East Coast Offense: The Limits of Statistical Models

This article is part of our East Coast Offense series.

If you want to explain how the world as we know it came to be, science has a much more plausible account than any of the world's major religions. A person arguing for any of the religious explanations will need to resort to beliefs that are not supported by observable evidence. As such, when it comes to the "how", scientific theories, bolstered, refined or refuted by testable hypotheses, give us the best chance at understanding.

But science has its limitations. If you ask a scientist why the universe came to be, he has no more insight than the layman. Some might argue there is no "why" - it's just atoms and molecules, biology, chemistry and physics - but there is no evidence for that view, either. Science does not lead us to believe there is no "why" - it simply has no opinion on the matter. The "why" is outside the scope of science. In fact, it's a question more suited to religion.

Understanding the scope of each discipline helps us avoid making category errors wherein we use the wrong framework for the question at hand. While not as well-defined as the science/religion dichotomy, I think a similar error is confusing the stats vs. observations debate that fills my Twitter feed every week and has now even spilled into the mainstream.

If you're developing a card-counting system for blackjack, a statistical model is a great tool. The model - using virtual cards and outcomes - is almost identical to the real-life game. You know exactly how many cards are in the deck, their frequency, their values and all the decision points in the game. There are no tired 18s that might not beat a 17 on a given day, no dealers that use different rules, no relevant environmental conditions. If you argue against a stats guy that you should hit on 17 against a 10, you're going to be wrong even if you get a four.

By contrast if you're a Pro Bowl NFL quarterback, you're much better off studying the defenses, practicing your plays and trusting your instincts than going through spreadsheet-based decision tree of what to do in each situation. There are so many variables - different pass rushers, formations and personnel in different downs and distances every play, in different venues every week - that you're better off trusting Tom Brady or Aaron Rodgers to read and react than to try and get him to do what your Real Live QB™ model says.

These are extreme cases, of course. The former is perfect for a statistical model, and the latter is terrible. But they serve as a framework to evaluate the analytics debate - are troglodyte coaches sticking on soft-17 against a 10 despite overwhelming evidence, or are analytics gurus purporting to know more than Bill Belichick about building an NFL roster?

The challenge then is to locate where a particular question lies on the blackjack to playing-QB-in-real-time spectrum before deciding what decision-making tool is optimal. I'll offer a couple examples:

(1) Going for two, down eight points after a score. We saw Doug Pederson and Pat Shurmur do this, and the math is fairly clear. The odds of missing both two-point conversions (a loss) are less than making the first one (a win if you make the PAT the next time.) Now, this isn't blackjack - there are personnel differences on both sides in each situation, and maybe weather plays a part - but it's usually similar enough that barring extreme circumstances, you should follow the math. (There are rare blackjack card counts where you depart from basic strategy too.)

(2) Always drafting a QB early in the first round if your team has a need. In my opinion, this is wrong. There is so much variation in QB quality, in the league's macro environment, in the talent at other positions, that saying "always" is the stats camp making a category error. They simply do not have the certainty to which they purport. And I'm not talking certainty of outcome - no one has that. I'm talking about process.

This is not blackjack where an ace is an ace, no matter what. Even if something is often or usually correct, it can be wildly off base in a specific instance. If we knew nothing else about the situation, then we would play the odds and do the thing that's correct most often. But we know quite a bit about the particular players and the league environment, so trusting a rule of thumb rather than looking at the specific evidence is foolish. It might well be the case that you look at the evidence and still draft the QB - in fact, I'd expect that to be the case most of the time. But to ignore evidence in favor of your rule is bad process. You should ascertain whether a given situation presents an exception - and you must be prepared to go against the grain if the specific evidence warrants.

I've found a lot of stats-splaining™ in my Twitter feed is dedicated to refuting people making weak arguments against going for two, or maybe that QBs aren't usually more valuable than running backs. It's easy to quote-tweet someone whose observation-based argument is obviously flawed and proclaim victory for your camp.

But I'd instead encourage people to understand the difference between situations where a statistical model is especially useful and those where the reality is far more complex. Once we locate the question on the spectrum (rather than pretending we can push all questions to the blackjack side of it), we can apply the appropriate analytical tools to answer it.

Often, we'll be left with uncertainty and forced to trust our instincts, but it's better to get comfortable with that than worship a false idol that's beyond the scope of its efficacy.

Week 9 Trivia

With Patrick Mahomes on pace for 52 TD passes and Andrew Luck on pace for 46, this week's trivia is all the QBs to throw 37 or more TD passes in a single season. (It's probably too easy, so I gave you only 150 seconds to answer it.)

Guessing the Lines

Week 8 Observations

• The Saints won comfortably, but Drew Brees barely lifted a finger – 23 attempts, 120 yards, 5.1 YPA, one TD, one pick and no sacks. The only players to participate at all in the offense were Alvin Kamara, Michael Thomas and Mark Ingram.

•Kirk Cousins, by contrast, had a big fantasy day – 359 yards, 8.8 YPA and two TDs, but he threw a crushing pick-six and took four sacks.

• As usual Adam Thielen got his 100 yards – 7-7-103-1, but he lost a fumble near the end of the half that turned the game around. Stefon Diggs (11-10-119-1) led the team in receiving while Latavius Murray (13-56-1) ran well and caught five of six targets for 39 more yards.

• The Cardinals comeback and cover were epic. I had them as my best bet and an entry in the Supercontest, and they looked dead to rights down 15-3 in the fourth quarter.

• I noticed Larry Fitzgeraldlooked healthy again in the Thursday night debacle against Denver, and he built on that game with a 12-8-102-1 line. He won't get yards after the catch or beat anyone deep, but he can still catch the ball in traffic.

•Christian Kirk made a nice catch for the go-ahead score in the last minute of the game. He had a modest 7-3-42-1 line, but he's the future there with Josh Rosen. Rosen looked shaky at times, but he engineered two final drives (three really, but Jermaine Gresham fumbled for no reason on one of them) and delivered accurate balls.

• Under new offensive coordinator and play caller Byron Leftwich, David Johnson got more work – 16 for 59 on the ground and four catches for 41 yards, despite leaving briefly to get checked for a concussion. It's not 2016 Johnson, but it's a start.

•C.J. Beathard can escape a pass rush, but he's inconsistent and herky-jerky in the pocket. Besides the one big TD to Marquise Goodwin, it was dink and dunk to Kendrick Bourne (10-7-71) and George Kittle (8-5-57.)

•Matt Breida led the team with 16 carries, Alfred Morris got six and Raheem Mostert only four.

•Todd Gurley went 25 for 114 and 7-6-81-1, leading the team in both rushing and receiving. He also left an easy TD on the table when he slid down short of the goal to kill the rest of the clock in the closing seconds. It was the right football play (the only risk being a fumbled snap as the Rams knelt on the ball), but had he scored, the Rams would have been up eight and the extra point would have sealed it at nine. I suppose missing the extra point is more likely than a fumbled snap, but I could have used the extra TD.

• Gurley only had the chance to run out the clock because RB Ty Montgomery foolishly returned the prior kickoff from the end zone and fumbled it, costly Aaron Rodgers a chance to drive for the game-winning FG. While Aaron Jones was leading a three-man RB committee, I suspect it's now down to two. Speaking of which Jones led the team with 86 yards and a TD on 12 carries, while Jamaal Williams had four carries for nine yards, and Montgomery two for six, so it might be Jones' job anyway. But it's Mike McCarthy, so I say that tentatively.

•Jared Goff didn't seem to do much, but still wound up with 295 yards, three TDs and no picks. He did take five sacks, though. Brandin Cooks (8-3-74) and Robert Woods (7-5-70) split the non-Gurley production evenly, but Josh Reynolds, filling in for Cooper Kupp (5-3-42-2) got the other TDs.

• Aaron Rogers put up solid numbers – 286 yards, one TD, no picks, 9.5 YPA with three sacks – despite missing his final chance. Davante Adams led the team with 7-5-133, while Marquez Valdes-Scantling caught the TD.

• The Raiders were hanging with the Colts until a late fumble by Doug Martin essentially ended their chances. Derek Carr had his best real-life game of the year (244 yards, 8.7 YPA, three TDs, no sacks), but he didn't target his ostensible top receivers Jordy Nelson (4-1-14) or Martavis Bryant (seven snaps, one target that was nullified due to a penalty) much at all. It was Jared Cook (5-4-74-1) and Jalen Richard (8-8-50) for the most part.

• Martin ran well (13-for-72) before killing the team with his fumble, but no one else saw more than two carries.

•Adam Vinatieri returned from a groin injury to set the all-time NFL record for points scored. It's odd they made such a big deal about Brees' passing yardage record, but little about this. I get that he's a kicker, but scoring points is fairly fundamental.

•Andrew Luck had a good game – 239 yards, 7.7 YPA, three TDs, no sacks and no picks. His top target was Jack Doyle (7-6-70-1) in his first game back from a hip injury. Eric Ebron (3-3-37-1) caught another TD, but he's back to second fiddle. T.Y. Hilton's 5-1-34 line was disappointing, as Dontrelle Inman (7-6-52) led the wideouts.

•Marlon Mack had his second straight monster game – 25-132-2, with two catches for 21 yards. He's a already a top-10-ish back in this offense.

• I've called for Eli Manning to be replaced, but what will it take for the Ravens to move on from Joe Flacco? Flacco's not in Eli territory yet, but he put up 4.9 YPA, took two sacks and threw two picks Sunday. And the Ravens have a first-round pick from this year, Lamar Jackson, who played well in garbage time, on the roster.

• No Ravens receiver did much, but Willie Snead led the team with an 11-5-54 line. Alex Collins went 11-49-1 and caught two passes, but fumbled again. Javorius Allen caught three passes, but didn't receive a single carry.

•Cam Newton had his usual – 219 yards passing, two TDs, 7.6 YPA with 10 carries for 52 yards and another score on the ground. He's almost always a top-five-ish fantasy QB by season's end.

•Christian McCaffrey had only one TD all year heading into the game, but scored twice against the Ravens, once on the ground, and once through the air. He had only 56 yards from scrimmage though on 14 carries and four catches.

• First-round rookie D.J. Moore broke out in Week 8 with a 6-5-90 line and two carries for 39 yards. He should be more involved in the offense going forward, but this is a tough team to read with so many versatile parts. Greg Olsen went 4-4-56-1 too, adding another valuable option to the offense, while Devin Funchess caught only three passes.

• James Connor had another massive game against the Browns – 24-146-2 and 6-5-66. It's hard to imagine the Steelers benching him entirely even if Le'Veon Bell were to return in the next couple weeks. Honestly, I think the only thing stopping the Steelers from cutting Bell loose at this point is the possibility he'd sign with the Patriots.

•Antonio Brown went 8-6-74-2, while Juju Smith-Schuster went only 6-4-33. The Steelers didn't need a ton of passing production this game.

•Baker Mayfield didn't do much – only 5.0 YPA, 180 yards, two TDs, one pick and two sacks. Jarvis Landry had a comically inefficient 12-8-39 line while Antonio Callaway and Seth DeValve scored TDs. David Njoku didn't receive a single official target, though he drew a pass interference penalty.

•Nick Chubb went a modest 18 for 65 with two catches for 10 yards. Duke Johnson also did next to nothing. Even with Mayfield, Cleveland is again shaping up to be one of the league's worst offensive environments. Maybe that will change with Hue Jackson gone, though.

•Eli Manning led all passers Sunday with 316 yards, but it was on 47 attempts for only 6.7 YPA. And that doesn't include the seven sacks, two picks and the fact that most of his production was once again in garbage time. At least two of the sacks were due to Manning holding the ball for much too long, and his overall numbers would have been even worse but for Odell Beckham making ridiculous catches on bad (and dangerous) passes. I've never seen a quarterback so inept under even the slightest pressure and yet so inaccurate even when the pocket is clean. I summed up my Manning thoughts in a Twitter post:

I don't know when the zombie apocalypse will arrive, but I do know Eli will be the first to be caught

— Christopher Liss (@Chris_Liss) October 28, 2018

•Saquon Barkley had no room to run (13-for-38) but 10-9-73 pays the bills in PPR. Beckham led the team with 11-8-136, while Sterling Shepard, Bennie Fowler and Evan Engram (who dropped one of Eli Manning's rare good throws on fourth down) split the rest.

•Alex Smith turned in another Alex Smith game – 5.6 YPA, one TD, no picks. He's unstartable in fantasy.

•Adrian Peterson had a monster day, finishing off with a game-sealing 64-yard TD run. He fumbled, but went 26-149-1 and caught another seven-yard TD.

•Jordan Reed got 12 targets, but pulled a Jarvis Landry (12-7-38), while Josh Doctson went 5-5-49.

• Pat Mahomes continued his all-time QB season with 303 yards, four more TDs, 8.9 YPA, one pick and two sacks. Sammy Watkins led the team with a 9-8-107-2 line, Tyreek Hill went 4-3-70 before leaving with a groin injury, and Travis Kelce 10-6-79-1.

•Kareem Hunt had a modest day – 16 carries for 50 yards, but he went 6-5-36-1 as a receiver, and broke multiple tackles and hurdled a defender on his 23-yard TD catch.

•Case Keenum played decently – 262 yards, two TDs and one pick, but he took five sacks for 40 yards. Courtland Sutton, who could see a bigger role should Demaryius Thomas get traded this week, led the team with 4-3-78, while Emmanuel Sanders went 4-4-76 and Thomas 7-3-30.

•Phillip Lindsay (18-95-1) ran well as did Devontae Booker (9-78) – and both had roles in the passing game (Booker, 4-4-23; Lindsay, 3-3-17.)

•Russell Wilson attempted only 17 passes all game, but got 248 yards and three TDs (14.6 YPA.) He also had two rushes for 15 yards.

•Doug Baldwin had a disappointing 3-2-26 line, while David Moore (4-4-97-1) scored again. Ed Dickson, back from an injury, went 2-2-54.

•Chris Carson led the way on the ground – 23-105-1 with two catches for 19 yards, while Mike Davis had 10 carries and Rashaad Penny (the team's first-round pick who played well two weeks ago) had none.

•Marvin Jones (10-7-117-2), Kerryon Johnson (8-6-69) and Golden Tate (12-7-50) had 30 targets between them, but Kenny Golladay went just 1-1-12 for some reason. The Lions got next to nothing on the ground (Johnson 8-for-22, LeGarrette Blount 3-for-3.)

• It was a tough beat if you had Bengals -4.5. Ryan Fitzpatrick (194 yards, two TDs, 12.9 YPA, 18 rushing yards, no picks and one sack) relieved Jameis Winston (276 yards, 7.9 YPA, one TD, four picks and five sacks) and secured the cover, though fell short of the win. It's unclear who will start next week, but Fitzpatrick made a strong case.

• Banged up all week, Peyton Barber went 19-85-1 while Ronald Jones had two carries for minus three yards.

•Mike Evans went 13-6-179-1, and DeSean Jackson (who might be traded this week) caught a 60-yard TD. O.J Howard (4-4-61-1) is still the top tight end, and it's hard to know what's more mystifying, the Bucs drafting Ronald Jones in the second round or signing Cameron Brate (2-1-13) to a six-year deal with $18 million in guaranteed money.

•Joe Mixon had a big game – 21-123-2 and three catches for 15 yards – against a weak Bucs defense. It's a two-man show in the passing game with Tyler Boyd (10-9-138-1) and A.J. Green (7-5-76-1.) I can't remember seeing Boyd on any preseason sleeper list, even though he's a former second-round pick in Year 3.

• Mitch Trubisky keeps putting up big fantasy numbers, but is less inspiring in real-life. He's an excellent scrambler though – six carries for 51 yards – and that and a 70-yard TD pass to Tarik Cohen paid the bills.

•Jordan Howard went 22-81-1, but dropped his only target. No one else in the game did anything noteworthy on either team.

•Carson Wentz threw for 295 yards and three TDs against the Jaguars in London. He's match-up proof, though he threw a pick, lost a fumble and took four sacks. He also added 28 yards on the ground.

•Josh Adams led the Eagles with nine carries for 61 yards, though he nearly lost a crucial fumble, but the review showed he was down. He should continue to push for a bigger role. Wendell Smallwood had eight carries for 24 yards, but caught a 36-yard TD. Corey Clement had only four carries for six yards.

•Jordan Matthews led the team with a 5-4-93 line, as the Jaguars are more vulnerable to the slot. Zach Ertz (6-4-26-1) and Alshon Jeffery (5-4-35) were held in check.

•Blake Bortles didn't play badly – 286 yards, 7.0 YPA, one TD, zero picks and four sacks. He also led the team with 43 rushing yards and should retain his job after the bye.

• The receivers didn't do much – T.J. Yeldon led the team with a 9-7-83 line, while Donte Moncrief was again the most productive wideout (7-4-54) and Dede Westbrook (6-2-31-1) scored the lone TD.

• I know the Eagles have a stout front seven, but it's odd how little Carlos Hyde (six carries for 11 yards, two targets, no catches) was used. The Jaguars head into a Week 9 bye, and Leonard Fournette is allegedly back for Week 10. With T.J. Yeldon also around, it's hard to see how Hyde fits unless Fournette aggravates his hamstring injury.

•Tom Brady played a clean game – 324 yards, 7.2 YPA, no TDs, no picks and two sacks. He did what he had to, but the Bills are a terrible fantasy matchup because they don't score and they have a decent defense.

• The Patriots used Cordarrelle Patterson (10 carries) over Kenjon Barner (2). James White got his usual eight carries, a rushing TD and went 13-10-79 in the passing game.

•Julian Edelman was Tom Brady's other favorite target, going 10-9-104. For someone who was allegedly being punished for tardiness, Josh Gordon seemed to be on the field a lot, though he was a modest 8-4-42 and dropped a would-be downfield throw.

•Rob Gronkowski's run of meager production continued – he went 8-3-43 and failed to concuss anyone after the play.

• Aside from the pick-six, I thought Derek Anderson played well, throwing the ball accurately, avoiding a few sacks. He finished with 290 yards, 7.4 YPA, zero TDs, one pick, one fumble and three sacks. He left the game in the closing minute after being sacked twice on the final series, but it's unclear whether he was injured.

•LeSean McCoy did too much dancing and not enough straight-ahead running but still caught six of eight targets for 82 yards and had 95 YFS. Chris Ivory ran better and finished with 45 YFS.

•Zay Jones, Andre Holmes, Charles Clay and Benjamin split the rest of the receiving yards, with only Jones 8-6-55 making a modest impact.