NFL pressure rate explained: why sacks are the wrong stat

A pass rusher has a quiet year. Half the sacks he managed the season before, no highlight reel worth the name, and by December the phone-ins have decided he is finished. Then somebody opens the pressure column and finds he beat his blocker at almost exactly the rate he always has. Nothing about the player changed. Quarterbacks got the ball out faster against his defence, two of his hits were wiped out by penalties elsewhere on the field, and three of the previous year's sacks were cleanup on somebody else's rush.

That gap between what a rusher does and what the box score records is what pressure rate exists to close. It is why the statistic is now the first thing a scouting department opens, ahead of the sack total, and it is why the phrase turns up in every broadcast. NFL pressure rate is usually explained in a single line: pressures divided by pass-rush snaps. The line is correct and nearly useless, because every word in it is doing more work than it looks. What counts as a pressure depends entirely on who is counting. What counts as a pass-rush snap depends on which plays you are willing to throw away. And the same two words describe at least three different numbers, published side by side, that cannot be compared with one another.

What follows is the mechanism underneath: how a pressure is decided, how the models actually work, which denominator you are looking at, and the specific places the number stops meaning what you think it means.

What a pressure is, before anybody counts it

Strip away the providers and a pressure is a simple idea. On a passing play the defence wants the quarterback to throw badly or not at all. Anything a rusher does that interferes with the throw is worth something, and the sack is only the most extreme version of it.

Three events sit under the umbrella, and they are ordered by severity rather than by kind.

A sack ends the play with the quarterback down behind the line. A quarterback hit means the throw was released but the rusher put him on the ground immediately afterwards. A hurry is everything else that mattered: the rusher forced him off the spot, made him reset his feet, shortened his stroke, or arrived close enough that the throw came out early. Adding those three together gives total pressures, and that sum is the numerator of almost every published pressure rate.

The definitional weight sits on the hurry, because a hurry is a judgement about a counterfactual. Nobody disputes a sack. Everyone can see a hit. A hurry asks whether the throw would have been different had the rusher not been there, and that is an opinion until you build a model to replace the opinion. Every argument about whose pressure numbers to trust is, underneath, an argument about hurries.

Two secondary distinctions are worth having in your head from the start.

The first is that pressure is credited to a defender, but pressure is experienced by an offence. A rusher's pressure rate and a quarterback's pressured-dropback rate are different quantities with different denominators, and a great many published comparisons are quietly mixing the two.

The second is that a pressure is not necessarily a good rush. A defensive tackle who is engulfed for three seconds and then walks past a tired guard while the quarterback climbs into him has generated a pressure. A defensive end who beats a tackle in eight tenths of a second and misses because the ball is already gone has not. Both facts are true, both are recorded, and the difference between them is why the tracking systems added timing measures on top of the raw count.

• 75Pressure probability at which Next Gen Stats records a pressure
• 2.5Seconds a block must hold to count as a pass block win
• 4Minimum pass rushers for a snap to qualify as a true pass set
• 2Seconds under which a throw is filtered out of true pass sets

Definitional constants published by the systems themselves, not season figures.

The three numbers that are all called pressure rate

This is the confusion that ruins most arguments about pressure, so it is worth settling before anything else.

Player pressure rate is one defender's pressures divided by his own pass-rush snaps. It answers the question you actually want answered about an individual: how often does this man win. It is small by nature, because on any given dropback most rushers do not get there.

Defensive pressure rate is the share of opposing dropbacks on which the quarterback was pressured by anyone at all. Eleven defenders are pooled into one number. This is much larger than any individual rate, for the obvious reason that a defence gets four or five attempts per snap and a player gets one.

Pressure rate allowed is the same quantity viewed from the other sideline: the share of a quarterback's dropbacks on which he was pressured. It is filed under the offensive line, which is only partly fair, and we will come back to why.

Put a single set of numbers through all three and the scale of the problem is obvious. A defence that pressures the quarterback on two dropbacks in five is doing very well. A rusher who won on two rushes in five would be the most dominant player the sport has recorded. Same fraction, different universe.

There is a fourth version, and it is the one that most often appears without a label: pressure rate on a filtered subset of snaps. Strip out screens, play-action and quick game and every rate in the table goes up, because you have removed the plays on which pressure was never available. A rusher's number can move by a third on the strength of nothing but the filter.

60 pressures over 500 pass-rush snaps12%

45 pressures over 300 true pass sets15%

24 quick pressures over 500 pass-rush snaps4.8%

60 pressures over 560 team dropbacks10.7%

A constructed season for a single edge defender. The pressure count and the snap count both change with the filter, which is the point. No real player is being described.

Show the numbers

Invented example: one rusher, four denominators, four rates

Item	Value
60 pressures over 500 pass-rush snaps	12%
45 pressures over 300 true pass sets	15%
24 quick pressures over 500 pass-rush snaps	4.8%
60 pressures over 560 team dropbacks	10.7%

Four defensible calculations, one player, one season, and a spread wide enough to move him from good to elite and back again. None of them is wrong. Only one of them is comparable with the number printed next to another player's name, and the only way to know which is to read the column header properly.

How a computer decides the quarterback was pressured

The NFL's own system does not ask a human whether the throw was affected. It computes an answer, and the way it does so is more interesting than the number it produces.

Player tracking supplies the raw material: the position, speed and orientation of every player on the field, several times a second. On its own that is a cloud of coordinates. Turning it into pressure takes three separate models stacked on top of one another.

The first model has to work out who is even involved. On any snap some defenders rush and some drop into coverage; some offensive players block and some run routes, and a running back may do either depending on what he reads. Next Gen Stats uses a graph neural network for this, treating the twenty-two players as a set of relationships rather than as isolated dots, which is the right shape for the problem. Whether a linebacker is rushing is partly a fact about the linebacker and partly a fact about everyone around him.

The second model pairs them up. It identifies which blocker is engaged with which rusher, so a pressure can be attributed to one defender and debited to one lineman, and it recognises when two or more blockers work on the same rusher over the course of a dropback. Double teams are not a footnote here. A defensive tackle who draws two blockers on half his snaps is doing something valuable that his own pressure rate will never show, and the matchup model is what makes that visible at all.

The third model produces the number. It is a random forest, and it estimates for every rusher, in tenths of a second, the probability that this particular rush becomes a pressure. The inputs are the things a coach would look at: how far the rusher is from the quarterback, how fast he is closing, where the quarterback is moving, and how the blocking in front of him is holding up.

Then a threshold. When a rusher's pressure probability goes above 75 per cent, the play is recorded as a pressure. That single cut-off converts a continuous, wobbling curve into a binary event that can be counted and divided.

A constructed rush drawn to show the shape the model produces and the moment the 75 per cent line is crossed. Values are illustrative, not measured.

Show the numbers

Invented illustration: pressure probability across one dropback

Item	Value
0.5s	3%
1.0s	9%
1.5s	24%
2.0s	47%
2.4s	68%
2.6s	81%
3.0s	92%

Reading that curve back is the fastest way to understand what the newer pressure measures are for. The height of the curve is how close the rusher came. The moment it crosses the line is time to pressure. The area above the line is how long the quarterback spent under duress. A rush that peaks at seventy per cent and never crosses records nothing at all, even though a coach watching the tape would call it a good rep.

That last point is the honest limitation of any threshold system. It converts a near miss into a zero. The counterweight is that it does so consistently, for every rusher, in every stadium, without a charter's Tuesday afternoon mood entering the data.

1. 1Tracking data arrivesPosition, speed and orientation for all twenty-two players, sampled several times a second across the whole play. At this stage there is no such thing as a pass rusher, only coordinates.
2. 2Roles are classifiedA graph neural network separates rushers from defenders dropping into coverage, and blockers from receivers running routes. Get this wrong and every number downstream is wrong.
3. 3Blockers are matched to rushersA second model pairs each rusher with the man or men blocking him and flags double teams and chip help. This is what allows one pressure to be credited to a defender and charged to a lineman.
4. 4Pressure probability is computedA third model estimates, in tenths of a second, how likely each rusher is to disrupt the throw, using distance, closing speed, quarterback movement and the state of the block in front of him.
5. 5The threshold converts it to an eventThe instant a rusher's probability passes 75 per cent, the play is logged as a pressure for him. The crossing time becomes time to pressure, and crossings inside 2.5 seconds are separately marked as quick pressures.
6. 6The denominator is chosenPressures are divided by the rusher's own pass-rush snaps, by team dropbacks, or by a filtered subset such as true pass sets. The choice made here moves the published rate further than any player ever does.

The tracking route. A charting service reaches a similar output through a human decision at step four rather than a modelled one.

Charting, tracking and win rate are measuring different things

Three families of pressure data are in general circulation, and treating them as interchangeable is the most common error in the whole subject.

Charting services put trained analysts in front of every snap and ask them to decide. Their advantage is that a person can see intent. A charter knows the tackle gave ground deliberately to invite a rusher upfield, that the quarterback stepped up because the play design told him to rather than because anyone chased him, and that the guard's hand placement was beaten a half second before the whistle. Their disadvantage is that people are not identical to each other and are not identical to themselves in week seventeen.

Tracking models replace the judgement with geometry and a threshold. They are perfectly consistent and completely literal. They will happily record a pressure by a defender who was standing unblocked because the protection slid the wrong way, and they cannot see that a rusher was beaten by technique while remaining physically close.

Win rate metrics ask a narrower question and are cleaner for it. ESPN's version, built on the same league tracking feed, does not ask whether the quarterback was disturbed. It asks whether the blocker held or the rusher got past, judged at a fixed 2.5 seconds. A blocker who sustains for 2.5 seconds or more is credited with a win; a rusher who beats his block inside that window gets one instead. The pocket is treated as a polygon drawn around the blockers, and the rusher wins by penetrating it within a certain distance of the quarterback.

That design decision matters more than it sounds. Win rate deliberately ignores the outcome. It does not care whether the quarterback threw a touchdown or was buried, only whether the block held long enough, which makes it a purer measure of the individual matchup and a worse measure of what the defence achieved. Screens are thrown out of the calculation entirely, since the whole intent of a screen is to let the rush through, and unblocked rushers are credited to the rusher and to the team without being charged against any particular lineman.

Each of the three systems is internally coherent. None of them is a translation of the others. A rusher can sit high in win rate and middling in pressure rate because he beats his man early and the quarterback throws before it matters, and a rusher can invert that pattern by getting home late against long-developing plays. Both patterns describe real players. Neither is a data error.

Why 2.5 seconds is the hinge

Almost every timing rule in the subject points at the same moment, and the reason is worth having.

A conventional dropback has a rhythm. The quarterback takes his steps, hits the top of the drop, works his read, and throws. On ordinary concepts that whole sequence lives in the region of two and a half seconds, which is why ESPN chose that figure for its win rates: it approximates the average time to release on standard dropbacks, and in testing it did the best job of separating the blockers and rushers already known to be good.

Pressure arriving before that point and pressure arriving after it are different events with different consequences.

Early pressure attacks the read. The quarterback has not yet worked through his progression, so he throws to the first thing available, checks down, or leaves the pocket without knowing what is downfield. The play is broken before the play has happened.

Late pressure attacks the throw. If a rusher gets home at three and a half seconds, the quarterback has already had time to find his second and third options. Either the ball is gone, or the coverage has held so well that nobody is open, which is a compliment to the secondary rather than to the rush.

This is why quick pressure rate is tracked separately from pressure rate, and why the two rank teams differently. A defence generating a high overall pressure rate but few quick pressures is probably getting there because the coverage is buying time, not because the rush is winning. A defence with the reverse profile has a rush beating protections cleanly, whatever the total says.

It also explains a piece of coaching that looks like cheating. The fastest way to reduce your pressure rate allowed is to throw the ball sooner. Quick game, run-pass options, screens and designed rollouts all subtract from the time the rush has available, and every one of them makes the offensive line look better. That is a real benefit, not a trick, but it belongs to the play caller rather than the tackle. The relationship runs the other way too: an offence built on shot plays and deep drops hands its rushers extra time and will post a worse protection number behind identical blocking.

The quarterback is half of his own pressure rate

Pressure rate allowed is filed under offensive line performance, and that filing is wrong often enough to be a problem.

Four things move it, and only one of them is the line.

Time to throw. Pressure is cumulative. Every additional tenth of a second gives the rush another tenth to work with, and the probability curve above shows what that looks like. A quarterback who consistently holds the ball an extra half second will be pressured substantially more often behind identical blocking, and the tape will show his linemen winning their reps and then losing them.

Pocket discipline. Quarterbacks who drift backwards, bail early, or climb into a rusher they cannot see turn contained rushes into pressures. Quarterbacks who step up through a clean lane while the ends run past behind them make good rushes disappear. This is a genuine, coachable, individual skill, and it shows up in the data as a property of the offensive line.

Protection calls and personnel. Keeping a back or a tight end in to block subtracts a receiver from the route. That is a real cost, it is a choice, and it lowers the pressure rate. An offence that empties the backfield to flood the coverage is accepting more pressure on purpose.

The receivers. If nobody gets open, the throw does not come out, and the sack that follows was created by the coverage. This is the coverage sack, and it is charged to the protection by every counting system in existence, which is one reason sack totals are such a treacherous way to evaluate a defensive front. A meaningful share of a defence's sack production can be manufactured by its secondary.

The practical consequence is that pressure rate allowed should be read as a property of the whole passing operation, and that comparing two quarterbacks on it without adjusting for time to throw is close to meaningless. It is the same class of error as judging a receiver on catch rate without asking about target depth, and it is corrected the same way, by conditioning on the thing that drives it. The expected points framework is one route to that, since it prices the outcome of a pressured dropback rather than the pressure itself, and completion percentage over expected attacks the same problem from the throwing side.

True pass sets, and the filter that makes the number honest

Once you accept that not every dropback offers the same opportunity, the next step is obvious. Stop counting the ones that do not.

The best known version of this is the true pass set. It is a filter applied before the division, and it removes the plays on which the matchup was never a fair test of blocking or rushing.

Out go screens, where the offence wants the rush to come. Out go play-action passes, where the rusher's first step is compromised and the timing is deliberately unusual. Out go run-pass options and short drops, where the ball leaves almost immediately. Out go snaps with fewer than four rushers, since a three-man rush against five blockers is not a pass-protection test. And out go throws released in under two seconds, which are decided before the rush has had a chance to matter.

What remains is the closest thing football has to a controlled experiment: a standard drop, a normal rush count, and enough time for the matchup to resolve. Rates computed on that subset hold up better from season to season and travel better from college to the professional game, which is precisely why draft evaluation leans on them so heavily. A tackle who protected well in an offence built on quick throws has told you very little. The same tackle's record on true pass sets tells you a great deal.

The cost of the filter is sample. Cutting away a large share of dropbacks leaves fewer plays behind, and a rate computed on fewer snaps moves around more from random variation. A prospect with an outstanding true pass set number over a modest count of snaps is weaker evidence than the headline suggests, and that is the specific trap in a lot of draft-week analysis. The scouting process around the combine carries the same tension: cleaner tests, fewer of them.

Why pressure rate predicts better than sack rate

The strongest argument for pressure rate has nothing to do with fairness and everything to do with sample size.

A sack is a rare event. Even a productive edge rusher records them at a rate of roughly one a game across a season, and there are only so many dropbacks in a Sunday. Rare events carry enormous variance: whether the quarterback slipped, whether the ball came out a tenth of a second late, whether a teammate flushed him into your arms. Over a full season those coin flips do not average out anything like as well as people assume, which is why sack totals swing violently for players whose underlying performance did not move.

Pressures happen several times a game for the same player. More events means the random component averages down faster, and what is left is closer to the thing you were trying to measure in the first place.

The consequence, and it is the finding that changed how front offices evaluate pass rushers, is that this season's pressure rate predicts next season's sacks better than this season's sacks do. A rusher with high pressures and low sacks is a buy. A rusher with the reverse is very likely to disappoint, and the market has still not fully priced this in, which is why the discount exists at all.

The bridge between them is pressure-to-sack rate, the share of a player's pressures that convert into sacks. It is a tempting statistic and mostly a trap. Conversion depends on the quarterback's escapability, on whether teammates arrive at the same moment, and on where on the field the rush happens. Rusher-to-rusher differences in conversion are real but small, and over one season they are swamped by noise. Treat an unusual conversion rate as something likely to revert rather than as a discovered skill and you will be right far more often than not.

This is the same logic that governs the rest of modern football analysis. Prefer the frequent, controllable input to the rare, contingent outcome. It is why fourth-down models work from conversion probability rather than from what happened last time, and why efficiency measures like DVOA are built on per-play success rather than on wins.

What a pressure is actually worth

Pressure rate tells you how often. It does not tell you how much, and the two questions have quite different answers.

The severity ladder is steep and it is not linear. A sack is enormously damaging, because it combines lost yardage with a lost down and often knocks the offence out of field goal range. A hit on the throwing motion is the next most valuable thing a rusher can do, because it degrades the throw itself while the ball is in the air. A hurry that moves the quarterback off his spot is worth real value. A hurry that arrives as the ball leaves is worth close to nothing.

Weighted pressure metrics exist for this reason. The common shape is to count sacks at full value, hits at something close to it, hurries at a fraction, then divide by pass-rush snaps as usual. Whether the weights are worth the added complexity is a live argument. They make the metric more descriptive of what happened and slightly less predictive of what will happen next, which is the standard trade in this field.

The interaction with coverage is the part most published analysis skips. Pressure and coverage are not separable phenomena. A rush that gets home in four seconds needed the secondary to hold for four seconds. A quarterback who throws in 2.1 seconds was not beaten by a great rush; he was let off by a coverage that gave him an immediate answer. This is why pressure rate should be read alongside how long the defence is asking its rushers to work, and why a defence with a mediocre pressure rate and superb coverage may be perfectly well constructed. The shells that buy that time, including the two-deep structures built on it, determine how much rush the front actually has to generate.

Blitzing: buying pressure with bodies

The cleanest way to raise a pressure rate is to send more rushers. It is also the most expensive.

The arithmetic is unforgiving. There are eleven defenders. Every additional rusher is one fewer body in coverage, which means either a receiver is unaccounted for or a zone gets larger. Send five and somebody is one on one. Send six and the protection is outnumbered, but so is the secondary, and a quarterback who identifies it before the snap has a free throw against a defence with no deep help.

Blitzing therefore raises the probability of pressure and raises the cost of failure at the same time. A defence that generates pressure with four rushers has solved the problem for free: it keeps seven in coverage, disguises what it is doing, and hands the quarterback no easy answer. That is why the market for pass rushers who win alone is the second most expensive in the sport, behind quarterbacks, and why so much of the salary cap ends up allocated to the defensive front.

Reading blitz numbers and pressure numbers together is where the useful information sits. A defence with a low blitz rate and a high pressure rate has a genuine front. A defence with a high blitz rate and a high pressure rate is manufacturing it, which works until an offence is patient enough to make it pay. A defence with a high blitz rate and a low pressure rate is in serious trouble, since it is paying the coverage cost without collecting the benefit. The specific mechanics of who is sent and who replaces them are a subject of their own, covered in the pressure packages defences build.

The tactical middle ground is the stunt: two rushers exchanging paths to confuse the protection rather than adding a body. Stunts and games generate pressure without conceding a coverage defender, at the cost of time, since the exchange itself takes a beat to develop. On the data side they create an attribution problem, because the man who penetrates is frequently not the man who beat anybody. The looper walks through a gap that the crasher opened, and the pressure lands on the wrong name.

Where the number breaks

Every metric has conditions under which it stops describing what it claims to describe. Pressure rate has more than most, and knowing them is the difference between using the statistic and being used by it.

Unblocked pressure. A rusher who is not blocked, because the protection slid the wrong way or a back missed his pickup, will be credited with a pressure. He did nothing. Team-level numbers are inflated by these more than player-level numbers, and a defence with an unusually good pressure rate against a poor protection team has learned less than it thinks.

Double-team credit. The interior lineman who occupies two blockers all afternoon appears on nobody's leaderboard. His edge rushers do. This is the single largest systematic unfairness in the statistic, and it is why interior defenders are chronically undervalued by raw pressure rate and why the matchup models that flag double teams matter so much.

Cleanup and coverage sacks. When a quarterback holds the ball four seconds against perfect coverage and is eventually dragged down, the rusher who arrives gets the credit. He may have produced the fourth-best rush on the play.

No opponent adjustment. Most published pressure rates are raw. A rusher who faces weak tackles all season looks better than one who does not, and the tracking-based win rates are explicit about not correcting for opponent quality or for the extra difficulty added by play action and rollouts.

Garbage time. Pressure rate climbs when everyone in the building knows a pass is coming. A defence that spends its season ahead by three scores in the fourth quarter accumulates cheap pressure against opponents who must drop back and cannot run. Rate statistics do not care about the score, and this one probably should.

Scheme-inflated denominators. A defensive end in a scheme that rushes him on nearly every dropback will post a lower rate than one deployed selectively in obvious passing situations. Neither is necessarily the better player. The second is being used more efficiently and is being rewarded for it by the statistic.

Threshold effects. The 75 per cent cut-off, like all thresholds, produces a discontinuity. Two nearly identical rushes can land on opposite sides of it. Over a season that washes out for a player with a normal snap count, and it does not wash out at all for a rotational player with a small one.

Reading a pressure rate table without being fooled

Four questions, asked in order, will get you most of the way to a defensible reading of any pressure number you meet.

Whose denominator is it? Player pass-rush snaps, team dropbacks, or a filtered subset. Until you know which, you cannot compare the number with anything. This one check disposes of most of the bad arguments on the subject.

Who counted it? A charting service, a tracking model, or a win-rate system. They disagree by design, and their disagreements are informative. When a rusher rates highly in one and poorly in another, the gap usually has a football explanation worth finding.

How fast did it arrive? Total pressure rate and quick pressure rate rank teams differently, and quick pressure is the one that describes a rush winning on its own. If only one number is published, assume it is the flattering one.

How many snaps is it built on? Filtered rates in particular are computed on far fewer plays than the headline implies. A striking rate over a small count is a hypothesis, not a finding.

Answer those and the statistic does what it was built to do. It tells you which pass rushers are genuinely winning, several times a game, in a way a sack total collected on a dozen lucky Sundays never could. That is the whole argument for it. Not that pressure matters more than sacks, which it plainly does not, but that pressure is measured often enough to be believed, and sacks are not.

The rest of this sport's numbers submit to the same test. Ask what the denominator is, ask who decided, ask how often the event happens, and most of the arguments answer themselves. More explanations built that way, on this sport and others, sit in the American football archive.