 matchups are the most likely place, closely followed by the No. 12 vs. No. 5 and then there is a bigger gap in the number of upsets between that and the No. 14 vs. No. 3 games. It continues as you'd expect – down to the No. 15 vs. No. 2 matchups, which have only had eight upsets since 1985. The best advice is to pick four or five upsets in this round and keep most of them for the No. 11 or No. 12 seed teams.

The second round has 2.9 upsets per year, with the No. 7 vs. No. 2, No. 10 vs. No. 2 and No. 11 vs. No. 3 games being the most likely spots for an upset. Beyond this, upsets are much less likely but they're still very possible. If you pick between six and ten upsets in your bracket, keep most of them in the first two rounds, but if you're going with more upsets, one or two later on is the best approach.

_Feeling the March Madness spirit? Check out our tips and tricks for filling out a bracket, and read why it's so tough to pick a perfect bracket._