One measure, in full · Usage: what holds
Carry share
Definition
One definition, the column that binds its range, and what the figure is a share of.
lag-1 autocorrelation of a player's weekly deviation from his own season mean; 0 means this week's surprise says nothing about next week's
usage_stability.within_lag1.carry_share
- Source
pbp; n counts independent players- Bound column
- the whole archive, 1999-2026; no input binds it
- Columns it reads
pbp.rusher_player_id|rush_attempt- Denominator
- The team's total: teammates split one denominator.
- Season range
- Current 1999–2026. Current. Its inputs cover 1999 to 2026 and reach the season in progress.
- Refreshed
- This file was generated Tue, Oct 27, 9:33 AM ET. It does not state a refresh schedule.
Distribution
No distribution: the file is one figure for the whole league, not a value per player.
Does it carry?
Yes: the interval sits wholly right of zero.
Week-to-week carry-over
r = 0.24
[0.22, 0.26] · n = 1,101 · 44,459 rows
Carries over- Carry sharethis measure0.24 [0.22, 0.26]
- Carriesfor contrast0.11 [0.09, 0.14]
- Receptionsfor contrast−0.04 [−0.05, −0.03]
Carry share, among usage stats
Carry share: r 0.24 week to week (interval 0.22 to 0.26, 1,101 players). A week above his own average carries into the next.
Whether a player's week above his own season average is followed by another: the correlation of each week's deviation from his own mean with the next week's. An interval wholly above zero carries; one that reaches zero, or sits under it, does not.
- Carries0.11
- Carry share0.24
- Reception share−0.04
- Receptions−0.04
- Target share−0.01
- Targets−0.02
analytics/nfl/usage_stability.within_lag1.carry_share.json · generated Tue, Oct 27, 9:33 AM ET
This stat, all three ways
| Measure | Estimate | Interval | n |
|---|---|---|---|
| Week to week | 0.24 | 0.22 to 0.26 | 1,101 players |
| Between players | 73% | 71% to 75% | 1,101 players |
| Naive | 0.80 | 0.79 to 0.81 | 1,101 players |
How it is measured
- Week to week
- lag-1 autocorrelation of a player's weekly deviation from his own season mean; 0 means this week's surprise says nothing about next week's
- Between players
- share of total variance sitting between player-seasons rather than within them; high means the metric separates players
- Naive
- pooled lag-1 correlation of raw weekly values across all players; high because good players are high every week, which is why it is published beside the demeaned figure rather than instead of it
- Interval
- Interval from resampling players with replacement, 2,000 times. Built at 95%.
- Sample
- n counts independent players, never weekly rows (44,459 weekly rows from those 1,101).
Leaderboard
No leaderboard: the file is one league-wide figure, not a value per player.
What the file does and does not say
Four limits, stated plainly.
- Interval from resampling players with replacement, 2,000 times. Built at 95%. That coverage is nominal: this page does not check it against outcomes.
- n counts 1,101 independent players, not the 44,459 rows behind them. Rows inside one player share a context, so counting them as the sample would draw the interval too narrow.
- It covers 1999 to 2026, and says why in the producer's words: the whole archive, 1999-2026; no input binds it.
- It is one figure for the whole league, pooled. It does not say what any one player does, and it does not split by position. A share of a team total: teammates split one denominator, so two teammates' intervals side by side are not two independent estimates.