The Perfect Bracket Project: Upsets, part 2

Crazy as it may sound, we are within a month of Selection Sunday (aka the Greatest Day in the Sports Year). Given that I’ve promised to equip you with parameters that maximize the chance of filling out a perfect bracket by then, I better get rolling. Expect my posts to become more frequent (Lord willin’ and the creek don’ rise).

In my previous post I took you on an exploration of the NCAA Tournament upset. We examined minor upsets, major upsets, and bracket busters. We saw that the average tournament features about 18 total upsets, but that average is on the rise (8 of the past 10 tourneys have yielded 19 or more). Our first few parameters are taking shape, but let’s dig a little deeper and see when in a typical tournament upsets are most likely to happen.

We’ll start, unsurprisingly, with a chart:

You might have heard the narrative that the first weekend of the tournament is where chaos reigns, then the cream rises to the top and things unfold mostly in line with expectations on the second and third weekend. A cursory glance at this chart would lead you to assume that narrative is true; the first and second rounds do feature the most upsets, and the frequency continues to decrease round by round.

But that’s not the whole story. Of course the early rounds have more upsets; they have more games. The first round, in fact, contains over half of a tournament’s games (32 of 63), so you’d expect at least half of a tournament’s upsets to happen here. Maybe more than half, if the aforementioned narrative is true.

However, the first round upset average is 8.36, while the total tournament average is 17.82. So 47% of a tournament’s upsets happen in the round where 51% of the games are played. That statistical difference, albeit slight, runs in the opposite direction of our narrative. Even when we add in the second round, filling out that chaotic first weekend, we get an average of 12.98 upsets, or 73% of the total average after 76% of the games have finished. Again, a small statistical difference, but the narrative is fairly well-countered.

The next chart offers a different look:

Here is where we see something surprising: the Elite Eight has, by quite a margin, the highest rate of upsets. The other rounds, minus the national final, have consistent upset rates, ranging in the 24-30% range. So the question must be asked: why is nearly every other Elite 8 game an upset, at least by seeding?

The answer may be due to simple bracket dynamics: we tend to see closely-seeded teams match up in this round (1’s vs. 2’s; 3’s vs. 4’s, etc.). Such matchups are coin flips. Let’s dig in a bit to see if that’s truly the x-factor, though:

Some initial observations from this chart:

·      One-seeds are a modest 62-41 in the Elite 8 (60% winning percentage), but they make their hay vs. 6, 7, and 10 seeds (a 17-3 record).

·      Amazingly, 11 seeds have more wins against 1-seeds (4) than 6, 7, and 10 seeds do combined (3).

·      Again on those amazing 11-seeds: they have an identical winning percentage in Elite 8 games as do 1-seeds (6-4, or 60%).

·      The lone 8 vs 6 and 8 vs. 7 matchups happened in the same tournament (2000). The 8 seed won both games (North Carolina over Tulsa, Wisconsin over Purdue).

So, about that coin-flip matchup assumption: in Elite 8 games that feature teams only 1 seed line apart, the higher seed (a 2 vs. a 1, for example), has a slight edge at 29-28. When there is a difference of 2 seed lines, favorites are 25-18. Combined, “upsets,” or games won by the higher seed in these close matchups, occur 46% of the time. This is only slightly higher than our overall upset rate of 43.5%, and arguably within a statistical error range.

Since these matchups constitute well over half of Elite 8 games (100 of 156 games in total), this probably does explain the high frequency. It should be noted, though, that those pesky 11-seeds have engineered 4 bracket buster upsets (seed difference of 10 or more) that contribute to that percentage in a small but not insignificant way.

Before closing the door on our detailed look at upsets, I want to make one more deep dive. We’ve seen what kind of upsets there are, and how frequently each happens, on average (minor, major, bracket buster). We’ve seen how many the average tournament delivers, in total and per round. Now let’s put these together and see what we can learn from plotting the per-round averages of each upset type.

The data really shows us something here. By strict averages, we see 9.51 minor upsets, 7.28 major upsets, and 1.05 bracket busters annually. Our prior charts showed that upsets tend to be scattered fairly evenly throughout the tournament, with the Elite 8 being especially frisky. But these charts reveal that the different types of upset scatter quite distinctly.

Major upsets and bracket busters overwhelmingly cluster into the first 2 rounds. In fact, in 39 tournaments there have only been a total of 30 majors and bracket busters in the sweet 16 and beyond—less than one per tournament. Over 90% of all major upsets happen in the first 2 rounds. This is evidence that actually backs up the narrative of early round chaos and late round calm we pointed to above.

Minor upsets, meanwhile, pepper the entirety of the tournament. By percentages:

·      38.5% of all minor upsets happen in round 1

·      18.5% happen in round 2

·      22.6% happen in the sweet 16

·      14.8% happen in the elite 8

·      4% happen in the national semifinals

·      1.3% happen in the final

Perhaps it shouldn’t be surprising that upsets take on this pattern. Rounds one and two not only offer more games, but more built-in opportunities for majors and bracket busters. Each regional bracket automatically contains 3 possible major upsets (6 v 11, 5 v 12, 4 v 13) and 3 potential bracket busters (3 v 14, 2 v 15, 1 v 16), with only 2 natural minor upset opportunities (7 v 10, 8 v 9). Meanwhile, even if expected results hold and the second round is filled with chalk matchups, 2 of these 4 hold major upset possibility (1 v 8, 2 v 7). And first round upsets generally create major upset potential in the following round; one upset in a sub-regional bracket quadrant means that the seed differential for the next game will be 8 (a 3 v 11 or 2 v 10 matchup, for instance). A chalk bracket yields no such major upset possibilities in the sweet 16 and beyond. Major upset opportunities in these later rounds are only born out of some degree of first weekend chaos, and even then they aren’t guaranteed. When a bracket truly blows up you might see a sweet 16 featuring a 9 v 5 and 6 v 10 matchup—wild, sure, but a higher seed win in both cases would fall into the minor upset category.

In fact, when a high seed (6 or above) advances to the final 4, it usually benefits from a domino effect of upsets that give it at least one favorable matchup (a minor upset opportunity or even a situation in which they are favored) along the way. This, for example, is what happened for 9-seed Florida Atlantic in 2023 (drawing 16-seed Fairleigh Dickinson in round 2) and 11-seed NC State last year (drawing 14-seed Oakland in round 2 and 4-seed Duke in the regional final, though that still qualified as a major upset).

Ok. Now that we’ve exhausted our analysis of the upset, we’re ready for our first few Perfect Bracket Parameters™:

PARAMETER 1: Predict between 16 and 22 total upsets.

We can add to this Parameter 1a—Upsets should be divided into the following ranges:

·      7-12 Minor Upsets

·      6-11 Major Upsets

·      At least 1 Bracket Buster (but no more than 3)

PARAMETER 2: Don’t predict more than 2 Major Upsets in the Sweet 16 and beyond.

And Parameter 2a: But be sure to scatter minor upsets throughout the tournament.

PARAMETER 3: Pick at least one Elite 8 upset.

These parameters reflect the tournament’s major trends, though it should be noted that occasional outliers do happen. We’ve had tournaments without an Elite 8 upset before (but only 4 out of 39), and had tournaments with more or less total upsets (as well as upset types). But these parameters will get us in the ballpark of a typical tournament. And getting in the ballpark is all we can ask for. Remember, we’re trying to harness Madness here, aiming to select the exact right page from a Texas-sized book instead of an Earth-sized one.

But these aren’t our only parameters. Next up will be a look at a data point I call the Madness Factor, which will help us understand when a tournament is just a little mad and when it’s truly insane, and which will help us be even more thoughtful in scattering the upsets across our brackets.

Explore the charts from this post, courtesy of Infogram.