Every Friday the injury report lands and every Sunday somebody tells you a team is too banged up to cover. It is the most widely available piece of NFL information that is also treated as an edge, which should be the first clue.
Here is what it is worth, measured three ways on 4,509 games.
One: count the bodies
Take the difference between the two teams in players listed Out, and compare it with how the game finished against the closing spread. Each additional absent player is worth about 0.17 points, in the direction you would expect — the shorter-handed team does slightly worse. The t is about -1.80.
Suggestive, and not significant. It is also the weakest possible way to ask the question, because three special-teamers and a starting left tackle are not the same loss, and a count treats them identically. The average gap between two teams is only 1.61 players, so most of this variable is noise about depth players.
Two: weight them by how much they actually play
So weight each absent player by the share of his team's snaps he normally takes. A full-time starter counts as a whole unit; a rotational player as a fraction.
The effect roughly doubles, to about 0.34 points per starter-equivalent lost, across 3,485 games. The t moves to -1.97.
This is the result worth sitting with. A better-built instrument found a larger effect and still did not clear — and that is far more informative than the first flat answer, because the obvious objection to a negative result is "you measured it badly". Measured better, it is still not an edge.
Three: just place the bet
Numbers about correlations are easy to argue with, so here is the version that settles it. Back the healthier side whenever the weighted gap is at least the given size.
| Weighted gap | Games | Covers | Non-covers | Cover rate | Return per bet |
|---|---|---|---|---|---|
| 0.75 starters | 1,845 | 935 | 861 | 52.1% | -0.61% |
| 1.50 starters | 746 | 377 | 351 | 51.8% | -1.14% |
A bet at -110 needs 52.4% to break even. At roughly a full starter of difference the healthier side covers 52.1%, which is 0.27 standard errors from break-even and returns -0.61% per bet. Widening the filter to 1.50 starters does not help: 51.8% and -1.14%.
Counting bodies instead of snaps gives the same answer — at a gap of 3 or more players, 53.2% and +1.50% per bet.
Those returns are the house margin. Not a small edge, not a reversed edge: the margin, which is what a market that has already read the information looks like from the outside.
What this does and does not say
It does not say injuries do not matter. The snap-weighted coefficient is real and it is roughly 0.34 points per starter, which over a season is the difference between several wins and several losses.
It says the number already contains them. By Sunday the injury report has been public since Friday, every book has read it, and the line has moved. The information is real and it is not news.
The practical version, for anyone writing or reading a preview: the injury report explains why a line is where it is. It does not tell you the line is wrong.
Method
Absences are rows designated Out in nerddawg.nfl_injury_reports, by team and week, from 2009. Snap shares come from nerddawg.nfl_snap_counts, which begins in 2013 — which is why the weighted sample is smaller than the counted one. A player's workload is his mean share of his team's offensive or defensive snaps for that season, whichever is higher; special-teams snaps are excluded, because a full-time special-teamer is not a full-time contributor.
The two tables are joined on the league player id. Of 16,414 absences, 3,576 could not be matched to a snap record and were weighted zero rather than dropped, so a team is never credited with a lighter burden because somebody could not be identified — a match rate of 78.2%. The mean weighted burden is 0.95 starter-equivalents per team-week. Slopes are ordinary least squares against the against-the-spread margin. Pushes are excluded from cover rates rather than counted as half. The two weighted thresholds were fixed before the records were examined. Measured 2026-10-13 and recomputed whenever this page is rebuilt.
