As college basketball changes, so does our model to predict upsets in the NCAA Tournament (2024)

The full NCAA Tournament bracket has been released with UConn leading the way as No. 1 overall

In our 19th year of writing about NCAA tournament upsets, we’re feeling a bit nostalgic. The landscape that we’ve come to know is shifting faster than a Tristen Newton crossover. By this time next year, for all practical purposes, the Pac-12 will no longer exist. Soon, the tournament field might include 76 teams. Another massive wave of players will have switched schools amidst the perpetual free agency that is the transfer portal. And NIL deals — where players receive monetary compensation for use of their name, image and likeness — will continue to change the way teams are constructed while administrators struggle to figure out how to govern the process.

In short, college basketball is in a period of chaos.

We’re unsure what the new world order will ultimately mean for March Madness. But we’ve learned one grand lesson in all these years of studying upsets: Embracing chaos increases the likelihood of unexpected outcomes. No matter what conference UCLA calls home nor how much money a top recruit might earn, that fact is not likely to change, especially when it comes to the style of play that leads underdogs to topple more talented foes.

Get your March Madness printable bracket

It’s all about employing a high-risk, high-reward style. Our statistical model, Slingshot, has found patterns among successful Bracket Breakers that have stood the test of time. Some tend to shoot lots of three-pointers, risking a lower-percentage shot for an extra point. Others send additional bodies to pursue offensive rebounds, gambling that their extra shot attempts will outweigh increased transition opportunities for their opponent. Other teams extend their defense, pressuring the ball and disrupting the passing lanes hoping that they will force bundles of turnovers. And many play a slow tempo: The fewer total possessions in a game, the less chance that the better team has to separate itself. As we often say, if you’re playing one-on-one against Steph Curry, would you rather play to 11 or two?

The best underdogs — like VCU in 2011, Abilene Christian in 2021 or Furman last year (all favorites of our model) — have employed several of these strategies. They turn games into a chaotic ball of stress for their opponent and, in a one-and-done scenario, that’s often enough to end a favorite’s season.

And top seeds are even more vulnerable when they don’t take strategic steps to insulate themselves from underdogs’ preferred tactics. Safe giants avoid turnovers, dominate the boards at both ends and limit opponents’ three-point shooting. Think of it this way: Over the course of a six-game run, most teams will have to win a game where they don’t shoot well. The safest overdogs have other ways to win when things go wrong. The most vulnerable — like several recent Virginia teams — don’t employ methods to create extra possessions and crumble when their shots stop going down (or their opponents get hot).

Put this all together, and we can assert that Samford, for instance — recently crowned the champion of the Southern Conference but just the 101st-best team in the country — will make a dangerous opponent in the NCAA tournament. The Bulldogs rank 10th in the country in steals and 131st in offensive rebounding percentage, and take and make a lot of long-range shots (40.9% 3PA/FGA, ranking 90th; 39.3% 3-point FG%, ranking 7th). Our model loves their “Bucky Ball” style of play, not because it is hyper-aggressive or fun (though it’s both) but because it involves multiple intelligent gambles to build the volume and value of Samford’s shots. Slingshot estimates their underdog traits will boost their tournament performance by a banging 6 points per 100 possessions.

Of course, how any team performs once it gets to the Big Dance depends not just on its own style but also that of its opponent. Favorites and longshots each cluster into groups of teams with similar characteristics. The cool part is that matchups matter greatly among these various statistical families; patterns emerge when overdogs and longshots from the different groups meet one another in the tournament. So Slingshot looks not only at power ratings that are up- or downgraded by key traits, it takes cluster collisions into account.

We’ll discuss this in more detail once we all have actual matchups to analyze. But here are a few effects to keep in mind.

For more Underdogs, listen toPeter and Jordan’s podcast.

First, any style that builds possessions is better than nothing for high seeds. Every year, some teams earn high seeds even though they’re not particularly strong at offensive or defensive rebounding or at avoiding or creating turnovers. We call them “Generic Giants,” and they lose in big upsets about twice as often as all other favorites. They’re often terrific shooters — they pretty much have to be. But in March, one wrong-footed stretch can send you packing. (Red flag this year: Kansas.)

Second, even giants who build possessions are at risk to a group of underdogs we call Slow Killers. This statistical family grinds down and suffocates its foes, relying on second-chance shots and, when they have to, threes to keep themselves in close games. They’re responsible for a long line of upsets that were memorable but not easy for opposing coaches or media pundits to explain, like Iowa State over LSU and Wisconsin in 2022, Xavier over Florida State and Arizona in 2017, Wichita State over Kansas in 2015 and Harvard over Cincinnati in 2014. In 2021, 11-seed UCLA, 12-seed Oregon State and 13-seed North Texas, who posted a combined nine tournament wins, all belonged to this cluster. We’ve said before that this isn’t a very analytical way to put things, but what successful Slow Killers have in common is that they drive superior opponents crazy; they discombobulate giants used to controlling games.

Third, longshots that rely on offensive rebounding can be very effective underdogs — except against favorites with the same key strength. Then it’s like playing your older brother, which is never fun for the little guy. Last year, Providence ranked 16th in the country in offensive rebounding, but its fate as an 11-seed was basically sealed when it drew Kentucky, which ranked No. 1 in that category. (And their matchup played true to form, with the Wildcats crushing the Friars on the boards.)

In addition to these considerations, we’ve bolted two more factors onto the ensemble of calculations that Slingshot performs. One is something we call Similarity, where our model looks at a matchup and measures which contests from our spreadsheets (which go back to 2007) are its closest statistical matches. We’ve learned that similar games often tell us something that bit-by-bit analysis of data points doesn’t capture, sometimes with eye-popping results.

For example, last year, we found that in the 10 most similar games to Virginia’s first-round contest against Furman, underdogs won six times and outscored favorites by an average of 4.1 points per 100 possessions. We had never seen that for a 4 vs. 13 matchup. And while there were other reasons to smell an upset brewing, Similarity gave us even more confidence to keep cracking Virginia jokes. Since last year, using similar games has been part of Slingshot’s predictive model.

Slingshot understands pace better than “slow tempo is good for underdogs,” which is true but incomplete. A team looks fast when it runs a quick offense. And offensive pace drives a team’s ranking in overall tempo as measured by possessions per game. But we have learned that it also matters how a longshot plays when it doesn’t have the ball. Indeed, a string of very deep underdogs sporting wide gaps between their average possession length on offense and defense have pulled off big upsets in recent years. To name a few: Ohio as a 14-seed in 2010 and a 13-seed in 2012, Florida Gulf Coast in 2015, 13-seed Buffalo and 13-seed Marshall in 2018 and 11-seed Syracuse in 2021. This is a fascinating group of teams with an underappreciated ability to play fast on offense while making their opponents grind out possessions at the other end. And we now incorporate metrics for possession length into Slingshot, too.

Slingshot is constantly evolving. And in the ever-changing college basketball world, its next discovery might be soon. Perhaps, as we build a greater sample size of upsets in the transfer portal era, we’ll find that continuity matters. Maybe, as expansion creates mega-conferences, we’ll find that conference strength plays an increasingly important role in projecting upsets. Maybe NIL will increase the disparity between the haves and have-nots, leading to fewer upsets altogether. Or maybe it actually will trigger more unexpected results.

We’re excited to see what comes next. In the meantime, it’s time to embrace the chaos. Bring on the upsets.

Thanks to John Harris, Kevin Hutson and Liz Bouzarth of Furman University for research assistance.

(Photo illustration: Sean Reilly / The Athletic; Photos: Justin Casterline / Getty Images and John Byrum / Icon Sportswire via Getty Images)