The data
190 permanent head-coaching stints began in 2000–2026; 180 have at least one completed season and are used for fitting. 42 interim-only stints are excluded.
Sources
- Games and coaches: nflverse results with the credited head coach, corrected for mid-season changes using Wikipedia departure dates.
- Why each coach left: Wikipedia’s yearly “NFL season” articles, with a quote and URL per departure.
- Birth dates: Wikidata. Pre-1999 jobs: coach infoboxes. GMs and owners: Wikipedia.
Terms
- Fired
- The club ended the job: firings, “mutual” partings, non-renewals, and reassignment out of the role.
- Censoring
- We know a coach lasted at least t seasons unfired, not what came after. Active coaches, resignations and retirements are censored.
- Hazard
- The chance of being fired after season t, given he reached season t.
- Hazard ratio
- How a covariate multiplies that per-season risk. 0.6 = 40% lower risk.
- Counting process
- One row per coach-season, interval (k−1, k], so covariates like win % can change every year.
How long tenures last
Half of coaches are fired by the end of season 4. Only 17% make it through season 8.
Kaplan–Meier median: 4 seasons (95% CI 3–4). First-time NFL head coaches: 4 seasons (95% CI 3–4). Previous NFL head coaches: 3 seasons (95% CI 3–4). Log-rank test for a difference: p = 0.566.
| Seasons | At risk | Share not yet fired | 95% CI |
|---|---|---|---|
| 1 | 180 | 89% | 84%–94% |
| 2 | 152 | 73% | 67%–80% |
| 3 | 116 | 52% | 45%–60% |
| 4 | 78 | 36% | 30%–45% |
| 5 | 49 | 31% | 25%–40% |
| 8 | 19 | 17% | 12%–25% |
| 10 | 9 | 11% | 7%–19% |
What’s known on hiring day
Age, experience and the team’s prior record barely predict tenure: concordance 0.54, where 0.50 is a coin flip.
| Covariate | Hazard ratio | 95% CI | p |
|---|---|---|---|
| Age at hire (per +5 years) | 1.13 | 0.98–1.30 | 0.083 |
| Previous NFL head coach | 0.84 | 0.53–1.33 | 0.451 |
| Team win % the season before hire (per +10 pts) | 1.05 | 0.95–1.16 | 0.383 |
Checking proportional hazards
The Cox model assumes each hazard ratio is constant across tenure. cox.zph checks whether each covariate’s scaled Schoenfeld residuals trend over time; a small p-value means the effect drifts.
Baseline model
Violations: none at the 5% level.
| Term | Chi-square | p |
|---|---|---|
| Age at hire (per +5 years) | 0.26 | 0.609 |
| Previous NFL head coach | 3.21 | 0.073 |
| Team win % the season before hire (per +10 pts) | 0.20 | 0.654 |
| Global test | 7.20 | 0.066 |
Time-varying model
Violations: Change in win % vs last season (per +10 pts).
| Term | Chi-square | p |
|---|---|---|
| Win % this season (per +10 pts) | 1.89 | 0.169 |
| Change in win % vs last season (per +10 pts) | 7.87 | 0.005 |
| Made playoffs this season | 0.19 | 0.660 |
| Point differential per game (per +1) | 2.57 | 0.109 |
| New GM or owner this season | 2.12 | 0.146 |
| Previous NFL head coach | 0.00 | 0.994 |
| Age at hire (per +5 years) | 0.64 | 0.422 |
| Global test | 13.89 | 0.053 |
Holding this season’s win % fixed, a bigger “change from last season” means last season was worse: the term measures two bad years in a row. Its effect differs early and late in a tenure, so the model splits it.
| Covariate | Hazard ratio | 95% CI | p |
|---|---|---|---|
| Win % this season (per +10 pts) | 0.57 | 0.45–0.72 | < 0.001 |
| Change in win % vs last season, seasons 1–2 | 0.94 | 0.80–1.10 | 0.438 |
| Change in win % vs last season, season 3+ | 1.26 | 1.11–1.42 | < 0.001 |
| Made playoffs this season | 0.17 | 0.06–0.48 | < 0.001 |
| Point differential per game (per +1) | 0.98 | 0.92–1.04 | 0.432 |
| New GM or owner this season | 1.08 | 0.72–1.62 | 0.722 |
| Previous NFL head coach | 1.09 | 0.74–1.60 | 0.680 |
| Age at hire (per +5 years) | 1.10 | 0.97–1.24 | 0.136 |
Split model PH check: global p = 0.415; violations: none at the 5% level. The hot-seat estimates use this model.
Season by season
Win and you stay. Each +10 pts of win % (about 1.7 wins) cuts that season’s firing risk by 41%; a playoff trip cuts it by 82%. Only 4 of 138 firings came right after a playoff season.
Time-varying Cox model on 734 coach-seasons and 138 firings, using only what was known when each decision was made. Standard errors clustered by stint. Concordance 0.87.
| Covariate | Hazard ratio | 95% CI | p |
|---|---|---|---|
| Win % this season (per +10 pts) | 0.59 | 0.47–0.74 | < 0.001 |
| Change in win % vs last season (per +10 pts) | 1.14 | 1.02–1.26 | 0.016 |
| Made playoffs this season | 0.18 | 0.07–0.51 | 0.001 |
| Point differential per game (per +1) | 0.97 | 0.92–1.03 | 0.373 |
| New GM or owner this season | 1.09 | 0.72–1.66 | 0.671 |
| Previous NFL head coach | 1.13 | 0.77–1.65 | 0.532 |
| Age at hire (per +5 years) | 1.10 | 0.97–1.24 | 0.152 |
How tenures end
Firing is the dominant exit: by season 8, 77% of coaches have been fired and 10% have left another way.
Cumulative incidence gives the real-world share of coaches who have left each way by season t, allowing for the fact that a coach who retired can no longer be fired.
| By end of season | Fired | Resigned, retired or other |
|---|---|---|
| 2 | 27% | 4% |
| 4 | 61% | 7% |
| 6 | 73% | 10% |
| 8 | 77% | 10% |
The hot seat: risk of being fired after 2026
Each team’s 2026 season is projected from results through week 6 plus .500 football and an even point differential the rest of the way. Playoffs aren’t decided, so each coach gets two numbers.
Model estimates, not predictions. The model knows wins, playoffs, tenure, age, experience and front-office turnover. It knows nothing about Super Bowl rings, contracts, quarterback injuries or owner temperament, and the baseline risk for very long tenures is pooled and imprecise.
| # | Coach | Team | Season | Record | Proj. win % | If team misses playoffs | If makes |
|---|---|---|---|---|---|---|---|
| 1 | Dan Quinn | WAS | 3 | 1-3-1 | 0.441 | 50% | 11% |
| 2 | Jim Harbaugh | LAC | 3 | 0-5 | 0.353 | 38% | 8% |
| 3 | Shane Steichen | IND | 4 | 2-3 | 0.471 | 32% | 6% |
| 4 | Todd Bowles | TB | 5 | 1-4 | 0.412 | 31% | 6% |
| 5 | Dan Campbell | DET | 6 | 3-2 | 0.529 | 25% | 5% |
| 6 | DeMeco Ryans | HOU | 4 | 1-4 | 0.412 | 24% | 4% |
| 7 | Dave Canales | CAR | 3 | 2-2 | 0.500 | 24% | 4% |
| 8 | Nick Sirianni | PHI | 6 | 2-3 | 0.471 | 23% | 4% |
| 9 | Andy Reid | KC | 14 | 4-0 | 0.618 | 20% | 4% |
| 10 | Sean Payton | DEN | 4 | 4-2 | 0.559 | 19% | 3% |
| 11 | Zac Taylor | CIN | 8 | 3-2 | 0.529 | 18% | 3% |
| 12 | Matt LaFleur | GB | 8 | 3-2 | 0.529 | 13% | 2% |
| 13 | Kellen Moore | NO | 2 | 1-4 | 0.412 | 11% | 2% |
| 14 | Jeff Hafley | MIA | 1 | 0-5 | 0.353 | 11% | 2% |
| 15 | Brian Schottenheimer | DAL | 2 | 2-3 | 0.471 | 10% | 2% |
| 16 | Mike Macdonald | SEA | 3 | 3-3 | 0.500 | 10% | 2% |
| 17 | Sean McVay | LA | 10 | 2-3 | 0.471 | 10% | 2% |
| 18 | Mike Vrabel | NE | 2 | 3-2 | 0.529 | 9% | 2% |
| 19 | Robert Saleh | TEN | 1 | 0-5 | 0.353 | 9% | 2% |
| 20 | Aaron Glenn | NYJ | 2 | 2-3 | 0.471 | 9% | 2% |
| 21 | Kevin O'Connell | MIN | 5 | 5-0 | 0.647 | 9% | 2% |
| 22 | Ben Johnson | CHI | 2 | 3-2 | 0.529 | 6% | 1% |
| 23 | Kyle Shanahan | SF | 10 | 5-0 | 0.647 | 6% | 1% |
| 24 | Mike LaFleur | ARI | 1 | 1-4 | 0.412 | 5% | 1% |
| 25 | Kevin Stefanski | ATL | 1 | 2-3 | 0.471 | 5% | 1% |
| 26 | Mike McCarthy | PIT | 1 | 3-2 | 0.529 | 5% | 1% |
| 27 | Liam Coen | JAX | 2 | 4-1 | 0.588 | 5% | 1% |
| 28 | Todd Monken | CLE | 1 | 3-2 | 0.529 | 4% | 1% |
| 29 | John Harbaugh | NYG | 1 | 3-1-1 | 0.559 | 3% | 1% |
| 30 | Klint Kubiak | LV | 1 | 3-2 | 0.529 | 2% | 0% |
| 31 | Joe Brady | BUF | 1 | 4-1 | 0.588 | 2% | 0% |
| 32 | Jesse Minter | BAL | 1 | 4-1 | 0.588 | 2% | 0% |
Limitations
- Small sample: 138 firings, so intervals for anything beyond winning and playoffs are wide.
- Seasons are a coarse time scale with many ties (Efron’s method is used).
- Some coding is judgment: which partings count as firings, who held final personnel say without a conventional GM, and dates known only to the month or year. The research files keep the raw wording so these can be re-coded.