All scores →
Loading…
Loading…
Loading…

Why upsets happen: the maths and the myths

By 4Basketball Staff · · 13 min read

The arithmetic behind NCAA tournament upsets, from single-game variance and compounding odds to the extra spread that three-pointers and slow tempo create, plus the popular explanations that do not hold up.

Every March the same ritual plays out. A team that was supposed to win by fifteen gets beaten by a team nobody had on their bracket, and within an hour the explanations arrive. The favorite was overconfident. The underdog was hungry. The star had a bad night, the referees let them play, the other coach out-schemed his rival. Some of these explanations are true in a particular game. Almost none of them is the main reason upsets happen, because the main reason is simpler and less dramatic: basketball games are decided by a modest number of events, and a modest number of events produces a lot of randomness.

That randomness is not a flaw in the tournament. It is the tournament. A single-elimination bracket takes a sport whose games have a real but limited amount of chance in them, strips away the repetition that normally smooths it out, and asks sixty-eight teams to survive a chain of independent trials. The upsets that result are not malfunctions. They are what the arithmetic predicts, and a bracket with no upsets in it would be the strange result.

This article works through that arithmetic in plain terms, with a few numbers you can check on a calculator. Then it goes through some of the popular explanations for upsets and asks which ones hold up. The aim is not to drain the fun from the event. Understanding why 12 seeds beat 5 seeds makes the next one more interesting, and it makes your bracket better.

What counts as an upset

Start with a definition, because it is slipperier than it looks. The usual meaning is simple: a team with a higher seed number beats a team with a lower seed number. A 12 beating a 5 is an upset, and a 9 beating an 8 is, technically, one too.

But seeds are rankings, and a ranking says nothing about how big the gap is. The difference in quality between a 1 seed and a 16 seed is enormous. The difference between an 8 seed and a 9 seed is close to nothing, which is why we treat that game as a coin flip. Between those extremes the gaps change from line to line, and they change from year to year depending on who is in the field. A better definition of an upset is a result that went against the ratings: a team that was rated clearly lower beating a team that was rated clearly higher. Seeds are a convenient shorthand for that, but they are only a shorthand. We explain how the committee produces them in how the bracket is built.

This matters because a lot of the confusion about upsets comes from treating the seed difference as if it measured the strength difference. When you compare teams on a points scale, such as the point spread implied by efficiency ratings, the picture is clearer. We look at those ratings in college efficiency metrics. The rest of this article uses the margin of victory as the measure of the gap between teams, because it connects directly to the math.

The core idea: a game is a noisy measurement

Imagine two teams, and suppose we know exactly how good each is. A good rating system will tell us the expected margin: perhaps Team A is better by six points on a neutral court. If the teams played a thousand games, Team A would win by an average of six. But they play once. In any one game the actual margin will be scattered around the expected margin, sometimes much bigger, sometimes smaller, sometimes in the opposite direction.

How scattered? A reasonable rule of thumb is that the standard deviation of a college basketball game's final margin around its expected value is somewhere around ten or eleven points. That means that in about two games out of three the result will land within eleven points of the expectation, and in the other third it will be further away. The exact figure depends on the sample and the era, but the order of magnitude is well established.

With that one number you can estimate the chance of an upset for any gap. Assume a standard deviation of 11 points and a bell-shaped distribution, which is a good approximation. The favorite wins when the actual margin is above zero. The table below shows the rough chance that a favorite wins at various expected margins.

Expected margin for the favorite Chance the favorite wins (roughly)
3 points 61 percent
6 points 71 percent
10 points 82 percent
15 points 91 percent
20 points 96 percent
25 points 99 percent

These are approximations built on one assumption, and the real world is messier. But they line up well with what the tournament has produced. A 5 seed against a 12 seed is typically a favorite by a margin in the middle of the table, which corresponds to the roughly two-in-three record of 5 seeds. The history of 1 seeds against 16 seeds, which is about two losses in roughly 160 games, is consistent with a gap of 20 to 25 points. The numbers are not a coincidence. They are the same arithmetic.

The key insight is how flat the middle of that table is. A favorite by six points loses nearly three games in ten. A favorite by ten loses nearly two in ten. A team does not have to be close in quality to the opponent to have a real chance. It only has to be within a couple of dozen points, which almost every team in the field is.

Why basketball is both predictable and not

The sport sits in an interesting spot. It is more predictable than baseball, where the best team in a season loses a large share of its games, or football, where a season is only a few games long and a single bounce can swing a result. A good basketball team wins a high percentage of its games, because there are many possessions in a game and the better team converts its edge over a long sequence. We touch on why possessions are the unit that matters in points per possession.

But the tournament is not a season. A season is a long series of trials, in which luck evens out. The tournament is a short sequence in which any one trial can eliminate you. A team that is excellent over thirty-five games can still lose a single forty-minute contest, and in a tournament that single contest is the whole story.

There is a simple way to see how much a longer format helps the better team. Suppose the better team has a 70 percent chance in each game. In a single game it wins 70 percent of the time. In a best-of-seven series, where it needs four wins, it wins about 87 percent of the time. If the better team has a 60 percent chance per game, a seven-game series raises it to about 71 percent. A series filters out chance by repeating the test. The NBA playoffs, which we describe in how the NBA playoffs work, use that filter, and the best team usually wins. The NCAA tournament deliberately does not use it.

The compounding problem

The second piece of arithmetic is how win probabilities multiply. A team that must win six games in a row to take the title needs to beat all six opponents. If its chances in each game were independent, the probability of winning all six is the product of the individual chances.

Chance of winning each game Chance of winning six in a row
70 percent about 12 percent
80 percent about 26 percent
90 percent about 53 percent

In practice, opponents get tougher as the tournament goes on, so a top team might be a 95 percent favorite in the first round and a 65 percent favorite in the Elite Eight. But the lesson holds. Even a team that is a large favorite in every game it plays has a fair chance of losing somewhere along the way. A team that is the best in the field, but only by a modest margin over the second-best, is unlikely to win the whole thing. The surprising thing about a tournament is not that favorites lose. It is that any favorite ever wins six in a row.

The same compounding works in reverse when you look for at least one upset. If each of four 12 seeds has about a one-in-three chance of beating its 5 seed, the chance that none of them wins is about 0.65 to the fourth power, which is roughly 18 percent. So in four years out of five, at least one 12 should win. That is why the pattern feels reliable even though no single 12 seed is likely to win. We follow the history of that pattern in the 12-5 upset.

The shape of the noise: threes, tempo and the smaller sample

Not all games carry the same amount of noise. Two features of how a team plays can widen or narrow the scatter, and both are worth understanding because they connect to the styles of many famous upset teams.

Three-pointers add variance

Take a team that shoots threes at 35 percent and another that shoots twos at 52.5 percent. Each shot is worth the same on average: 1.05 points. But the spread of results is very different. A two-pointer that goes in 52.5 percent of the time has a variance of about one point squared per attempt. A three-pointer at 35 percent has a variance of about two points squared per attempt, roughly double. A team that takes many threes is therefore a higher-variance team: its good nights are very good and its bad nights are very bad.

For a favorite, higher variance is a risk. For an underdog, it is a lifeline. A weaker team that cannot beat a stronger one by out-executing it on average can sometimes beat it by shooting well on a night when the stronger team shoots poorly. That helps explain why so many Cinderella teams are built around guards and shooters, a pattern we describe in Cinderellas: the greatest runs by double-digit seeds. The broader change in the sport is described in the three-point revolution in the numbers, and one effect of that revolution is that the tournament may be slightly more random than it was in the era of more two-point shots.

Fewer possessions give the underdog more room

The second effect is tempo. Suppose a favorite is better than the underdog by a fixed amount on every possession. Over a game with more possessions, the favorite's advantage accumulates in proportion to the number of possessions, while the noise grows only with its square root. A longer game therefore lets the edge outgrow the noise. A shorter game does the reverse.

You can make this concrete. A game with 75 possessions per team gives the favorite more time to express its edge than one with 60 possessions. The ratio of edge to noise is proportional to the square root of the number of possessions, so the slower game cuts the favorite's effective edge by roughly 11 percent. That is not huge, but it is enough to move a 71 percent favorite a couple of points closer to a coin flip. It explains a piece of folk wisdom that holds up: teams that slow the game down, whether through a deliberate offense or a patient defense, are making the contest more random, and an underdog usually wants a more random contest.

Uncertainty about the teams themselves

There is a third source of noise that is easy to overlook. All of the above assumes we know exactly how good each team is. We do not. Our estimate of a team's strength is based on its results, and those results come from a schedule. A team that went 30-3 in a small conference has played mostly weaker opponents, so its true strength is harder to pin down than that of a team that played a dozen games against top opposition. A 12 seed from a small league has a wider range of plausible true strengths than a 5 seed from a major conference. That extra uncertainty widens the odds of an upset, because the underdog might be better than its rating says. It is also why the committee's judgments on Selection Sunday, described in Selection Sunday: how the committee works, are inherently imperfect: they are estimates built from uneven schedules.

The myths

With the arithmetic in place, we can look at popular explanations for upsets. Some of them have a kernel of truth, and some do not. The test is simple: does the explanation add anything beyond the variance already in the game?

"They wanted it more"

The hunger narrative is the most common and the least useful. Every team in the tournament has spent a year getting there, and the players on a 1 seed are just as motivated as those on a 16 seed. A game's outcome is far more strongly linked to shooting percentage, turnovers and rebounding than to any difference in desire. Hunger sounds like an explanation, but it is a label we put on the result after the fact.

"Momentum" and "hot teams"

Teams that win their conference tournaments are often said to carry momentum into the bracket. There may be a small effect from playing recent competitive games, and from confidence, but it is hard to separate from the simple fact that teams that win tournaments are good. The claim that momentum overrides the ratings is not well supported, and the sheer number of hot teams that lose in the first round suggests that it is not a strong force.

"Experience wins in March"

The idea that veteran teams beat young ones has a reasonable basis, because veteran players tend to commit fewer turnovers and execute sets well. It is also a feature of the profile of many mid-major upset teams. But experience is correlated with quality in many other ways, and the effect is probably smaller than the stories suggest. The truth is that experience is one input among many.

"Seeds in the middle are underseeded"

The pattern that 5 seeds lose to 12 seeds roughly a third of the time has led some to conclude that the 12 seed is systematically underrated. The arithmetic suggests another explanation: the 5-12 matchup has a smaller gap than the seed numbers imply, and the uncertainty about 12 seeds is greater. There is no need for a conspiracy, and no guarantee that the pattern will continue at the same rate.

"The best team always wins in the end"

This is the opposite myth, and it deserves equal skepticism. As the compounding table shows, a team can be a large favorite in every game and still be unlikely to win the whole tournament. The fact that top seeds win more titles than anyone else is a measure of their edge, not of certainty. See the history of the one-seed for the evidence.

"Upsets are getting more common"

Fans often say that parity has increased and that the gap between big programs and small ones is shrinking. There are reasons to think the landscape has changed, from the transfer portal to the rise of analytics at smaller schools, and our look at analytics myths cautions against reading too much into any single measure. But the sample is small: a few dozen games per seed matchup over four decades. A short run of upsets is within the range of normal variation.

Why upsets feel more common than they are

One last effect is psychological. Upsets are vivid and favorites winning is forgettable. People remember the 15 seed that beat a 2 and forget the thirty other 15 seeds that lost in the same period. A single memorable buzzer-beater can shape how fans think about an entire seed line. That bias, combined with the heavy media attention on upset stories, leads many fans to overestimate how often they happen.

The corrective is to look at the count. In the 12-5 matchup, the underdog wins about once in three. In the 15-2 matchup, it has won about a dozen times in four decades. In the 16-1 matchup, it has won twice. When you fill out a bracket, those frequencies are the guide, not the stories.

What to take away

Upsets happen because a single basketball game is a noisy measurement, and the tournament asks teams to survive a long chain of those measurements. A favorite by six points loses almost three times in ten. A favorite by ten loses nearly two in ten. A team that wins each game with 80 percent probability wins six in a row about a quarter of the time. Add the extra variance from three-pointers, the extra room that slower tempo gives an underdog, and the uncertainty in our knowledge of the teams, and the surprise is not that upsets happen but that they do not happen more.

For anyone filling out a bracket, the practical advice is to think in probabilities rather than stories. Look at the size of the gap between teams, not just the seed numbers. Expect a few upsets in the places where the gap is smallest, and do not pay for them in the places where it is largest. We turn that thinking into specific tactics in bracket-picking strategy. The best brackets are built by people who accept that chaos is part of the design.

Search more on this

Keep reading

Tournament expansion: from 8 teams to 68

The NCAA tournament began with eight teams in 1939 and now invites 68. This is the story of every major expansion, why each one happened, what it changed in the bracket, and how the women's tournament grew alongside it.

4Basketball Staff · Aug 19, 2026 · 11 min read

The 12-5 upset: history, odds and the pattern

Why 12 seeds beat 5 seeds about a third of the time, how the matchup became the most famous first-round upset, which 12 seeds are worth picking, and how the odds work out across four games.

4Basketball Staff · Aug 8, 2026 · 11 min read