The pattern
A striker scores far more than usual in one season, then scores fewer the next. A rookie has an outstanding first year and a more modest second. A team wins far more than expected and then slips back. Commentators often call these “slumps”, “second-season syndrome” or “the pressure of expectations”. Sometimes those explanations are right. But a simpler statistical effect is often at work, called regression to the mean.
The mean here means the typical level, and regression means moving back towards it. The idea is that an extreme result, high or low, tends to be followed by one closer to the typical level.
Why it happens
Any performance is a mixture of two things: some underlying ability, and some luck on the day or over the season. Luck includes bounces, refereeing decisions, opponents, injuries to others, and how chances happened to fall.
When someone has an exceptional season, they are likely to have both good ability and unusually good luck. The ability is likely to persist. The luck, by its nature, does not. So the next season probably has the same ability with luck closer to average, which means results closer to average. Nobody got worse. The luck component just stopped contributing so much.
The same applies in reverse. A terrible season often includes bad luck, and the following season often looks better for no special reason.
A coin example
Imagine 100 people each flipping a fair coin ten times. Getting eight or more heads can happen by chance alone. The number of ways to get exactly 8, 9 or 10 heads in ten flips is 45, 10 and 1, which makes 56 out of 1,024 equally likely sequences. That is about 5.5%, so we would expect around five or six of the 100 people to hit eight heads or more.
Now ask those people to flip again. There is no reason for them to do better than anyone else, because the coin has no memory. Their expected result is five heads, the same as everyone. Compared with their first round, they have “regressed” sharply, and yet nothing about them changed.
Real sport is not a pure coin toss, because skill matters and varies between players. But the same logic holds in proportion. The more luck is involved in a measurement, the more strongly extreme results regress.
Where to expect it
Regression is strongest when:
- the measure is noisy, such as goals scored, or shots made over a short stretch;
- the sample is small, such as a handful of matches, which links to small sample sizes;
- the starting result is extreme, because the further out a result is, the more of it is likely to be luck.
It is weaker for stable, skill-dominated measures, where results from year to year are more consistent.
Common misreadings
Blaming a curse or pressure. The “cover athlete curse” and “second-year slump” are often partly regression. These stories feel convincing because extreme performances are what get attention in the first place.
Crediting a change. If a team is struggling, replaces its coach, and then improves, it is tempting to credit the coach. But a team picked out for being unusually poor would be expected to improve somewhat anyway. Without comparing against similar teams that did not make a change, it is hard to tell the difference. This overlaps with correlation versus causation, because a change that coincides with a recovery is not necessarily its cause.
Treating streaks as forecasts. Regression is not the same as the gambler’s fallacy. Regression does not say that bad luck must be followed by good luck to compensate. It says only that future results tend to be closer to the typical level, because nothing is pushing them to stay at the extreme. The law of averages is a loose label that often confuses the two.
Confusing it with the law of large numbers. The law of large numbers says averages stabilise as more results are collected. Regression to the mean is the reason a single extreme observation tends to be followed by a less extreme one. They are related but not identical.
What to do with the idea
When you see an exceptional number, ask how much of it might be luck. Check how many games or events it covers. Look at longer records, and at related measures that are less noisy. For forecasting, a sensible estimate usually sits between the player’s recent extreme result and their long-run record, leaning more towards the long-run record when the recent sample is small.
It also helps to resist narrative explanations that go beyond the data. A player whose form drops after a great run has not necessarily lost anything. And a performer who returns from a bad run has not necessarily fixed something. Often the numbers are doing what the numbers do.
Regression to the mean does not make skill unimportant. It reminds us that outcomes are shaped by ability and chance together, and that the best guide to ability is usually a long record rather than the most striking recent result.