Why Pass Rush Win Rate Lies to You

Pass rush win rate has become the default metric for evaluating pass rushers. ESPN built it, Twitter runs with it, and it shows up in draft coverage every spring like clockwork. PRWR measures something very specific though, and the way it gets cited usually strips out all the context that makes it useful.

What PRWR actually measures

ESPN’s version tracks whether a rusher beats his blocker within 2.5 seconds of the snap. A “win” means beating the block. Not getting to the quarterback, not generating pressure, not doing anything that actually shows up on the stat sheet. Just beating the block. The data comes from NFL Next Gen Stats player tracking.

The 2.5-second threshold is doing a lot of heavy lifting in this metric. A rusher who wins at 2.6 seconds gets nothing. A rusher who wins at 2.4 seconds but the QB already threw the ball still gets credit, even though the play was completely unaffected.

The 2025 numbers

The 2025 season gave us a clean example of why this gets messy. Two edge rushers, both full-season starters, both productive players, with very different stories once you look past the win rate.

import pandas as pd

rushers = pd.DataFrame({
    "player": ["Nik Bonitto", "Josh Hines-Allen"],
    "prwr": [17.1, 19.2],
    "pressures": [76, 95],
    "sacks": [14.0, 8.0],
})

print(rushers.to_string(index=False))

Hines-Allen had a higher win rate and generated 19 more pressures over the season. Bonitto still ended up with nearly twice as many sacks. Part of that is volume and snap count distribution, part of it is how well a player converts pressure into something that actually brings the quarterback down vs. just getting close. Hines-Allen was winning his reps and creating disruption on paper, but Bonitto was finishing plays. PRWR captures the first part and completely misses the second.

You see the same kind of thing at the top of the leaderboard. Will Anderson Jr. led the NFL in PRWR at 26.2% with 93 pressures, which is dominant by any measure. Myles Garrett finished at 24.6% and set the single-season sack record with 23. Garrett was double-teamed on more than half his pass rush snaps and chipped on over 20%, which tanks your win rate. The metric penalizes him for being so disruptive that offenses build their entire protection scheme around him.

The 2.5-second cutoff

The threshold itself creates bias. Speed rushers who win with a quick first step look great. Power rushers who collapse the pocket at 3+ seconds look worse than they actually are.

Jeffery Simmons is a good example. He’s a 305-pound interior lineman who physically moves guards into the quarterback’s lap. In 2025 he had 11 sacks and a 91.4 PFF pass-rush grade, which is elite for a defensive tackle. His PRWR was 19.4%. Solid for the position, but it doesn’t come close to reflecting what he does to an offense on a snap-by-snap basis. A bull rush that takes 3.1 seconds and caves in the entire pocket doesn’t register as a win.

The Box quote

I think about the George Box quote a lot. “All models are wrong, but some are useful.” PRWR is useful. It measures a real thing, and at the extremes it’s genuinely predictive. A guy with a 30%+ win rate is almost certainly elite.

But in the 15-22% range where most starters live, treating it as ground truth is exactly the kind of false precision that makes people distrust analytics in the first place.

One number is never going to be enough. That’s true in statistics and it’s true in football.