The singular beauty of the NCAA Basketball Tournament lies in its unpredictability. Right here, right now, as the big dance is about to get underway, every underdog can dream of joining history’s most triumphant Cinderellas.
Depending on how many gray hairs you have, you might remember heartstoppers like Villanova overcoming Georgetown for the national title in 1985. Or Butler, an 8-seed, and VCU, an 11-seed, both making the Final Four in 2011. Or UMBC in 2018 becoming the first 16-seed to beat a 1-seed in the men’s tourney. Or Notre Dame’s Arike Ogunbowale sinking an overtime jumper to topple seemingly unbeatable UConn on the women’s side that same year.
The triumph of underdogs like these, no matter how brief, affirms for athletes and fans alike that anything can happen — which is one of the reasons we all love sports. To quote the poetry we will all be hearing for the next month: “’Cause inside you knew / That one shining moment, you reached deep inside / One shining moment, you knew you were alive.”
But what if we told you that upsets are more predictable than the world commonly assumes?
Don’t worry. We’re not here to bludgeon the romance out of the NCAA Tournament with a stack of spreadsheets. We can’t — and wouldn’t want to — make March Madness fully rational. But we have studied the men’s tournament for two decades with an eye towards the decisive details of how Davids lay low Goliaths.
We started forecasting upsets in 2006 with our Giant Killers series at ESPN, and we’re still going strong with Bracket Breakers here at The Athletic, 20 years later. We utilize a statistical model (nickname: Slingshot) that we developed with the help of John Harris, Kevin Hutson and Liz Bouzarth from the Furman University Mathematics Department to help forecast upsets. There are ways to use the patterns common to favorites and underdogs to improve your brackets — and get a little smarter about sports in general.
Cinderella’s top traits
Across sports, underdogs win by playing high-risk/high-reward styles. The wider the variation in a team’s scoring, the better its chances to topple a superior foe. In basketball, that means the feistiest underdogs build extra possessions by crashing the offensive boards and forcing turnovers. They launch lots of 3-pointers.
Of course, these tactics can produce inconsistent results. When they work, they boost the quantity or value of a team’s possessions. When they don’t, they can put a team on the wrong side of a blowout. But in a win-or-go-home scenario like March Madness, it pays to take a puncher’s chance.
Would-be Cinderellas also tend to benefit from keeping games slow. That’s just arithmetic. Each possession is an individual battle. The fewer possessions, the fewer battles the lesser team needs to win. If you’re playing one-on-one against Steph Curry, you’d rather play to one than 11.
In contrast, giants are best protected from upsets when they can play hassle-free and convert their skill advantage into points without getting trapped by chaos or trickery. The strongest favorites amass offensive rebounds (giving themselves second chances) and avoid turnovers (keeping weaker teams from getting extra possessions and limiting easy transition opportunities).
Slingshot digests all of this info in three basic steps. First, it calculates power rankings, rating every Division I squad by its margins of victory and strength of schedule. Next, our model uses regression analysis to determine which statistical characteristics are associated with teams that over- or underperform in tournament games and determines the importance of each of those variables. Then it looks at the latest crop of tournament entrants and adjusts each team’s basic strength according to how much it shares those key traits.
This lets us say that Drake is the best potential giant-killer this March. The Bulldogs rank 67th in the NCAA in our basic power ratings, but they’re also the best team in the nation at forcing steals (14 percent of opponent possessions) and 18th-best at seizing their own missed shots (36.8 percent). And they play at the pace of your grandma entering a WiFi password (just 59 possessions per game, the only D-I team below 60). Slingshot estimates this constellation of talents would amp Drake’s strength in a tournament game by a whopping 14.9 points per 100 possessions, the most of any team this year (or last).
The dance partner matters
Of course, there are myriad differences among teams beyond their overall strengths. Akron, which has launched nearly 1,000 three-pointers this season, plays nothing like Yale, which attacks the paint and relies on rebounding to build possessions. Indeed, using a technique called cluster analysis, Slingshot has found that favorites and long shots each bunch into four distinct families that share statistical features. (This rundown has more details.)
Here’s what’s really interesting about that: Collisions among members of these various clans produce widely different results. The various sets of teams have diverse effects on one another, and our model can measure those impacts. Strong offensive rebounding, for instance, generally makes a team a better underdog. But long shots that define themselves by that trait almost always get smoked when they have to face a top seed with the same dominant skill, as Providence learned against Kentucky in the 2023 tournament. For underdogs, that particular matchup is like playing your big brother. This is a way of quantifying something we all know but isn’t easy to gauge, which is that matchups matter.
We will dive further into cluster effects once we have actual matchups to examine on Selection Sunday. But we want to mention a couple of patterns now because they crop up so frequently.
First, for high seeds, any style that builds possessions is better than no style at all. There are always some teams that land in the uppermost regions of brackets despite not being particularly strong at rebounding at either end 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 outstanding shooters (they almost have to be). But that’s a double-edged sword because one cold stretch can send this kind of Goliath home for the summer.
Further, the underdogs we call “Slow Killers” are especially dangerous. This cluster of statistical cousins grinds down superior foes, limiting their field goal percentages while relying on second-chance shots and, when necessary, hoisting 3s to hang tough in close games. Slow Killers’ disruptive efforts have led to a long line of NCAA upsets that traditional analysis didn’t predict and found hard to explain. Some examples: Harvard over Cincinnati in 2014, Xavier over Florida State and Arizona in 2017, and Iowa State over LSU and Wisconsin in 2022.
Both trends were in full force in the upset-happy tournament of 2021. That year, Ohio State, a 2-seed with dangerously weak turnover and rebounding stats, was taken down by No. 15 Oral Roberts. And Abilene Christian, Oregon State and UCLA, all Slow Killers seeded No. 11 or below, won a combined nine games. This year’s brackets could be combustible for similar reasons. Slingshot sees Michigan, Oregon and Wisconsin as generic and classifies Drake, McNeese and UC San Diego — three of its very favorite long shots — as Slow Killers.
And a little bit of history repeating
Slingshot includes two more sets of data in its suite of calculations. For one thing, looking at past games similar to a contest we are studying often yields significant results. A favorite example: In 2023, we found that in the 10 matchups most similar to Virginia vs. Furman, the Davids beat the Goliaths six times and outscored them by 4.1 points per 100 possessions. We had never before seen anything like that for a game involving 4- and 13-seeds, and it helped us be, ahem, cavalier about UVA’s chances. For every tournament game, our model now examines which teams and matchups in our spreadsheets (which date to 2007) are the most similar.
We have also moved toward a more sophisticated analysis of tempo. Prompted by some fascinating research, we began splitting apart teams’ average possession length (APL) on offense and defense. Lo and behold, many deep dogs share an important trait. Ohio in 2010 (seeded 14th) and 2012 (seeded 13th), Florida Gulf Coast in 2013 (seeded 15th), Buffalo (seeded 13th) and Marshall (seeded 13th) in 2018 and Syracuse (seeded 11th) in 2021 all had huge gaps between their offensive and defensive APL. Essentially, they played fast when they had the ball but had such strong defenses that they forced opponents to use larger chunks of the shot clock. And they all won in the tournament.
Slingshot now incorporates this knowledge — one reason that this year it likes Utah State, which ranks 52nd in the country in average possession length on offense and 361st on defense.
Listen and subscribe to Underdogs, hosted by Jordan Brenner and Peter Keating, everywhere you get your podcasts, including Apple and Spotify.
(Top photo: Tim Nwachukwu/Getty Images; Illustration: Will Tullos, The Athletic)
