Upsets make March Madness special.
In 2018, Virginia was the top overall seed heading into the tournament. They got shocked by UMBC, as a 16 seed beat a 1 seed for the first time in men’s NCAA tournament history.
The only thing better than watching these upsets is picking them in your bracket. How can we use analytics to do that?
Maybe others have some insight. For the past two seasons, The Athletic has written about Bracket Breakers, a math model that predicts March Madness upsets.
Let’s focus on one result: teams that pull off the upset tend to shoot a lot of threes. This makes sense, as shooting more threes should increase the variance in points. An increased variance favors the underdog.
Last year, I decided to test this and other hypotheses about upsets. However, I didn’t restrict myself to tournament games like the Bracket Breakers model.
Instead, I considered all college basketball games in which a team closed in the markets as an underdog of six or more points but won outright. As a data scientist, I like a larger sample size.
In the 2021-22 season, college basketball teams took 37% of their field goal attempts from three. Underdogs of six or more points that won had a three point rate 1.7% lower. Pulling off the upset is not about shooting more threes.
In addition, I’ve found no statistical relationship between a team’s three point rate and variance in points per possession. The idea that shooting more threes increases variance is not supported by the data.
Then what leads to upsets? Making three pointers.
During the 2021-22 season, underdogs of six or more points that won made 5.5% more of their threes than their season average. Favorites that lost made 5.5% less from behind the arc.
This difference in three point shooting from both teams resulted in a seven point swing. This was by far the largest effect that I found.
For example, there is also an effect from two point field goals. Like from behind the arc, underdogs make a higher rate of two pointers while favorites make a lower rate. This resulted in a two point benefit for the underdog.
Fantastic. Three point shooting leads to upsets. All we need to do is predict three point field goal percentage.
However, this is difficult. In a landmark study called The 3-point line is a lottery, Ken Pomeroy found no correlation in a team’s three point percentage from early to late season. This was true for both offense and defense.
The results on offense should be particularly surprising. Shooting is a skill, right? How can it not be predictive?
That’s a nuanced question for another day. For March Madness, let’s conclude that the randomness of three point shooting makes it difficult to predict upsets.
Except for one team. I’ll get into that team that bucks three point regression next week.
Get your March Madness cheat sheet
The follow up to this email will come out on Tuesday, March 14, 2023. It will not be posted on the site.
To find out which team has bucked three point regression, sign up for The Power Rank’s free sports betting email newsletter. This is a free service that strives to be:
You’ll also get my March Madness cheat sheet that makes it drop dead easy to fill out your bracket (this will also not appear on the site). And you’re going to need analytics more than ever in 2023 with such parity in college basketball.
In addition, you get a free pdf of my book How to Win Your NCAA Tournament Pool.
To get this free service delivered straight into your inbox, enter your best email address and click on “Sign up now!”
Leave a Reply