Philadelphia Eagles and the Dangers of Disestablishing the Run
Establishing the run is a silly, outdated idea. The Philadelphia Eagles are trying to prove this year that disestablishing the run is even sillier.
If you have watched the Eagles at all this season, you know their offense consists of:
- Wide receiver screens.
- Deep shots along the sidelines, the most successful of which are often Jalen Hurts underthrows which result in pass interference.
- More wide receiver screens.
- Holding and false start penalties.
- RPOs, except with no "R," which means there is really no "O."
- Occasional read-options where Hurts almost always keeps the ball.
- Rare handoffs to Miles Sanders. The Lincoln Financial Field crowd responds to each one as if Bruce Springsteen just walked onto the field. And started calling plays.
To get an accurate sense of how rarely the Eagles run, I took their run-pass ratio from the first halves of games (they sat on the clock a bit against the Falcons and Panthers) and reclassified 11 Hurts scrambles as passing plays. I did the same for the whole league, of course. Here are the NFL's lowest true first-half run-pass ratios:
Walkthrough hereby proposes a hypothesis: if your team runs the ball less often than the teams with Patrick Mahomes or Tom Brady at quarterback, then you are not running the ball enough.
And here's the kicker: the Eagles rank third in rushing DVOA. They refuse to do the one thing they may be able to do as well as teams like the Browns.
One rationale for the Eagles' low run-pass ratio is that they use screens the way most teams use handoffs. But the Eagles' screen game has netted diminishing returns. Per Sports Info Solutions, the Eagles executed eight screens for just 17 yards against the Panthers and three screens for 10 yards against the Buccaneers. If we expand screens to the colloquial sense of "passes at or behind the line of scrimmage," the Eagles executed seven of them for 16 yards against the Buccaneers. The more a team noodles behind the line of scrimmage, the more likely the defense is to just hold back and tackle the receiver for a minimal gain.
Also, the Eagles have been flagged for having an ineligible receiver downfield six times. If you're trying to revolutionize the screen game, you'd better at least execute properly.
The Eagles use play-action on about 19% of their first-half snaps. Their adjusted net yards/attempt on play-action of 5.5 ranks 25th in the NFL. There is a ton of evidence that suggests there is no correlation between play-action success and rushing success, but that principle hasn't really been tested at such extreme tolerances. At some point, even the oldest-school defensive coordinator will watch the film of a team that only runs 24% of the time and say, "Fellas, ignore the run keys and just drop into coverage." Indeed, it looked as though Cowboys and Buccaneers defenders might have done just that when Sanders was finally given his second or third opportunity of the game and ripped off a few big runs.
There's a theoretical ideal NFL run-pass ratio range which depends on a team's personnel and the game situations they face. It may well hover below 30.0% for a team with Brady or Mahomes at quarterback, established playmakers across the skill positions, and Bruce Arians or Andy Reid supervising the game plans. The Eagles have novice coaches, a quarterback with 10 career starts, and speedy but inexperienced wide receivers. Even after the Zach Ertz trade, the strength of their offense is Sanders, the offensive line, and tight end. They need to be "balanced"—we're talking about running the ball at least one-third of the time in neutral situations here, not caveman stuff—if they hope to remain competitive and give Hurts a fair chance to develop.
There's a valid argument to be made that the Eagles' underlying problem isn't a low run-pass ratio, but the fact that Nick Sirianni and his coaches may have no idea what they are doing. That leads us directly into our next segment.
Leaderboard of the Week
Every Thursday, Walkthrough examines a random (and usually obscure) leaderboard from Football Outsiders, Sports Info Solutions, or elsewhere on the analytics Interwebs in search of deep truths and wisdom.
Let's investigate passes over the middle of the field. The data comes from Sports Info Solutions and only includes passes marked "middle" in their database, so we are essentially working between the hashmarks.
We'll start with the yardage leaders in passes over the middle of the field:
Sam Darnold likes to throw to DJ Moore out of the slot. Moore is 7-for-7 for 83 yards on passes over the middle, mostly quick slants and other in-breaking routes from the slot. Robbie Anderson, on the other hand, is just 1-for-4 for 7 yards and two drops over the middle, including a few incomplete bombs. Chuba Hubbard is 4-for-5 for 30 yards on passes clearly designed for Christian McCaffrey. As with all things Panthers-related, don't be surprised if their statistics dip once their early wins become diluted in the data.
Mark Andrews is 6-of-10 for 69 yards over the middle. Lamar Jackson has also gotten Sammy Watkins and Marquise Brown more involved on such passes: 9-of-13 for 168 combined. The big story with Jackson, of course, is that he is no longer relying exclusively on passes over the middle.
Christian Kirk is the middle man for the Cardinals: 9-of-10 for 141 yards. Maxx Williams caught six passes over the middle before his injury; Zach Ertz will soon assume that role. Cardinals receivers have very clearly defined roles: DeAndre Hopkins is the go-to guy, usually on the left side; Kirk works underneath between the numbers; A.J. Green the boundary (usually on the right side); Rondale Moore is Little Screens Dude; the tight ends do tight end stuff. It's possible that the roles are too clearly defined and that opponents will soon figure out countermeasures. That's part of what happened to the Cardinals last year, but they appear to have improved in a variety of areas this season.
Noah Fant is 7-for-7 for 60 yards in this category. Courtland Sutton is 4-for-4 for 90 yards. As the data suggests, the Broncos offense right now consists of Teddy Bridgewater waiting in a collapsing pocket for one of them to get open in the 5- to 15-yard window. Like the Panthers, the Broncos may be headed for a team-wide statistical cliff due to injuries and their creamy early schedule.
The Chiefs lead the NFL in attempts over the middle. Travis Kelce is 10-of-12 for 148 yards. Jet sweep passes are technically "passes over the middle," and Mecole Hardman has three glorified end-arounds for 33 yards in this category. The high interception rates of the Chiefs, Ravens, and Cardinals are likely the residue of their daring, unconventional offenses: throw a lot of passes into heavy traffic, and some of them are going to get picked off.
Now for some real fun: the teams with the fewest yards on passes over the middle.
Those Saints numbers are a big red flag for a team with playoff aspirations. They simply have no presence over the middle of the field, and there's no tight end or wide receiver on the active roster who looks like a potential weapon over the middle. Oh, and all three of the interceptions were by Jameis Winston, so we can't blame The Lovechild for this one.
The Bears offense is utterly dysfunctional in ways that become more mystifying every time we peel back a layer. The Jets are … you know. To ease back on the Carson Wentz snark for once, he has been working the sidelines fairly well while playing on bum ankles. The Colts aren't actively avoiding the middle of the field like the other teams on this list.
And then there are the Eagles, who have attempted the fewest passes over the middle of any team in the NFL. You know, the team that had Ertz until last Friday and still has Dallas Goedert at tight end? The team that only runs 23.8% of the time? The team with Miles Sand … forget it. And take a look at that ANY/A, which is higher than any team on the over-the-middle passing leaderboard. (The Seahawks lead the NFL in ANY/A for this category, though not for long).
So the Eagles appear to excel at rushing and at throwing over the middle. They also appear built to do both those things well, even with Ertz gone. Yet they do both of those things far less than any other team in the NFL.
That's a quality control issue. It's a coaching experience issue. It's a freakin' common sense issue.
Sirianni better figure out the obvious in a hurry, because his career, Hurts' career, and the future of the franchise depends on it.
Five to Watch
This week's edition of Five to Watch focuses on emerging stars and unheralded free-agent additions on defense.
Chidobe Awuzie, Cornerback, Cincinnati Bengals
Awuzie has been one of the Bengals' best players on either side of the ball this year. And it's not like he hasn't been tested: Awuzie shadowed Davante Adams for most of the Packers loss and battled him more or less to a draw. (Adams caught some passes, but Awuzie recorded an interception). Awuzie also took care of Justin Jefferson in Week 1.
Awuzie has been shadowing No. 1 receivers frequently this season and may draw Marquise Brown this week against the Ravens. He will also likely be tested as a run defender and in second-stage coverage when Lamar Jackson does his thing. One thing is certain: Mike Nolan should not be allowed within 500 feet of even a Pop Warner field after making a Cowboys defense with Awuzie and Trevon Diggs look ridiculous last year.
Rashan Gary, ER, Green Bay Packers
Gary developed slowly in the shadows of Za'Darius Smith and Preston Smith for two years after the Packers made him the 12th overall pick in the 2019 draft. But Gary has stepped up in Za'Darius' absence with three hurries against the Bengals in Week 5 and four against the Bears in Week 6 (per Sports Info Solutions).
The Packers face a Washington Football Team that does a pretty good job defending their quarterback (if nothing else). Gary must convert some of those hurries into sacks before the playoff race heats up. Sunday is as good a time as any to start.
Terrell Lewis, ER, Los Angeles Rams
Lewis could barely get on the field after the Rams drafted him in the third round last year. Stepping up in relief of injured Justin Hollins, Lewis now has a sack in three straight games and nearly intercepted a pass against the Giants.
Lewis left Alabama as a size/speed project with a raw game and injury concerns. With their top-heavy payroll, the Rams need mid-round picks like Lewis to pay off in a big way, not necessarily to beat the Lions, but to keep up with the Buccaneers and (it still feels so strange to type) the Cardinals.
Elijah Molden, DB, Tennessee Titans
Speaking of teams that need young players to step up, the Titans need all the help they can get if they hope to turn Monday night's win over the Bills into more than a one-game blip. Molden, a third-round pick out of Washington, was billed as a versatile, pint-sized Budda Baker-type over the middle of the field, but he delivered more big whiffs than knockout blows early in the season. Molden is now coming around: he forced a fumble against the Jaguars, then both stuck his nose into run defense and held his own against Cole Beasley (who beat other Titans defenders for big plays) against the Bills.
Elijah Molden with the forced fumble that leads to the #Titans first touchdown of the game 😤💪 pic.twitter.com/lteexeVRsJ
— Titans Radio (@titansradio) October 10, 2021
The Titans face the Chiefs this week, and both cornerback Kristian Fulton and first-round pick Caleb Farley are injured. You can fill in the rest.
Bobby Okereke, Linebacker, Indianapolis Colts
Okereke had an awful game against the Ravens in Week 5, but unheralded third-year linebackers are bound to have awful games when forced to cover Mark Andrews all night. Okereke bounced back with 14 total tackles against the Texans last week. Like Darius Leonard, Okereke rarely leaves the field and gets tasked with some tricky coverage assignments.
Okereke and Leonard will catch a break with George Kittle on IR this week. Still, the 49ers will do lots of things to stress the Colts linebackers, from lining Deebo Samuel up in the slot or the backfield to designing some runs for Trey Lance (either as a starter or a Wildcat). If Okereke keeps playing like he has for most of this season, the Colts just might be able to crawl back into the playoff picture.
Thursday Night Sportsbook: Denver Broncos (+2) at Cleveland Browns
This spread has bounced around all week, going as high as Broncos +4.5. That will happen when both injury reports require a six-minute download. It plunged to +2 within an hour of Baker Mayfield getting ruled out on Wednesday afternoon and could fall lower by kickoff. (It actually dropped a half-point during copy editing).
Walkthrough normally loves a veteran backup like Case Keenum in his first start of the season, but the Browns will also be without Nick Chubb and Kareem Hunt. Tackles Jed Wills and Jack Conklin have not been ruled out, nor has Odell Beckham, and Jarvis Landry is expected back soon. There might be some wisdom to the Browns just resting all of them on Thursday night, taking one on the chin, and returning from the mini-bye closer to full strength. If that's what they are thinking, you don't want to be the one who wagered on them, even if the line slips closer to even odds.
Walkthrough also has little faith in the Broncos, so we're going to pass on them as less-than-field-goal favorites. But since TNF always needs a little action, we'll take the Broncos to score over 20.5 points at +100. The Browns will likely go three-and-out a lot, giving Broncos field position that their local-train-to-nowhere offense can at least convert into multiple field goals.
53 comments, Last at 25 Oct 2021, 10:03am
#15 by Harris // Oct 21, 2021 - 3:36pm
It's impossible to judge Hurts because Sirianni isn't helping him at all. It's difficult to judge Sirianni because he clearly has no interest in building an offense to maximize Hurts' strengths. That said, a coach who appears to have no interest in making changes to one of the worst offenses in the league tells you a lot about him.
#21 by Pat // Oct 21, 2021 - 4:20pm
So, it's actually funny, because I actually completely disagree with Mike's actual idea (that the Eagles need to run more). Philly's offense is almost entirely "pointlessly attack the area within ~2 yards of the line" combined with "chuck it downfield if some guy got hilariously open or you're effin' desperate."
And I think the reason for that is a combination of the fact that Hurts just isn't that accurate (so you can't run plays into tight defenses) and the Eagles receivers just aren't good. There are lots of plays they should be running (high-lows, double-moves at ~10 yards deep-ish) but if they ran them, they'd be horrible.
#22 by theTDC // Oct 21, 2021 - 7:18pm
Brian Schottenheimer may have a very well deserved reputation for being a bad coach, but I do remember when he coached my Rams. Those years they typically had something like the 20th best offence, and I thought Schotty was holding them back. Well he left and then the offense got way worse under Fischer. Turns out his conservative quick passing game was just what the talent (Bradford at QB, Amendola at slot WR, terrible OL, no real other weapons) called for.
*After looking it up, the Rams went from ~-30% DVOA in 2011 before Schotty, to averaging around -8% DVOA with Schotty, to immediately dropping to -16% then -35% in 2015 and 2016, the year and year after he left.
Now Schotty might have just gotten lucky. He seems to have the same offence everywhere, and it just happened to match the rams roster pretty much perfectly. It's easy as a fan to over-evaluate the talent on the roster, and blame coaches for not calling certain plays. Well sometimes the coach is an idiot, but they do see their roster in practice. Sometimes the guys just don't have the talent to do the thing the fans are hoping for.
#23 by theTDC // Oct 21, 2021 - 7:22pm
Yet another example of confounders that you mentioned. Deep passing is good, but also teams that have:
1) A great QB with a strong, accurate arm.
2) A great OLine able to hold up in pass protection
3) Receivers that can win contested jump balls down the field. (Also fast like prime DJax to get open down there)
4) Others I can't think about.
...throw the ball deep more than bad teams that don't have any of those pluses. So it's hard to figure out just how good deep passing is.
(Also, sometimes the playcall is a dumpoff to the HBack if no one is open downfield, and this shouldn't be considered a short pass playcall, but rather a deep pass playcall that failed).
#35 by BJR // Oct 22, 2021 - 6:36am
And, as you imply with your final sentence in parenthesis, deep passing success depends not only on the context of the offence, but the context of the play itself. Just in order for a deep pass to be attempted, implies that certain things must have gone right early in a play. E.g. the defense has given the correct look; the pass protection has held up; receivers had a mostly free release, etc. 'Failed' deep passes often look like sacks or checkdowns, or even runs that were checked into before the play, rather than incompletes or interceptions.
(I believe I first read Pat making this point on these boards a short while ago, so issuing credit here.)
#38 by Pat // Oct 22, 2021 - 11:32am
Another really interesting similar point to that is the run audible: if Peyton Manning checks to a run because he knows the pass they have won't work, if they only get 1-2 yards from the run, how is it fair to compare that run play to a pass play? That's literally a failed pass play, not a run play.
#39 by Pat // Oct 22, 2021 - 11:37am
He's a second year QB. He's a good backup, not a project.
I don't think they should move on from him yet, but that's only because they need the money. Total rebuild time. Start drafting OL/DL/CB (which they should've freaking done last year), wean yourself off the painful contracts, grab cheap WRs/RBs/TEs and churn through them. When you've got a solid new base, swing for the fences for a new QB and hope you hit.
#44 by DIVISION // Oct 22, 2021 - 5:47pm
If we go by games played. I'm only going off what I've seen of him in Siriani's offense. Maybe he just needs more time?
Your'e right about Philly, though. Complete rebuild time.
Trading Ertz to the Cards is the start of that.
#48 by Noahrk // Oct 23, 2021 - 11:41am
If history teaches us anything is that any QB can eventually make it if given enough time. But what I've sen from Hurts is not encouraging. I'd cut ties, unless as Pat suggests, the cap makes it forbidding.
#52 by Pat // Oct 25, 2021 - 9:55am
It's not that the cap makes it forbidding (he's a 2nd round rookie, his contract is free) - it's that it'll cost more to replace him with a QB of equal value. Hurts is ~replacement level, meaning around $5-10M in value on the free market. He costs way cheaper than that.
The Eagles need at least until 2023, ideally 2024 to shed most of the dead weight on the roster (they're leveraged really bad). Hurts is the perfect kind of QB for that - cheap, and one that you can plausibly sell to people as a "developing star" with just enough competence to sell it.
Unfortunately Roseman's an idiot and I'm sure they'll do something stupid, like trade the house for Watson to try to win next year or something.
#53 by Pat // Oct 25, 2021 - 10:03am
If we go by games played. I'm only going off what I've seen of him in Siriani's offense.
The coaches are clearly not giving him plays where they trust his judgement or accuracy. That's the entire reason why the offense looks so weird. You virtually never see receivers hit in stride, for instance. And if a second-year QB can't pick things up quickly (at least enough for coaches to trust them) to me that's an indicator that his ceiling's just not high.
Trading Ertz to the Cards is the start of that.
Yeah, I won't buy that they're in rebuild mode until they start trying to get rid of actual talent. Ertz was clearly expendable with Goedert on the team. They've got $100M guaranteed invested in Fletcher Cox (31), Lane Johnson (31), and Brandon Graham (33). If they're all on the team week 1 of next year, they're screwed.
#31 by justanothersteve // Oct 21, 2021 - 9:09pm
It's difficult to judge Sirianni because he clearly has no interest in building an offense to maximize Hurts' strengths.
Your statement, in itself, is a judgment. That statement alone shows Sirianni should not be any team's OC. It's the job of the OC to construct an offense around the players as theTDC points out above with his Schotty example. I've always admired Belichick as being one of the best at building offenses and defenses around his players' strengths and covering up their weaknesses, even from game-to-game.
#45 by DIVISION // Oct 22, 2021 - 5:49pm
We could make the same statement regarding Shanahan and Trey Lance, right?
He sure didn't look interested in creating a gameplan for Lance versus Arizona.
It looked like he was giving away the game on offense.
#13 by theTDC // Oct 21, 2021 - 3:18pm
And for the most famously awful football statistical analysis we have Malcolm Gladwell. Gladwell claimed that there was no correlation between where a QB was drafted and their NFL success. In order to “prove” this hilariously incorrect statement, he showed a study that showed only a small correlation between QB rating and draft position.
Now first of all, small correlation is not no correlation, but everybody could have just told Gladwell that late round prospects that suck never see the field, whereas even the worst first round prospects get a chance to play (and suck). Gladwell was even saying garbage like “you know it’s so funny that teams spend all this money on scouting, when science has proven that nobody really knows anything about the draft.”
I feel like there are some great people doing analytics for football. There are also Gladwell’s out there, who just want some way to feel smarter than the DumbDumb coaches in the NFL, who stupidly think that you should try to draft a QB high in the draft if you want a really good one.
#28 by DIVISION // Oct 21, 2021 - 8:09pm
When I read Malcolm Gladwell, I do so with a caveat. He's going to give me a different perspective that I perhaps haven't taken in to consideration before.
That said, I don't draw conclusions from his assertions. Anyone with a background in stats/research methods knows about confounding variable, sampling errors et al., so I don't think Gladwell is speaking to the analytics community with his books and articles.
He's speaking to the general public who have no idea what a correlation is, much less a positive or negative one.
#30 by theTDC // Oct 21, 2021 - 9:08pm
Malcolm Gladwell is speaking to an audience of chattering class midwits who want to feel smarter than everybody else, without actually being smarter. The type of people who believe themselves to be Critical Thinkers because the WMD Liars tell them they are. Who believe they are Science Lovers because Neil Degrasse Tyson tells them they are. Who believe themselves to be intelligent, because Malcolm Gladwell tells them how stupid everyone else is. These are the types of people who live in some sort of anti-engineering world, where the truth is irrelevant, only whose daddy was who and who is at what position at such and such corporation.
Gladwell's bloviating that NFL GM's are just a bunch of dumbdumbs who are stupidly wasting draft capital on the QB position is not a "perspective," it's hilariously stupid disinformation. Like it's just flat out incorrect. It's just wrong. It's so wrong that I feel stupid for even making the logical arguments for why it's wrong instead of just saying "LOLOLOLOL, people who read Malcolm Gladwell are retards."
I thought that Gladwell's career was over, in part because even the midwits who used to read him started to realize that when he started talking about a subject they actually were experts in that his analysis was bafflingly idiotic. There's only so many times these people could hear "you know I like Gladwell but as an expert on military history I had to say that his claim that underdogs win militarily almost 40% of the time is just too wrong to even begin." Or "Gladwell tried to say that the 1996 Kentucky Wildcat NCAA champions were an untalented underdog team, despite them putting all 5 starters to the NBA for a combined 60 seasons. Oh and also three bench players had nice NBA careers."
Once these people started hearing that Gladwell was basically full of shit it removed the value of the product he offered them. Which was not actually informing them, but rather allowing them to pretend to be informed. The unearned badge of Smart Critical Thinker. It just doesn't have the same value when you pull up some midwit Gladwellism at a cocktail party only for someone to say "but didn't he say that NFL QBs are equally as good in the seventh round as the first? Seems kind of full of shit."
But apparently DIVISION is a fan.
#43 by DIVISION // Oct 22, 2021 - 5:42pm
As I said before, I don't draw conclusions from his work. I think some of his writing is interesting and some is bullshit precisely because he draws conclusions from flawed research design. It looks cute in a book, but it's just entertainment.
That's how I see Malcolm Gladwell.
I first became aware of him through Bill Simmons, pre-Ringer. Simmons is the type of guy who buys in to Gladwell. This is true.
#12 by theTDC // Oct 21, 2021 - 3:10pm
All this reminded me of perhaps the most all time ridiculous analysis on this site. Some guy claiming that teams should never punt. And yes, you heard that correctly, there was an article on this website that claimed that NFL teams should never punt. He had this bizarre complicated math to “prove” this assertion.
But let’s try to run this through a real world scenario. Let’s say that you are up by 21 points late in the 4th quarter, facing 4th and 40 from your own goal line. Apparently you should go for it instead of punting. But actually you shouldn’t, because that’s absurd.
While I do very much like the analysis on PA/running correlation, the conclusions simply have to be overly strong. A play action is a run fake. Why would a teams ability to run be un correlated to their ability to sell a run fake?
I think the only reasonable conclusion is that currently NFL defences are overly respectful of the run, so even bad running teams still get PA benefit of the doubt, assuming they do in fact run the ball fairly often. That, and defences key up to stop the running attack of great rushing teams, which misleads, because they get gashed against PA, but not the run. There may be other explanations as well.
#14 by Will Allen // Oct 21, 2021 - 3:34pm
Then again, we had numerous occasions when the Brady Patriots would supply weeks of video to an upcoming opponent, of Brady&Friends killing defenses with a thousand short and intermediate passes, peppered with a few deep balls. The defense would come out completely devoted to the task of defeating that attack, and then the Patriots would come out and run the ball down their throat for 3 hours, resulting in about as boring a 27-10 outcome imaginable.
Doing something different than what the opponent planned for is usually good, and having personnel which can do a variety of things well facilitates that effort. In other words, having superior talent makes playcallers look more clever.
#17 by theTDC // Oct 21, 2021 - 3:49pm
Something that doesn’t seem to be brought up that often is that teams can “establish the run,” on tape in previous games. Back in 2017/18 when the rams had Gurley not injured, and a great OLine, McVay would often make the very first play of the game PA. Teams were geared up to stop the run, so they didn’t rush very well, but they did throw for easy yards time after time.
It’s just the analysis of “if teams are afraid of the run, run fakes work better,” seems so obviously true that any claims to the contrary need an enormous amount of work separating out confounders.
#46 by ChrisS // Oct 22, 2021 - 7:44pm
I think a lot of the effectiveness of PA is taking advantage of deeply ingrained instincts in defenders. Most learned at an early age to react immediately to the handoff and these instincts are hard to overcome. Just one or two steps in the wrong direction I the NFL put you way out of position.
#9 by Will Allen // Oct 21, 2021 - 1:56pm
The goal on offense, presnap, should usually be to compel the 11 defenders to be planning to defend as large an area as is possible to compel them, or to overload to defend a subset of the area in a predictable way.
Running, in varying amounts, depending on personnel, is important to that goal.
#7 by Joseph // Oct 21, 2021 - 1:32pm
The Saints lack of passing over the middle is three-fold: 1--No Michael Thomas or Trequan Smith, who were supposed to be the starting WR's; 2--Had to let their top two TE's go in the offseason b/c of the cap; 3--Their reliance on rushing--Winston's high in passing yards for the season is 238!
IMO, once Thomas comes back from IR, the passing in the middle will pick up. Even Smith being removed from IR on Tues might help some.
#5 by Pat // Oct 21, 2021 - 12:08pm
There is a ton of evidence that suggests there is no correlation between play-action success and rushing success, but that principle hasn't really been tested at such extreme tolerances.
Ooh! Ooh! This is my absolute favorite statistical mistake.
Pretty much everyone learns at some point that "correlation doesn't imply causation," right? But people tend to always think the opposite is true - that causation implies correlation. Except it doesn't. Even if two things are perfectly linked, you can dilute the correlation coefficient arbitrarily.
And, even better, in the real world you can reverse the correlation coefficient.
Think about it this way: imagine your stove. Imagine if you plotted "position of your dial" versus "temperature of water in pot."
Think about how you cook. When you put the pot on the stove, the water's cold. The dial's "off." You want it hot, so you turn the dial up. There are now a bunch of data points "water cold, dial high." That dial stays the same, and the water heats up. And when it gets to the temperature you want, you turn the dial down. So now you have "water hot, dial low."
Plot 'em, and what do you get? A slight negative correlation coefficient (actually mostly uncorrelated). Turning down your dial apparently heats up the water. Think about how crazy this is: these two things are totally linearly related, and we get a crazy result because we don't know why the dial was turned.
How does this relate to football? Ask this question: if play-action is working awesome... why would you bother running the ball?
#11 by theTDC // Oct 21, 2021 - 3:00pm
This is a great comment. I’ve always somewhat scratched my head at the whole “PA and running success aren’t correlated,” because it is so intuitively wrong. I know that sometimes the most brilliant insights can be unintuitive things that are correct, but statistics can mislead and give flat out garbage results. Off the top of my head I wonder:
1) Are we taking into account that teams with great rushing attacks and bad passing games might rush the ball a lot and not do that much PA? In other words, are we looking at the increase WITHIN a team’s passing game, PA pass to straight drop back, or just an aggregate for the entire league?
2) Are we counting rushing success, or just rushing amount? It makes sense to me that a team with a poor running game and Peyton Manning might actually rush very successfully for 4.5 yards/carry, at 15 carries per game. A team with Adrian Peterson and SomeGuyAtQB might rationally rush 30 times per game. Per play success might be the same, but I think the LB’s and safeties will be playing the rush a lot harder in the second case.
3) Sort of adding on to point 2, the real measure of the success of the run/PA correlation, is actually just how well the defence bit on the fake. Everything after that is subject to massive unrelated noise like QB and WR play. So a better statistic would be watching film and trying to plot how out of position the defenders are after the fake.
4) Adding to Tanier’s point, if we rushed 0 times, nobody would respect the fake. It’s an interesting statistical result that teams might be overly honouring the run, meaning that there is no correlation between running rate/success AT CURRENT LEVELS and PA pass efficacy. That’s interesting, but doesn’t support “there is no correlation,” because obviously if we extend the results further to 0 runs we get an absurd statement that teams will still respect the run fake without any threat of the run.
I think that analytics guys can sometimes get really carried away, and make extremely strong statements when more cautious ones would do. “Teams should probably do PA more, because current defences are overly respecting of the run,” is a reasonable statement, although we’d have to actually do film analysis of the run fake itself. But “there is no correlation between run success and PA success,” is the kind of thing that gets coaches to tune out. I mean it is literally a run fake.
#20 by Pat // Oct 21, 2021 - 4:01pm
It's the best you can do with what we have, but because literally all of it is confined to observables from the offense, you can't conclude anything.
Play-action passes are designed to take advantage and manipulate of the defense. Without measuring what the defense is doing, concluding whether or not it does anything or what's needed for it to do something is just not possible.
It's like asking why teams don't run "chuck deep downfield after the defense jumps offsides" play more often. That play's awesome, they should do it all the time.
#25 by theTDC // Oct 21, 2021 - 7:39pm
And thank you for that article, but I'm pretty sure I read it at the time, and it doesn't address the points I or Pat brought up. The purpose of the run fake is to displace the defence. The question of PA efficacy versus the run rate/efficacy can quite simply not be solved without film analysis. You need to see with your own eyes if the LBs are still biting on run plays for teams that barely run, because that's the specific thing we want to measure.
I personally believe that teams probably don't run PA enough. I can certainly believe the general claim there. But there is a constant running thread amongst some, not all but some, analytics nerds that they've got this proof that "Hurr Durr establish the run," thinking is objectively wrong. I mean, I can believe that they're wrong, but there are probably some extremely bright borderline geniuses coaching football. McVay, Belichick, maybe Harbaugh. I can believe that there are lots of >120 IQ people coaching, and if they're saying things like "you need to be balanced and establish the run," coupled with watching tons of tape trying to exploit the opposing teams weaknesses, well that means something more than raw statistical analysis.
I still say, I would love if someone did a study where they charted how out of position the LB's, and secondary got due to Play Action. Maybe even if the DL stopped rushing the passer and instead tried to hold their gaps. That kind of study is what I want to see.
#32 by dank067 // Oct 21, 2021 - 9:27pm
Look, I say this as a fan of a team who blew a 97% win probability in the 2nd half of the 2014 NFC Championship game and whose play calling head coach said after the game that they had set out a target of 25 carries (!!!) in the second half to secure the win - most NFL teams don't have a very sophisticated run game, and they aren't being especially strategic in how they deploy it nor how they tie it into play action.
(And the Packers only led by scores of 16-0, 16-7, and 19-7 most of that 2nd half, anyway... uggggggggggggggggggggggggggghhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh.)
There's plenty still to study on this front, but Josh Hermsmeyer has analyzed "wasted yards" by linebackers reacting to play action before. He didn't look at how it correlated with rushing success, but he did show that the effect of tricking linebackers doesn't wear off over the course of a game. Would love to see more work using these data. Here's the link:
I did notice some numbers coming out this past summer starting to show that running success on a specific type of run play (I think outside zone) positively correlated with play action passing success when showing the same run action - and that makes all the sense in the world. But I still strongly suspect it has very little to do with rushing volume, especially in neutral situations like first down.
#40 by Pat // Oct 22, 2021 - 11:45am
But I still strongly suspect it has very little to do with rushing volume
Volume numbers will never make sense to study, because you don't know why they're called in the first place.
Do you have a link to those numbers you mentioned? A focused study like that is probably the best way to see something, but it's still super hard because again, it's uncontrolled: if you're trying to manipulate something else, you don't naively expect a positive correlation between the two things because of the feedback.
#47 by Noahrk // Oct 23, 2021 - 11:36am
Great post, Pat, very instructive. The way I usually look at the PA situation is like an evolutionary war. The defense developed the behavior that PA exploits a gazillion years ago and it worked very well back in the day that defenses won championships. Defense has been slow to adapt to change in that respect, but ironically, the harder the behavior is exploited, the faster it will change. Like Tanier hints at, when PA is over-abused, the defense will stop biting.
Unless, of course, as you suggest, we don't know nuthin' and PA is used just right for reasons we don't see.
Speaking of adaptability, it always strikes me how coaches who on one hand seem like geniuses can be so slow to adapt even slam dunk strategies such as going for two when scoring a TD down by 14. I guess we all have our blind spots.
Or maybe rigidity as a HC behavior works for some reason?
#49 by Will Allen // Oct 23, 2021 - 3:19pm
Coaches who aren't extremely secure, which is the vast majority of them, probably have short term incentive to rigidly stick to conventional wisdom. Head coaches who go 7-10 while being unorthodox are more likely to be fired than the 7-10 coach who adheres to convention. The irony lies in the reality that the less talented the roster, the more the coach should take a high variance approach.
#50 by jheidelberg // Oct 23, 2021 - 3:41pm
Not sure if you are a baseball fan Will, but whether or not, you would enjoy Brian Kenny's book, "Ahead of the Curve.". Your comment is similar to what he would say, that a baseball manager will do what is conventional, thinking of his next job interview, as opposed to doing what is unorthodox, and is analytically correct.
It is easy to do what "everyone else does" instead of new thinking which is what baseball and football analytics often require.
#51 by Pat // Oct 24, 2021 - 1:43pm
In some sense I feel like play action might be like the offsides free play: it only happens when the defense screws up. You could ask "why do defenses jump offsides, that's a terrible play" - but that's because you don't count sacks as "successful" offsides plays.
So I think it's a little tough to say that defenses are being dumb and slow to change: we don't know what behavior play action is exploiting, so we don't know what to compare it against. It's *really* naive to believe it's "all runs."
#19 by Pat // Oct 21, 2021 - 3:56pm
but statistics can mislead and give flat out garbage results
This effect specifically is a very common bias - it shows up in social sciences/economics all the time, because you can't generate controlled situations. In general, it's usually an unconfoundness issue (*) - as in, you've assumed that you're measuring all of the important variables ("unconfoundness," as in, there are no confounding variables). In the previous "dial on a stove versus temperature" issue, the confounding variables that aren't being measured are the heat capacity of the object on the stove (the temperature of the stove is related linearly to the dial position) and the difference between current temp and desired temp.
Football has a massive unconfoundness issue. Staggeringly massive. You can see it in literally everything that analytics-types try to test. Look at your own examples:
"the LB’s and safeties will be playing the rush a lot harder"
"just how well the defence bit on the fake."
"nobody['s defense] would respect the fake."
It's all defense. Football's issue is that the statistics contain basically nothing on the defense. And so the idea that you could ever evaluate a play choice as good or bad statistically is fairly crazy, because you're totally and completely missing what the play choice is actually trying to do.
Literally any information in the play-by-play on the defense would be better than what we have now. You could completely give up all the information on pass distance and direction and replace it with anything describing defensive alignment and you'd be better off.
(*: also called conditional independence, also known as "the underlying cause of most bullcrap studies that get pushed for political reasons")
#24 by theTDC // Oct 21, 2021 - 7:27pm
I heard someone remark once that there are only two things that social sciences can prove, the blindingly obvious, and the utterly bullshit. Trite, but worryingly true.
Anyway, my point with PA passes, also known as run fakes, is that the best way to see how the running game rate/success impacts play action is to do film study and mark how out of position the defence gets. After all, that's literally the entire point of the run fake. If the defence doesn't get out of position, then all you did was turn the QBs back to the play, giving him a harder time reading the defence. *
Looking at the overall success of PA plays is just the wrong way to go about things. It's asking for unseen confounders.
* Actually not even just that, a strong run game also encourages teams to load the box, which could result in easier passes, PA or otherwise.
#37 by Pat // Oct 22, 2021 - 11:23am
Yup, exactly. It's actually even a little harder than that, obviously, because you don't know what the defense was trying to do. Or possibly even what the offense was trying to do.
This actually gets back into my longstanding point that grouping plays by "run" and "pass" is just pointless nowadays anyway. We're so far past the point where this makes sense it's comical.
I mean, we're now somehow splitting passes into "normal pass" and "play-action pass." Why not split runs into "normal runs" and "draws" (which are pass fakes)?
And then we need to split out run-pass options, too. Those aren't runs, or passes. If a teams runs an RPO, QB reads run and they get 3 yards, runs it again and reads pass and gets 7 yards, you can't say "hey, you should pass every time, it's way better!" The team runs the first time because the pass wouldn't work. Same problem - missing the confounding variable of the output of the read.
But wait - scrambles are runs on pass plays. So they're not runs either. And screens are pass plays blocked like runs. So what the hell are they?
I don't even know how you could ever do this stuff.
#26 by Eddo // Oct 21, 2021 - 7:47pm
This all seems pretty accurate. I think we can probably say with a decent degree of confidence that play action increases passing success relative to non-PA passing, regardless of how well you run, but that it's much harder to say that all teams are affected equally.
At the very least, I think you'd need to take into account formations and tendencies, as well, thinking about your Colts/Vikings examples.
#33 by LionInAZ // Oct 21, 2021 - 10:19pm
This analogy is a poor one from a physics standpoint.
First, the addition of heat does indeed cause the temperature to rise, so the causation does lead to correlation.
However, you neglected two things when you turned down the dial:
1. The thermal inertia of the water. Water takes a long time to cool down because of thermal inertia. If you keep the heat low long enough, boiling water will stop boiling and cool down.
2. It also takes a large amount of energy to get the water from boiling to evaporation as steam, so the water will stay at constant 212 F until it has all been turned to steam. Once the water has boiled off, the applied heat will continue to raise the temperature of the pan until it gets much hotter than 212 F.
#36 by Pat // Oct 22, 2021 - 11:12am
It's... actually a really good example from a physics standpoint. Which I can say definitively, because it literally is a physics example. As in, I literally had someone try to do exactly this problem with a part from a physics experiment I built. Collaborator plotted heat load versus temperature, found a negative correlation and was confused as heck until I told them that's because it's temperature controlled, not heat controlled.
First, the addition of heat does indeed cause the temperature to rise, so the causation does lead to correlation.
I'm not sure you understand the bias issue I'm trying to say? There's a more common one which might help. I'll mention that at the end.
Of course adding heat to something causes temperature to rise. That's why I said these two effects are totally causally linked. Linearly, in fact!
The problem isn't in the theory. The problem is in the experiment. It isn't a controlled experiment! You're testing things in an environment you don't control, and specifically you're not measuring an incredibly important ("confounding") variable. The heat capacity is one thing (less heat capacity hides the effect), but the most important thing you're missing is that the temperature is controlled. That is, the dial position is a function of the difference between current temperature and target temperature. So the dial's position is actually negatively correlated with the difference in temperature. Hence the remnant negative correlation that shows up. In the simplest case if it's a (horrible, unstable) P-only control, the dial position is just D = -P*(Tcur - Ttarget) and you'd have a perfect negative correlation. The integral (I) terms dilute that.
The more "common" example that I've seen in economics is this: suppose you've got some disease, and you're trying to figure out if that disease causes death. You go out and find everyone who's contracted the disease, but virtually no one dies from it. You then say wow, and publish your results saying "this disease does not cause death."
And it's instantly rejected, because the reviewers say "um. You didn't even bother to find out what percentage of the people got treatment for it. Everybody gets treatment for this. That's why no one's dying from it." The correlation between the cause (the disease) and the effect (death) was eliminated due to a confounding factor (treatment).
Your experiment wasn't controlled: you didn't go out and say "pretty please can you get this disease and not get treated so I can see if you die?" In exactly the same manner, we're trying to conclude things about NFL head coaches without knowing what the confounding factors are. We need to go out and say "pretty please can you just run these kinds of plays regardless of what the opposing defense is doing?"
#4 by Coldmilk // Oct 21, 2021 - 11:49am
I love your articles. I’ve been reading them every week (or more) for 5 or so years now.
You have a knack for ripping teams back to earth, even when they look decent to the naked eye. You’ve been tearing my Seahawks apart for years, for example. I think most people tend to see good, flawed teams like this year’s Packers, Rams, Chargers, and Chiefs as legitimate contenders, but you seem to be a lot more selective with your Super Bowl favorites, more willing to point out weaknesses than heap praise on teams, even when they win.
Who do you think is legit this year? Any team you’re willing to praise? (Even the 6-0 Cardinals still get chuckles from you)
#6 by Mike Tanier // Oct 21, 2021 - 12:11pm
Gosh! I never thought of myself as ripping the Seahawks! I liked to rip Brian Schottenfailson, but I feel as though that's different.
Bills: not much to argue about except settling for field goals on MNF
Ravens: Love 'em but just keep bracing for playoff narrative bullsnot.
Rams: Very very good.
Cardinals: Can't find a real flaw besides "no track record" but also can't wrap my brain around them because they lack a track record.
Packers: Have about the same 15-20% chance of not self-destructing in the 2nd-3rd round as they always have.
Cowboys: Need to be more consistent on defense but are gonna cause a lot of havoc.
Chiefs: Mahomes won't go away quite this easily.
That's the Super Bowl list. For everyone else, the question is "what are you building toward" or "can you reach the playoffs and get a few wins, and is that a positive outcome for a team in your stage of development?"
#34 by Trizzle58 // Oct 22, 2021 - 1:04am
Even us Cardinals fans agree with this assessment. We are collectively doing the “I don’t know what to do with my hands” scene from Talladega Nights. I think a small part of us expects a loss to the Texans because “that’s the Cardinals for ya”.
#42 by DIVISION // Oct 22, 2021 - 5:37pm
The Cardinals of this season have so many vested veterans from other teams that I don't see letdowns like in past years.
Plus, with Watt and Hopkins primed for great performances against their old team, I expect a pretty standard 20+ point win.
They might score 50 against the Texans.
I want to see Watt score a TD on offense.
#3 by ImNewAroundThe… // Oct 21, 2021 - 11:47am
Their Dropback epa/play and success rate is still higher in the 1st half than their rush. Even just looking at y/a.
Maybe they need to elect to recieve more and start off with more runs but it seems like they're down quick and idk if running with "just" the 3rd best rush DVOA is the answer to get back into things.
They've completely flipped Hurts adot though.