statistical
Sample Size
Also known as: N, Sample Count
Sample size is the number of observations used to compute a statistic. Small samples produce noisy, unreliable estimates; large samples converge on the true underlying rate. Every statistical claim in betting must be evaluated against the sample size that produced it.
Sample size determines how much you should trust a statistic. The standard error of any proportion estimate is roughly sqrt(p × (1-p) / n), so the noise around an observed rate scales inversely with the square root of the sample. Doubling the sample halves the standard error by about 30%.
Stabilization points vary by statistic. In the NBA, three-point percentage stabilizes around 750 attempts (most of a season). Field goal percentage stabilizes faster, around 200 attempts. Free throw percentage stabilizes around 300 attempts. Usage rate and pace stabilize within 20-30 games. Most counting stats (points, rebounds, assists) stabilize in 30-40 games of regular minutes. Anything based on fewer than the stabilization threshold should be regressed toward a longer-window or league prior.
Worked example. A bettor sees a player averaging 8 rebounds per game across his first 5 starts (he was a bench player previously). The naive read: “He’s averaging 8 boards, the over 6.5 is a lock.” The sample size read: 5 games is nowhere near enough to update strongly off his prior. If his bench profile suggested 4.5 rebounds per game in those minutes, the regressed projection might be closer to 6 than to 8. The over 6.5 is far from a lock; it might even be a slight under.
Elite Picks reports both the raw recent-window stat and the regressed projection on every player. When sample size is small, the model explicitly shows how much shrinkage was applied so the user can see why the projection diverges from the box score average. Most recreational betting mistakes trace to ignoring sample size.
Example
A player going 6-for-10 from three (60%) across one game is meaningless. The same player going 60-for-100 (60%) across 25 games is a strong signal. Same percentage, drastically different sample size.