New to these reports? Start here
- Dotted-underlined words have a plain-English definition — hover or tap them. Every term is also on the glossary page.
- "Null" means we found nothing, not that something broke. Most reports here are negative results, on purpose — knowing an idea doesn't work is the point.
- Two questions get asked separately. First, is the effect real? Second, is it already priced into the betting odds? An effect can be completely real and still useless to bet on.
- A "calibration" row is a self-check. It runs the same method on something already known to be true. If that fails, the whole report is unreliable — so it's reported alongside the findings.
- If a confidence interval includes zero, the real effect might be nothing at all, so no claim gets made.
What injuries and suspensions actually cost (NFL)
2026-09-08. 2,176 team-weeks, 2022-2025, from nflverse weekly rosters joined to snap counts and closing lines.
Measuring absence
Players out are taken from roster status, and the two channels are kept apart:
- injured —
R01,R48,R04,R05. ⚠ A first pass used only R01/R48 and missed 1,837 player-weeks: R04 and R05 are further injured-reserve variants (Tank Dell 2025, Nyheim Hines 2023, Andrew Vorhees, Tyus Bowser).R02is deliberately excluded — it is Jason Kelce and Fletcher Cox in 2024, retirements parked on a reserve list. Correcting this moved the margin coefficient −1.46 → −1.39 and the line coefficient −0.51 → −0.50; no conclusion changed, because undercounting attenuates toward zero. - suspended —
R40, ⭐ verified rather than assumed. The longestR40spells are Deshaun Watson 2022 (11 weeks), Cameron Sutton 2024 (8), Grover Stewart 2023 (6, PED), Kareem Jackson 2023 and Tashaun Gipson 2024 — all league suspensions, no injuries among them.
Absence is weighted by snap share, not counted. Raw games-missed is dominated by players nobody plays: a team is missing 9.0 players in an average week but only ~2.6 full-time-equivalents.
Suspensions were meant to be the cleaner experiment, since the league sets the timing rather than the team's form. They are too rare to use: 212 team-weeks with any suspension, 104 with a meaningful one, and the coefficient spans zero in every specification below. That question is unanswerable at this sample size, not answered in the negative.
Injuries cost about 1.5 points per full-time player
Regressing a team's game margin on its net snap-weighted burden (own minus opponent's):
| estimate | 95% CI95% confidence intervalThe range the true value is plausibly in. If this range includes zero, we cannot rule out that the real effect is nothing at all. | ||
|---|---|---|---|
| net injury burden | −1.39 | [−1.66, −1.11] | excludes zero |
| net suspension burden | +1.43 | [−0.94, +3.94] | spans zero |
So one full-time-equivalent player more injured than your opponent is worth about a point and a half of margin. That much is unsurprising and is the uninteresting half of the question.
The market prices them — outside the last four weeks
The question that matters is whether burden predicts beating the closing line, which already knows who is out. Pooled, it looks like it does: −0.50 [−0.73, −0.26], excluding zero.
⚠ That pooled number is misleading. Split by when in the season:
| n | effect vs closing lineclosing lineThe final odds right before a game starts. It reflects everything the betting market knows, which makes it the hardest benchmark to beat. | ||
|---|---|---|---|
| weeks 1-9 | 876 | +0.06 | spans zero |
| weeks 10-14 | 462 | −0.34 | spans zero |
| weeks 15-18 | 478 | −0.70 | excludes zero |
The whole effect is in the last four weeks; weeks 1-9 sit on zero. The natural reading is not that the market misprices injuries — it is that by week 15 an injury burden marks a team whose season is over, which shuts players down and stops trying. That is a motivation story, and burden is its symptom rather than its cause.
Splitting burden by whether it was already on the books a week earlier points the same way: burden that appeared THIS week carries a bigger coefficient (−0.88) than burden knowable a week before (−0.36), which is what biasbiasWhether the misses lean consistently one way. A projection can have a good average error size but still be biased if it is almost always too high. Bias is often the more fixable problem." data-def="Accidentally using information that was not available at the time. It makes predictions look brilliant and is the single most common way a backtest fools you.">look-aheadlook-ahead biasAccidentally using information that was not available at the time. It makes predictions look brilliant and is the single most common way a backtest fools you. looks like — a player placed on IR after being hurt in the game being scored.
Conclusion: injuries are priced. Nothing here supports betting them, and the two ways it appeared to are a late-season motivation artifact and a within-week timing artifact.
The retrospective, which is what was asked
Mean snap-weighted burden and performance against the closing line, per team-season (128 team-seasons with >= 14 games):
| team | season | burden | margin | vs line |
|---|---|---|---|---|
| LA | 2022 | 6.67 | −4.53 | −2.85 |
| CAR | 2023 | 5.87 | −10.59 | −5.53 |
| ARI | 2025 | 5.84 | −7.82 | −4.12 |
| DEN | 2022 | 5.50 | −4.24 | −3.65 |
| CAR | 2024 | 5.19 | −11.35 | −4.50 |
| HOU | 2023 | 5.07 | +1.41 | +2.21 |
| CLE | 2024 | 4.84 | −10.41 | −6.32 |
Houston 2023 is the standout: sixth-most injured team-season in the sample and still beat its closing line by two points a game.
The 49ers:
| season | burden | rank | margin | vs line |
|---|---|---|---|---|
| 2022 | 2.38 | 18th | +10.18 | +4.53 |
| 2023 | 2.02 | 18th | +11.35 | +3.24 |
| 2024 | 4.34 | 3rd | −2.76 | −5.24 |
| 2025 | 2.81 | 13th | +3.88 | +2.18 |
⚠ The premise of the question needs correcting: 2025 was not San Francisco's injury season — 2024 was. In 2025 they were 13th of 32 in burden, mid-pack, and beat their number by 2.2 points a game. The heavily-injured year was 2024, 3rd-most in the sample, and they underperformed by 5.2 a game — the worst line-relative season of the four despite the burden being public.
Caveats
- The snap-share weight uses the full season's snaps, so a player's importance is measured partly after the injury. It does not run in the direction of the findings, but it is not clean.
- Roster status is a weekly snapshot; a designation made after a game is not distinguishable from one made before it. That is what the known/new split above is trying to bound.
- 4 seasons. Team-season rows are 128, and the interesting tails are single seasons.