What the term means
Regression to the mean is the tendency for an extreme result to be followed, on average, by a result closer to the typical level. A very high score tends to be followed by a lower one; a very low score by a higher one. Nothing pulls the results back. The pattern comes from selection: when you pick out the extreme cases, you are partly picking out cases that had unusually good or bad luck, and luck does not repeat on request.
Results are skill plus luck
Think of any measured result as two parts: something stable (skill, or the underlying chance) and something that varies from one attempt to the next (luck). When a result is very high, it is more likely than usual to contain a helping of good luck. The next attempt starts with the stable part again but draws fresh luck, which on average is neutral. So the next result is expected to be lower.
The more luck matters in a game, the stronger the effect. In a pure game of chance the stable part is the same for everyone, so even the best performer in one round has no reason to do better than the rest in the next. In games with real skill, regression is milder but still present.
A coin-flip demonstration
Imagine 100 people each flip a fair coin 10 times and count heads. The chance of getting 9 or 10 heads in 10 flips is 11 out of 1,024, since there are 10 ways to get exactly 9 heads and one way to get 10. That is about 1.1 percent. So among 100 people you would expect about one to produce a result that impressive.
Now give that top performer another 10 flips. Their expected number of heads is 5, the same as everyone else. Their first round was not a sign of a special ability; it was the extreme end of ordinary variation.
An observer who only watches the top scorer would see a drop from about 9 to about 5 and might blame pressure or complacency. The drop needs no explanation beyond arithmetic.
Regression is not a debt
It is easy to confuse this with the idea that the game must even things out. The two are different.
Regression says that the future is expected to look average. It does not say the future will compensate for the past. After 9 heads in 10 flips, the next 10 flips are still expected to give 5 heads, not 1. Over the combined 20 flips the expected total is 14 heads, which is 70 percent, still above one half, because the first batch is not erased. It is only diluted as more flips accumulate.
The gambler’s fallacy is the belief that a streak makes the opposite outcome more likely next time. In independent events such as coin flips, dice or roulette spins, it does not. Closely related is hot hand belief, the opposite error of expecting a streak to continue. For independent chance events neither expectation is supported.
Where it shows up
- Hot starts and cold spells. A player who wins several sessions in a row will usually look less impressive afterwards, and one who loses several will usually recover toward their normal level, without any change in ability.
- After a change. If someone tries a new approach right after a terrible stretch, the following improvement may be mostly regression. The new approach gets credit it has not earned. The reverse also happens: a tweak made after a hot run can seem to cause the decline.
- Rankings and records. Leaders in a short league table or a leaderboard season often look weaker the following period. This applies to claims about esports statistics as much as to casual games.
- Poker. Over short stretches, poker variance can produce winning and losing runs that say little about the quality of decisions. Results tend to drift toward the player’s true level over many hands.
How to avoid being fooled
- Ask how much of the result could be luck. The more random the game, the less a single extreme result tells you.
- Judge by larger samples. The sample size behind an average decides how much it should move your beliefs.
- Compare with a baseline. When evaluating a change, compare against a similar group that did not change, rather than against the person’s own worst or best period.
- Be wary of “before and after” stories that start from an extreme moment.
A related warning: in loot and reward systems, a long unlucky run is not evidence that something good is due, even though the typical result will improve simply because extremes do not persist. Chasing losses on the belief that an upswing must come is a well-known trap with no mathematical support.
A balanced view
Regression to the mean does not say that skill is imaginary or that every streak is luck. Real differences between players exist and do persist. The point is narrower: extreme results are a poor guide to the underlying level, and the more extreme the result, the more reason to expect a less extreme one next.
Games with a negative expectation do not become fair because your results regressed upward for a while. Outcomes are uncertain, and if gambling is causing you difficulty, a local support service can help.