NFL · Empirical Bayes
True-Talent Leaderboards
A raw single-season rate is a noisy guess at a player’s real ability, and the noise is worst for the smallest samples — so a naïve leaderboard is topped by whoever got lucky in the fewest tries. This one shrinks every rate toward its position-group prior by how much the sample can be trusted, and shows a 90% credible interval instead of a point. Toggle raw vs. shrunk to watch the flukes fall back to the pack.
Dropback success rate · 2018
Share of dropbacks (incl. sacks & scrambles) with positive EPA.
- Beats raw by
- 15.6%
- lower out-of-sample error
- RMSE raw → shrunk
- 5.5% → 4.6%
- odd vs. even weeks
- Split-half reliability
- 0.47
- how repeatable the raw stat is
- Stabilizes at
- 192
- dropbacks to trust the number
Each row is a player: the solid dot is the shrunk estimate, the hollow dot the raw rate, joined by the red pull of regression; the grey bar is the 90% credible interval. A hollow shrunk dot means the sample is below the stabilization line — the number is mostly the position prior. Switch to Raw rank and watch the small-sample names climb.
Dropback success rate · 2018 · full board
| # | Player | Team | Pos | dropbacks | Raw | Shrunk | 90% interval |
|---|---|---|---|---|---|---|---|
| 1 | Drew Brees | NO | QB | 508 | 55.5% | 52.9% | 49.8% – 56.0% |
| 2 | Patrick Mahomes | KC | QB | 606 | 51.6% | 50.3% | 47.3% – 53.2% |
| 3 | Tom Brady | NE | QB | 590 | 51.0% | 49.8% | 46.8% – 52.7% |
| 4 | Jared Goff | LA | QB | 595 | 50.9% | 49.7% | 46.8% – 52.6% |
| 5 | Philip Rivers | LAC | QB | 547 | 51.0% | 49.7% | 46.7% – 52.7% |
| 6 | Andrew Luck | IND | QB | 658 | 50.5% | 49.4% | 46.6% – 52.2% |
| 7 | Matt Ryan | ATL | QB | 655 | 50.4% | 49.4% | 46.5% – 52.2% |
| 8 | Ben Roethlisberger | PIT | QB | 701 | 49.8% | 48.9% | 46.2% – 51.7% |
| 9 | Carson Wentz | PHI | QB | 431 | 49.9% | 48.6% | 45.4% – 51.9% |
| 10 | Ryan Fitzpatrick | TB | QB | 261 | 50.6% | 48.6% | 44.7% – 52.4% |
| 11 | Jameis Winston | TB | QB | 407 | 48.9% | 47.9% | 44.6% – 51.3% |
| 12 | Cam Newton | CAR | QB | 501 | 48.7% | 47.9% | 44.8% – 51.0% |
| 13 | Kirk Cousins | MIN | QB | 648 | 48.3% | 47.8% | 44.9% – 50.6% |
| 14 | Derek Carr | LV | QB | 598 | 47.7% | 47.2% | 44.3% – 50.1% |
| 15 | Mitchell Trubisky | CHI | QB | 461 | 47.7% | 47.2% | 44.0% – 50.4% |
| 16 | Jimmy Garoppolo◦ provisional | SF | QB | 103 | 49.5% | 47.1% | 42.4% – 51.9% |
| 17 | Deshaun Watson | HOU | QB | 562 | 47.5% | 47.1% | 44.1% – 50.1% |
| 18 | Andy Dalton | CIN | QB | 384 | 47.4% | 46.9% | 43.5% – 50.3% |
| 19 | Matt Barkley◦ provisional | BUF | QB | 26 | 53.8% | 46.8% | 41.3% – 52.4% |
| 20 | Nick Mullens | SF | QB | 288 | 46.9% | 46.5% | 42.7% – 50.2% |
| 21 | Kyle Allen◦ provisional | CAR | QB | 31 | 48.4% | 46.2% | 40.8% – 51.7% |
| 22 | Lamar Jackson◦ provisional | BAL | QB | 186 | 45.7% | 45.8% | 41.6% – 50.0% |
| 23 | Chase Daniel◦ provisional | CHI | QB | 86 | 45.4% | 45.7% | 40.8% – 50.6% |
| 24 | Russell Wilson | SEA | QB | 471 | 45.6% | 45.7% | 42.5% – 48.9% |
| 25 | Colt McCoy◦ provisional | WAS | QB | 58 | 44.8% | 45.6% | 40.5% – 50.8% |
| 26 | C.J. Beathard◦ provisional | SF | QB | 189 | 45.0% | 45.4% | 41.3% – 49.6% |
| 27 | Nick Foles | PHI | QB | 203 | 44.8% | 45.3% | 41.2% – 49.5% |
| 28 | Baker Mayfield | CLE | QB | 515 | 45.1% | 45.3% | 42.2% – 48.4% |
| 29 | Aaron Rodgers | GB | QB | 647 | 45.0% | 45.2% | 42.4% – 48.0% |
| 30 | Marcus Mariota | TEN | QB | 374 | 44.6% | 45.1% | 41.6% – 48.5% |
| 31 | Joe Flacco | BAL | QB | 394 | 44.4% | 44.9% | 41.5% – 48.3% |
| 32 | Teddy Bridgewater◦ provisional | NO | QB | 25 | 36.0% | 44.7% | 39.2% – 50.3% |
| 33 | Dak Prescott | DAL | QB | 583 | 44.3% | 44.6% | 41.7% – 47.6% |
| 34 | Josh Johnson◦ provisional | WAS | QB | 102 | 42.2% | 44.6% | 39.8% – 49.4% |
| 35 | Alex Smith | WAS | QB | 349 | 43.8% | 44.6% | 41.1% – 48.1% |
| 36 | Cody Kessler◦ provisional | JAX | QB | 153 | 41.8% | 44.1% | 39.7% – 48.5% |
| 37 | Taylor Heinicke◦ provisional | CAR | QB | 58 | 37.9% | 44.0% | 38.9% – 49.2% |
| 38 | Blaine Gabbert◦ provisional | TEN | QB | 106 | 40.6% | 44.0% | 39.3% – 48.7% |
| 39 | Matthew Stafford | DET | QB | 597 | 42.9% | 43.6% | 40.7% – 46.5% |
| 40 | Eli Manning | NYG | QB | 627 | 42.7% | 43.5% | 40.6% – 46.3% |
| 41 | Case Keenum | DEN | QB | 620 | 42.7% | 43.5% | 40.6% – 46.4% |
| 42 | Derek Anderson◦ provisional | BUF | QB | 75 | 37.3% | 43.5% | 38.5% – 48.5% |
| 43 | Brock Osweiler | MIA | QB | 195 | 40.5% | 43.2% | 39.1% – 47.3% |
| 44 | Sam Darnold | NYJ | QB | 444 | 41.7% | 42.9% | 39.7% – 46.2% |
| 45 | Mark Sanchez◦ provisional | WAS | QB | 42 | 28.6% | 42.8% | 37.5% – 48.1% |
| 46 | Josh McCown◦ provisional | NYJ | QB | 116 | 37.1% | 42.6% | 38.0% – 47.2% |
| 47 | Blake Bortles | JAX | QB | 436 | 41.1% | 42.5% | 39.3% – 45.8% |
| 48 | Ryan Tannehill | MIA | QB | 308 | 39.9% | 42.2% | 38.6% – 45.9% |
| 49 | Jeff Driskel | CIN | QB | 195 | 38.0% | 41.9% | 37.8% – 46.0% |
| 50 | DeShone Kizer◦ provisional | GB | QB | 45 | 24.4% | 41.8% | 36.6% – 47.1% |
| 51 | Sam Bradford◦ provisional | ARI | QB | 87 | 31.0% | 41.2% | 36.4% – 46.1% |
| 52 | Tyrod Taylor◦ provisional | CLE | QB | 97 | 30.9% | 40.8% | 36.1% – 45.6% |
| 53 | Josh Allen | BUF | QB | 348 | 37.9% | 40.8% | 37.3% – 44.3% |
| 54 | Nathan Peterman◦ provisional | BUF | QB | 89 | 27.0% | 39.9% | 35.1% – 44.7% |
| 55 | Josh Rosen | ARI | QB | 438 | 35.4% | 38.6% | 35.4% – 41.8% |
How the shrinkage works
Two estimators
Rate stats (completion %, success rate, catch rate) use a beta-binomial model: a Beta(α, β) prior fit by marginal likelihood over each position group, then a Beta posterior per player. Per-play averages (EPA, CPOE, yards) use a normal-normal model with DerSimonian–Laird between-player variance. Both pull each player toward their group by exactly how thin their sample is.
Does it help? & the fine print
The trust panel’s numbers come from a leakage-free odd/even-week holdout: fit on odd weeks, predict even-week raw. Shrinkage lowers out-of-sample error for every stat. Priors are fit per season and per position (WR and TE separately), so a TE’s baseline isn’t a skill. Regular season only, 2016–2025. Rushing and receiving efficiency are heavily scheme-driven — read the wide bands as the honesty they are.
Source: nflverse play-by-play. Counted and computed deterministically — never modeled by a language model. Built by build_nfl_leaderboards.py (byte-reproducible).