Injury Rates: What Actually Predicts Missed Games

Every draft, some analyst insists high-volume backs are “due” for injury and light-workload wideouts are “safe.” So we tested it: across 2016 to 2025, does workload actually predict missed games at any position?

The short answer is no. This year's box score is a bad injury oracle. What does travel is the injury history a player already has.

Average injury-based games missed per regular season since 2016, normalized to a 17-game schedule (16-game seasons pre-2021 are prorated up):

QBs are the most durable position; RB, WR and TE all sit well above two games a year, with RB highest. A missed week counts as injury-based only when the injury report corroborates it: the player is on the report that week (any status — many real sit-outs are only ever logged “Doubtful” or “Questionable”), or pops on it the following week after getting hurt in that game. Mid-game exits count fractionally by quarters. Absences with no report adjacency — benchings, healthy scratches, role reductions — never count.

We fit negative-binomial regressions of games missed on every in-season volume, usage, and player-profile metric we track. Full list per position:

• QB: pass attempts/g, scrambles/g, designed runs/g, total rushes/g, sacks taken/g, QB hits/g, goal-line designed runs/g, deep-ball rate, shotgun rate, aDOT, age, years experience, height, weight, BMI, prior-year games missed, prior 3-year games missed, career dropbacks entering the season.
• RB: carries/g, targets/g, total touches/g, inside carries/g, outside carries/g, goal-to-go carries/g, short-yardage carries/g, offensive snaps/g, YAC per reception, age, years experience, height, weight, BMI, prior-year games missed, prior 3-year games missed, career touches (carries + targets) entering the season.
• WR: targets/g, receptions/g, middle-of-field targets/g, middle rate, deep rate, short rate, aDOT, YAC per reception, offensive snaps/g, age, years experience, height, weight, BMI, prior-year games missed, prior 3-year games missed, career touches entering the season.
• TE: targets/g, receptions/g, middle-of-field targets/g, middle rate, deep rate, aDOT, YAC per reception, offensive snaps/g, age, years experience, height, weight, BMI, prior-year games missed, prior 3-year games missed, career touches entering the season.

Exactly one workload metric holds up as a standalone signal:

• QB sacks / game: p10→p90 IRR of 1.90 (p = 0.003). A quarterback in the top decile of sack rate carries nearly double the expected missed games of one in the bottom decile.

A TE middle-of-field-targets effect that looks real in a naive sample (IRR ~1.9) collapses to noise (IRR 1.29, p = 0.10) once benchings and role reductions are excluded and tight ends who got hurt early stay in the sample — it was a survivorship artifact, not a durable signal.

The single largest predictor in the study is prior 3-year games missed. It clears p<0.0001 at every position (p10→p90 rate ratios of 2.1x at RB up to 2.6x at QB) and dominates every multivariate model we fit.

In practice: a player with 10 games missed over the last three seasons has roughly double the expected miss rate of a player with zero, at any position. It is boring, and it works.

When we throw everything into one model per position, career injury history still leads. A few workload effects survive:

• QB: sacks / g (in the joint model, sacks absorb the history signal)
• RB: injury history (outside carries barely marginal, nothing else moves)
• WR: injury history + career touches, plus a marginal aDOT lean
• TE: injury history + BMI (heavier tight ends miss more)

Per-position negative-binomial model, fit on 2019–2025 with prior 3-year games missed (total and severe), age, and QB sack rate as predictors. Applied to the FantasyPros top-150 ADP. Filter by name or position.

• Position baselines (17-game normalized, injury-based absences only): QB 2.06, RB 2.66, WR 2.43, TE 2.39. QBs are most durable; RB highest.
• RB carry volume does not predict missed games.
• QB sacks / game is the only workload signal that survives. A tempting TE middle-targets effect is a survivorship artifact.
• Prior 3-year games missed is the strongest, most durable predictor at every position, by a wide margin.
• The predicted 2026 miss counts above are conservative by design. They regress toward the position baseline for players with clean histories.

Sample: player-seasons 2016–2025 with adj_gp ≥ 0.25 (played a real snap), trimmed to each position's top 75% by per-game volume, with healthy fill-ins removed (0 missed games and fewer than 4 active weeks) and a pool floor of 20% of team offensive snaps — except players knocked out in week 1 Q1 (no snaps after Q1, on the week 2 report), who stay in. Games missed = regular-season injury quarters ÷ 4: a week counts only with report adjacency (on the injury report that week, or popping on it the following week), full absences count 4 quarters, mid-game exits count the quarters after the player's last appearance, and report-free absences (benchings, role reductions) never count. Prorated ×17/16 for 16-game seasons (pre-2021) so eras are comparable. Regression: MASS::glm.nb per position, response games_missed, predictors prior_missed_3y, prior_severe_3y and age (plus sacks/g for QB). Predictions clipped to [0, 17]. Full spec: scripts/build_injury_predictions_2026.R and scripts/analyze_injury_risk.R.