Variance and Sample Size
Last updated September 8, 2026
Every number on this site is computed over a small window, and small windows are noisy. Understanding roughly how noisy is the difference between using these figures well and being led around by them.
The counterintuitive part: the sport with the most games — baseball — produces the least reliable short-run numbers, because its per-game outcomes are the most random. Volume of data and reliability of a ten-game sample are not the same thing.
Why the same player swings between windows
A player whose true rate is somewhere near even will, over five games, quite often go 4-1 or 1-4 without anything having changed. Those are 80% and 20% hit rates describing identical form. Nothing is wrong with the data; that is simply what a five-sample draw from a near-even process looks like.
This is why we show three windows rather than one. Each is a different question, and the disagreement between them is more informative than any single figure.
Telling a real change from noise
The useful signal is not a high number — it is a persistent gap between the short and long windows that has an identifiable cause.
A player whose five-game rate sits well above his fourteen-game rate has either had a lucky stretch, or something structural changed: a teammate got injured and usage shifted, a rotation changed, he came back from injury, he was traded. The first is noise and will fade. The second is real and may persist. The game log almost always distinguishes them, because a real change has a start date and noise does not.
•
Noise: good and bad games interleaved, no obvious boundary.
•
Real change: a clear point in the log where the level shifts and stays shifted.
•
Check what happened around that date before assuming either.
Variance differs enormously by sport
Basketball counting stats accumulate over dozens of possessions per game, so they average out within a single night and produce comparatively stable numbers. Baseball gives a hitter three or four chances, and a good hitter fails most of them — so a single game says almost nothing.
Hockey and soccer sit at the noisy end for scoring markets, because goals are rare, while their volume markets — shots on goal, shots on target — behave far more like basketball. MMA is its own category, with a handful of fights spread across years.
Each sport guide on this site goes into its own variance profile, because the right amount of scepticism genuinely differs between them.
The gambler's fallacy, briefly
A player who has missed five in a row is not "due". Independent events have no memory, and a cold streak carries no corrective force. The same applies in the other direction to a hot streak.
Where a streak does mean something, it is because it reflects an underlying change — a new role, an injury, a different opponent quality — not because the sequence itself creates pressure toward a reversal. Look for the cause, not the pattern.