Which Stats Stabilize First

A statistic stabilizes quickly when most of its variation comes from the player and slowly when most of it comes from circumstance.

Strikeout rate for a pitcher is largely about the pitcher, so it becomes informative fast. Batting average on balls in play depends heavily on where fielders happen to be standing, so it takes an enormous sample before it says anything about the hitter.

Within a sport, the ordering is more stable than any exact threshold:

• Fastest: rates that reflect a repeatable action — strikeout rate, walk rate, three-point attempt rate, snap share.
• Middle: contact and quality-of-contact measures, shooting percentages.
• Slowest: outcome rates that depend on other people — batting average, batting average on balls in play, team-dependent counting stats, anything involving touchdowns.

The general rule: the more people involved in producing the number, the longer it takes to mean anything about any one of them.

Do not discard it. Regress it, hard.

A slow-stabilizing statistic in a small sample should be shaded strongly toward the player's own prior — his career rate, or the population average for his role. As the sample grows, the weight on the observed number rises and the weight on the prior falls.

That is precisely the operation our NFL Draft Kit baseline performs: a player with four games gets pulled hard toward the positional mean, and a player with seventeen barely moves.