What follows is a hypothesis. Like a good hypothesis, it fits the available evidence. Like a good hypothesis, it is testable and falsifiable. But it hasn’t been tested, no attempt has yet been made to falsify it, so, for now, it should be treated like a plausible idea and nothing more.
Also: this post is long. Taking long/complex concepts and making them short/simple is a lot of effort. When someone is paying me to write words, I put in that effort. When I’m writing words for free the choice is more “write it long/complex” or “don’t write it at all”. Apologies in advance.
If you are riding a bicycle sufficiently fast and you jam the brakes sufficiently hard, your wheels will lock up and stop rotating as you slow down. This might sound like a good thing, but it costs you traction (because your wheel is skidding along the road instead of maintaining constant contact) and you lose the ability to steer (since your wheel is no longer freely rotating the amount it can influence direction is greatly diminished). This is bad.
For the most part you’re not going to be getting a bicycle going fast enough for this to come into play. But cars routinely hit the kinds of speeds and braking forces required to lock up the wheels, which created an unsafe situation for many motorists. So science created Anti-Lock Braking Systems (ABS), designed to pulse the brakes in a way that prevents them from locking no matter how fast you’re going or how hard you’re mashing the brake pedal. This is a good thing that unambiguously increases motorist safety.
Because ABS is a good thing that increases motorist safety, you would expect motorists in cars that have ABS to be safer than motorists in cars that don’t have ABS, on average. You’d be wrong! A study of German taxi cabs (half of which were equipped with ABS and half of which were not), found that the crash rate for both types of vehicles was essentially the same. (Actually, ABS-equipped vehicles got in slightly more crashes.)
How could this be? Because the safety of the car is not the only thing that determines crash rates; the safety of the car’s operator also plays a substantial role. And studies find that drivers of cars that have ABS follow more closely than drivers without ABS, trusting the ABS to protect them if they need to stop suddenly.
This idea that gains in safety can be consumed by motorists was termed the Offset Hypothesis, or Risk Compensation. A stronger form of this theory is called Risk Homeostasis, which posits that human beings have a certain amount of risk they are willing to accept, and anything that reduces perceived risk will lead to them taking riskier behaviors until they are back at their initial risk set-point.
An example from daily life: go visit a playground. Any playground. I’m willing to bet that the slides are plastic instead of metal. That any metal support structure is coated in a thick layer of rubber. That the surface of the playground is either thick mulch, rubber mats, or some other impact-absorbing material. That no matter how hard you look there’s nary a tripping hazard to be found.
Reflect now on the playgrounds of your youth. If my childhood is anything to go by, plenty of playgrounds had raw dirt beneath the swings and slides. Many structures were exposed metal, not padded in the slightest. Slides were steep, and objects were unforgiving. (I recall a time where every park I visited had a tetherball pole. Nowadays I suspect they’re considered brain injuries waiting to happen.)
At my elementary school, the swings were secured to the ground (of course!), but it was loose enough that with concerted effort (multiple children swinging high and in sync) you could cause the entire set to tip just the tiniest bit. There was a slight mound over by the fence where a child could lay down and be out of view of the playground monitors, getting into all kinds of mischief. These kinds of peccadillos were litigated away over the years.
And beyond the features of the playground itself is the rise in adult supervision. Spend an afternoon trying to find a child at a park without a watchful adult nearby. Spend a week. I wish you luck.
This isn’t nostalgia for the “good old days”. Some things are better today than they were when I was a kid. Television is interesting, clothes fit again. If all this commitment to safety resulted in an increase in safety, then I’d say it was well worth it.
But it hasn’t. Even controlling for population growth, there were 25% more emergency-room visits related to playground accidents in 2012 than there were in 1980. Kids have a need to take risks and if we take away the obvious ones they only find newer and more creative methods.
(Aside: helicopter parenting might play a role in that. Did you know that going down a slide with a small child in your lap dramatically increases the chances the child will break his or her leg? I didn’t know that either until I broke my kid’s leg and the orthopedic surgeon told me that this is apparently common knowledge among orthopedic surgeons. Seems like something they should be telling parents! Not relevant to anything else, but consider this a PSA for anyone with small kids.)
So risk compensation. It’s probably a thing. And if the “risk homeostasis” hypothesis is true, that means we shouldn’t judge the effectiveness of a safety measure by an increase in safety, but by an increase in risky behaviors.
And now, football.
We are conditioned to think of sacks as an outcome of the competition between offensive linemen and defensive linemen on every play. Moreover, given that we attribute sacks to DLs much more readily than OLs, we’re conditioned to think of them as something a pass-rusher does more than something a blocker allows.
In other words, in terms of the hierarchy of who a sack “belongs to”, it goes pass-rusher > blocker > quarterback (who we generally picture as a sort of MacGuffin, an ancillary object that serves only to provide motivation and momentum to the true conflict).
Recent research (and not-so-recent research!) has greatly complicated this picture.
For instance, here’s a piece by Jason Lisk in 2009 titled
What quarterback rate stats stay most consistent when a quarterback changes teams? And wouldn’t you know it, the statistic that stays most consistent when a quarterback changes teams isn’t yards per attempt or touchdown percentage or interception rate… it’s sack rate. When a quarterback changes locales he tends to bring his sack rate with him and leave his other stats behind.
Here’s a follow-up from three weeks later titled
What quarterback rate stats stay most consistent when a team changes quarterbacks? that finds that yards per attempt and touchdown percentage stay somewhat consistent (i.e. they belong at least partially to the team itself), but there’s no correlation at all between a team’s sack rate with one quarterback and its sack rate with another quarterback, suggesting that the sack rate “belongs” to the quarterback himself and not to the offense as a whole.
This second study mirrored
an even earlier piece by Michael David Smith back in 2003 suggesting sack rate belonged more to the quarterback than the team around him. In 2011, Monte McNair
tracked the rate at which various quarterback statistics stabilized (represented 50% quarterback skill and 50% random chance) and found that sack rate stabilized significantly faster than any other so-called “quarterback stat”.
Last year, Lisk
returned to the question looking at QBs who didn’t change teams. He
followed up by asking what percent of sack rate was the responsibility of the quarterback, finding quarterbacks controlled 45% of the variance in sack rate vs. 44% in completion percentage, 40% in interception rates, 38% in yards per attempt, and just 34% in touchdown percentage (to name four stats that people don’t hesitate to consider “quarterback stats” despite being less in a quarterback’s control than sack rate).
Indeed, the body of research is now robust (and getting more robust all the time), and it all points in the same direction– the primary responsibility for taking sacks, hits, and even pressures falls not on pass rushers or pass blockers, but on the passer himself. There’s a small contingent of true believers in this concept, and we identify ourselves by our repeated assertions that sacks are a quarterback stat. (I’ve joked on Twitter that “sacks are a quarterback stat”
is my Freebird.)
So let’s say you’re ready to join me in the resistance and concede that individual quarterbacks to a large extent control their own sack rate. Awesome, and welcome! Now let me complicate the picture for you, because that’s just what I do.
Are Pressures The Best Way To Measure Pressure?
ESPN this offseason dropped a bombshell piece titled
Pass blocking matters more than pass rushing, and we can prove it, which I find to be a measured, nuanced, and well-reasoned piece of analysis. As far as I’m concerned, yes, they have proven it. Or at least they’ve successfully shifted the burden of proof onto anyone who wants to argue otherwise.
So… cool. We thought that the hierarchy went pass rushers > blockers > quarterbacks. Now we can just reverse that and say it’s quarterbacks > blockers > pass rushers and call it a day.
Except, well… it’s complicated. I mean, now we suddenly have to explain why GMs give so much money to pass rushers. Is it that GMs are dumb? No. GMs (
like coaches) are smart. I mean, “running backs don’t matter” Twitter likes having itself a victory lap acting like they discovered some new truth, but RB salaries have been trending hard down for a decade, so they’re really just saying the same thing GMs have been saying (with their dollars) for quite a while now.
And it’s really important to note that the ESPN piece doesn’t say that pass blocking matters more than pass rushing
in terms of pressure rates. The piece claims it matters more in terms of
overall winning percentage and
overall offensive/defensive effectiveness. Indeed, Eric Eager– resident data scientist at Pro Football Focus– tells me
their data suggests that defensive linemen contribute more to pressure rates than offensive linemen. Hooray for a complicated mess of conflicting data!
(One would also like to cynically note that each given play typically features more players devoted to pass blocking than players devoted to pass rushing; if pass blocking wasn’t a bigger contributor to team success that seems like it would be a misallocation of resources. The thing that 5/11ths of your offense is doing should have a larger impact on offensive success than the thing that 4/11ths of your defense is doing. Like, a 25% bigger impact, since you’re devoting 25% more players to doing it. Just from a naive/intuitive standpoint.)
How do we cut through this mess and find a coherent narrative through-line? My proposal is that we stop looking at sacks and pressures as isolated failures or successes and start looking at them as a general global risk.
At the NFL level, even the worst offensive linemen in the league are capable of keeping a quarterback upright long enough to run a play. A team with a terrible line against a stellar pass rush could call nothing but runs and 3-step drops, and they’d probably escape with few-to-no sacks surrendered.
The problem is that “nothing but runs and 3-step drops” isn’t a very effective offense, and “minimizing the number of sacks taken” isn’t the goal of football! Instead, teams call more aggressive plays, knowing that they’re accepting some level of risk but also that this risk comes with commensurate rewards.
Why are sacks a quarterback stat? Because each quarterback differs in terms of the amount of risk he’ll tolerate. Why does pass blocking influence wins even if it doesn’t have much impact on sacks taken? Because a line that is good at blocking allows for better and more aggressive play at any given level of risk. Why are pass rushers so prized by GMs? Because individual pressures really are a function primarily of the quality of the pass-rushers.
Again, as I noted at the top, “risk homeostasis explains everything about NFL pressure rates” is just a hypothesis. I don’t have any evidence to prove this. If someone comes along with evidence that says this isn’t what is happening, you should listen to them no matter how plausible this hypothesis sounds.
But once you start seeing pressure rates in terms of risk homeostasis, so many things start to make sense.
I’m going to use Russell Wilson as an example. In part because Russell Wilson is so hard to explain based on the traditional model where pressure is a result of DL vs. OL, with DL being the most important contributor, OL being the second-most, and QB being the least.
I’m going to use Pro Football Focus’ pass-blocking unit grades here as a proxy for how good a player’s pass protection was in any given year. Last year, per PFF, Wilson’s pass protection was either the best (by ordinal rank) or 2nd-best (by raw value) of his career, up with what he received his rookie year. On the other hand, 2016 and 2017 featured the two worst pass-blocking units of his entire career, truly dreadful outfits that ranked 30th and 32nd in the league, respectively.
Interestingly, when playing with the best pass blocking unit of his career, Wilson also had the highest sack rate of his entire NFL career (10.7%). Meanwhile, while playing with woeful protection in 2016 and 2017, Wilson set the two lowest sack rates of his career (7.0% and 7.2%). In fact, for Russell Wilson the correlation between pass blocking grade and sack% is a strong 0.66, which means for his career the better his pass blocking has been the more sacks he has taken!
Now, I don’t think PFF is perfect, and I do worry that there are some entanglement issues with grades like these. (An example: Denver ranked 26th in pass blocking in 2011 and 1st in pass blocking in 2012 with largely the same personnel. Was this a genuine improvement, or was this because the Broncos added one of the top sack-avoiding quarterbacks in NFL history? I don’t know.)
I’m okay with these entanglement issues because in this case they should be working in the opposite direction of our the observed effect. The more entangled you think pass protection grades are with quarterback skill and outcomes the more you should expect to see good pass blocking grades in seasons with low sack totals, which makes the observed phenomenon even more surprising.
How can we reconcile the fact that Russell Wilson had the best pass protection of his career and yet also had the worst sack rate of his career? Risk homeostasis! Because Wilson felt comfortable behind the line of scrimmage for the first time in a long time, he was willing to take bigger risks. Those bigger risks resulted in more sacks; he essentially “consumed” the safety improvements.
But he gained something in the process. Consider: on passes to Tyler Lockett in 2018, Russell Wilson had a perfect 158.3 passer rating and averaged 13.8 yards per throw. Wilson felt comfortable and therefore attacked down the field more successfully. That willingness to attack down the field resulted in more pressure, but it also resulted in a dominant season of downfield passing, which provided more than enough positive value to offset the negative value of the sacks. Better pass blocking didn’t translate to more sacks, it translated to better offense (with more sacks simply being a byproduct).
In fact, the sample size here is truly abysmal (just four seasons so please take this with a whole heaping mountain of salt), but I find it really intriguing that the correlation between Seattle’s pass protection grade and Tyler Locket’s yard-per-target average is an unbelievably strong 0.86. The better Seattle protects Wilson, the more efficient Tyler Lockett becomes. Again, small samples so this isn’t dispositive. But it’s a very interesting result, one that raises compelling questions and areas for further study.
The Risk Homeostasis Hypothesis:
Why does pass blocking correlate so well to overall team success despite the fact that it has minimal impact on the thing we would naively expect it to impact, pressure rates? Because the effects of good pass blocking show up somewhere else in the offense entirely. They show up in the quality and efficiency of throws that are being made!
Why do GMs keep throwing money at pass-rushers? Because pass rusher quality is pretty much the only thing that translates directly to an impact on the bottom-line pressure rate (short of replacing your starting quarterback with another one with a different risk set-point).
Why does study after study demonstrate that the preponderance of the responsibility for sacks taken falls on the quarterback himself rather than the players actively rushing and blocking? Because the quarterback is the guy with the football and therefore ultimately the guy who determines what level of risk he is willing to tolerate before getting rid of the football.
And perhaps we’d all have realized this a bit sooner if only we’d paid more attention to German taxi drivers and per-capita playground injury rates.