Cowboys Led 2021 DVOA Overachievers

Dallas Cowboys CB Trevon Diggs
Dallas Cowboys CB Trevon Diggs
Photo: USA Today Sports Images

NFL Offseason - Every year, we here at Football Outsiders do our best to accurately forecast the upcoming season. We run thousands of simulations taking into account major offseason personnel changes, players returning from injury, potential development from draft picks, continuity on the offensive line, and numerous other variables in an attempt to provide the best predictions out there.

Considering how difficult predicting the future can be, 2021 went quite well. There was a .74 correlation between our mean projected wins and teams' actual win totals, and a .73 correlation between our projected DVOA and teams' actual DVOA. For the most part, teams were more or less as good as we anticipated them being.

For the most part, that is, because there's always an exception or two that falls through the cracks, a team that unexpectedly gels or crumbles. You didn't exactly see too many people predicting the Cincinnati Bengals would come out of the AFC when all was said and done!

Over the next couple weeks, we'll take a look at the top overachieving and underachieving teams of 2021. Today, we're looking at the five teams that outperformed their DVOA projections the most, looking at just what we missed and whether or not last year's success was sustainable, or just a mirage.

A quick methodological note before we begin. Our rankings are based on how many standard deviations each team beat or fell short of its projection, rather than just looking at the raw numbers. When you run thousands and thousands of simulations, it's a very rare team indeed that has an average DVOA over 20.0% or performs better than 12-5. But each NFL season only happens one time, so outlier results not only happen, but are expected—someone keeps rolling sevens all season long, it's just difficult to predict who. We're not here to explain that the best teams were better than their projections because that's how simulations work. We want to focus on the teams that got grouped wrong to begin with: the bad teams that became average or the average teams that became good.

5. Buffalo Bills (+1.22)

Projected DVOA: 10.7%; +0.97 standard deviations
Actual DVOA: 27.6%; +2.19 standard deviations

Our projections had the Bills as the third-best team in the AFC, but in a tier firmly below the Chiefs—contenders, for sure, but not quite the cream of the crop. Instead, they led the league in VOA, only falling to second in DVOA because they played the easiest schedule of the season. The Bills skyrocketed right past "contenders" status and straight into, at the very least, co-favorites in the conference for the foreseeable future.

There was a fair bit of consternation when we released our projections, which had the Bills ranked 11th in offense a year after ranking fifth. The projections had a potential Josh Allen regression baked in to them … and, to be fair, that happened. Allen's DVOA fell from 25.9% in 2020 to 4.9% last season; he was still very good, but Allen's 2020 would be a career year for most passers, and our projections took that into account. What the projections missed, however, was the Bills running game coming to life midseason. Their 18.4% rushing DVOA from Week 10 on was best in the league as they found an offensive line combination they liked and decided to stick with Devin Singletary as their lead back rather than rotating him with Zack Moss. That took the pressure off of Allen and the passing game to do everything, and Buffalo's overall offensive DVOA jumped from 4.7% in the first half of the year to 14.4% in the second more in line with what Bills fans were hoping before when the season began.

But it was defensive overperformance which really led the way here. The Bills were projected with a -6.3% defensive DVOA, fifth-best in the league. Instead, they hit -18.1%, the second-best mark in franchise history. They managed this despite Tre'Davious White's season ending early due to an ACL tear. Levi Wallace, Dane Jackson, and Taron Johnson all stepped up to the challenge, while Micah Hyde and Jordan Poyer remained one of the best safety tandems in the league. Up front, rookies Greg Rousseau and Boogie Basham developed faster than anticipated, and Ed Oliver had a career year as the Bills rose from 14th to seventh in adjusted sack rate. They were just OK against the run, but that pass defense was a thing to behold.

Our way too early DVOA projections have the Bills atop the defensive rankings again next year, alongside a top-10 offense. That's enough to put them as the best team in the conference; nothing about 2021 screams "fluke." Plus, you know, adding Von Miller tends to help. Just ask the Rams.

4. Philadelphia Eagles (+1.25)

Projected DVOA: -10.5%; -0.96 standard deviations
Actual DVOA: 3.8%; +0.30 standard deviations

Our projections had the Eagles in the bottom 10 of all three phases of the game. That was spot-on defensively, but no one saw Philadelphia managing to generate an above-average offense in 2021.

Remember, Doug Pederson had just been fired, and Carson Wentz sent packing to Indianapolis. The Eagles were coming off of a -18.8% offensive DVOA and needed a full reboot offensively, to the point where swallowing Wentz's massive contract in exchange for future draft picks was a defensible idea. Stick Jalen Hurts under center, suffer through a rebuilding year while you see what you've got, and be loaded enough to bring in the guys you actually want in 2022.

Well, Hurts was … fine. A 0.4% DVOA isn't going to lock him in as the undisputed passer of the future, but it was a significant jump from his -17.6% mark in limited rookie action. He got to the point where the Eagles could use an upgrade at quarterback rather than needing an upgrade. Baseline competence at the quarterback position was a welcome change of pace, and that alone would have shot the Eagles past our projections. But the Eagles also jumped from 18th to third in rushing DVOA as Nick Sirianni and Shane Steichen pivoted midseason to focus their offense on a rushing attack which stretched defenses from sideline to sideline, always had the threat of the mobile Hurts breaking off and running, and opened up the deep crossing routes to DeVonta Smith that made up the majority of the Philly passing offense. The Eagles' offensive DVOA was projected at -6.2%; they finished all the way up to 8.2% and grabbed a wild-card spot in the process.

The Eagles and Hurts need to keep improving on offense to avoid getting stuck in seventh-seed no man's land. Hurts' success as almost an option player in November and December doesn't seem likely to be sustainable if he doesn't take a step forward as a passer as well. A receiver or two who could catch passes over the middle would likely help.

3. New England Patriots (+1.42)

Projected DVOA: 4.7%, +0.43 standard deviations
Actual DVOA: 23.3%, +1.84 standard deviations

We could probably just type "Mac Jones" in this entry and leave it at that. The general wisdom was that Jones had the highest floor of the all of the rookie quarterbacks, but I don't think anyone was expecting his ceiling to be as high as it was. Our projections had the Patriots 23rd in offensive DVOA at -5.9%, based in large part on the premise that Jones was a rung below the other rookie starters—we weren't even sure he was going to beat out Cam Newton until a week before the season began, after all.

Jones' 6.1% passing DVOA was 11th out of the 82 rookies who have qualified for our leaderboards since 1983; his 620 DYAR ranks 10th. Yes, he was working in an above-average situation for a rookie passer, but Jones was able to take his college accuracy and decision-making and translate them fairly seamlessly to Josh McDaniels' offense—far from a given for any rookie passer. Jones had an 85.2% catchable pass rate per SIS charting, firmly in the middle of the league; Zach Wilson and Justin Fields held the last two slots, while Trevor Lawrence was about halfway between them and Jones. And it wasn't all dink-and-dunks, either; Jones gradually got better at throwing deeper passes as the season went along. He's not at a level where he was singlehandedly leading New England to glory, but rookies are often lead weights as they adapt to the NFL game. Jones looked ready for prime time from Day 1, and the Patriots' offense benefitted.

It's not all Jones, mind you. New England's defense was projected at -6.7%, but actually finished at -12.8%. A large part of that is the difficulty in valuing all the players the Patriots added this year. The return of all of the 2020 COVID opt-outs plus some major spending in free agency meant that the Patriots had a net approximate value over replacement change of +31, the largest total added since at least 2003. When you're setting records for bringing talent in, it can be hard for projection models to really quantify the effect it will have. It turns out, getting a bunch of good players generally makes a team better than they were before. Who knew!

The Patriots may have some trouble living up to these numbers next season. Defensively, they have lost J.C. Jackson, Kyle Van Noy, and others. Their top outside receiver remains Nelson Agholor, which is somewhat less than ideal for Jones as he continues to grow. Jones is three steps ahead of the rest of the rookie class, but he needs to keep improving if he's going to be the long-term answer at the quarterback position. But suffice to say that the Patriots won't be projected 23rd again entering 2022.

2. Los Angeles Chargers (+1.42)

Projected DVOA: -8.0%; -0.73 standard deviations
Actual DVOA: 8.7%; +0.69 standard deviations

While they don't quite hit the top spot on our list, you could make the argument that the Chargers were our biggest miss. They went from the bottom 12 in our projections to the top 12 in the standings, after all. They still just missed out on the postseason, but with the possible exception of the Colts, no team went from projected afterthought to playoff contender quite as much as the Chargers did.

We projected the Chargers with a 3.8% defensive DVOA; they actually underperformed that at 4.8%. We projected them with a -2.8% special teams DVOA; they barely cleared that at -2.6%. We projected them with a -1.4% offensive DVOA and they jumped to 15.9%, fourth-best in the league. In Football Outsiders Almanac 2021, we wrote that "it is optimistic to assume Herbert can lead a top-three offense right now in order to make up for the defense." I mean, technically, we nailed it; fourth place is in fact not top-three.

Quarterbacks often make significant jumps between Year 1 and Year 2, but Justin Herbert was so good as a rookie that it was easy to think that he had already arrived, fully formed. How much could you really expect him to jump from his 10.2% DVOA as a rookie, eighth best among first-year quarterbacks in our database? Well, it turns out you could expect him to jump to 17.4% with a DYAR in the top five in the league, combining his never-questioned arm talent with a level of accuracy and decision-making that is frankly shocking, considering how he played at Oregon. The Chargers also saw a massive improvement in their offensive line, going from disastrous to passable, and they didn't end up missing Hunter Henry at all. Still, outperforming the projection was 99% Herbert's continued ascension into an upper-tier passer.

And the Chargers may well be better next season, with Brandon Staley having a year to both install his new defense and a busy free agency period to load up on players who can actually play. It's just a shame for them that they're playing in the AFC West. It's almost unthinkable that a player on Herbert's trajectory could go three straight years without making the postseason, right?

1. Dallas Cowboys (+1.94)

Projected DVOA: 5.6%; +0.51 standard deviations
Actual DVOA: 30.9%; +2.45 standard deviations

Our top overperforming team started the year projected to just squeak into the top 10 in DVOA. They finished the season tops overall for the first time since the dynasty years of the 1990s. It would be hard to imagine this season having gone much better for America's Team, assuming you stopped paying attention minutes after Week 18 ended.

The Cowboys outperformed projections on both sides of the ball. Offensively, it's a story of the Cowboys starting out incredibly strongly through six weeks, with a 24.9% offensive DVOA. Dak Prescott then injured his calf and the Cowboys fell to 6.4% the rest of the way as his limited mobility hampered what Dallas was trying to do. That 6.4% mark basically matches their preseason projection of 6.7%, but playing a third of the season at a high level is enough to keep Dallas' overall offensive DVOA at 13.4%, nearly seven points higher than projected. It was just an unexpected twist that the Prescott injury limiting the Cowboys' offense was a new one in 2021 as opposed to the severe ankle dislocation from 2020.

Defensively, however, Dan Quinn came in and turned things around. We had the Cowboys projected firmly in the middle of the pack with a 1.2% defensive DVOA. They finished the year second at -15.2%. Some of that is Quinn producing a much more comprehensible scheme than Mike Nolan, doing a much better job of putting guys in position to succeed and installing a system similar enough to what the Cowboys ran pre-Nolan to let the veterans stay comfortable. Phenom Micah Parsons was the much-deserved defensive rookie of the year, regardless of questions about his actual position. He had to play more snaps than ideal at linebacker because the defensive line of DeMarcus Lawrence, Randy Gregory, Neville Gallimore, and Osa Odihizuwa was an absolute monster of a unit. And then you had the turnovers—we can argue all we want about Trevon Diggs' underlying coverage abilities, but it turns out intercepting 11 passes is generally speaking a Good Thing, as is the defense as a whole leading the league with 34 takeaways. That ball-hawking, harassing defense turned Dallas from a good team to one with thoughts of Super Bowl glory. Maybe next year?

The Cowboys entered the offseason first in our way too early projections, but it hasn't exactly been a spectacular offseason to date—it feels like half the team is in "win now!" mode while the other half is dumping players like Amari Cooper to try to save cap space. I would expect the defense to come somewhat back down to Earth as takeaways regress towards the mean. But they kept Quinn, which is huge, and the odds of Prescott getting hurt three years in a row has to be low. Dallas should be right in the mix once again next season.

Comments

29 comments, Last at 31 Mar 2022, 11:06am

1 The projection for LAC in FO…

The projection for LAC in FO Almanac 2021 was obviously way too low (they were projected to barely higher than DET) and BUF was due to the projections being cautions, but the others all definitely overperformed our expectations.

2 When you run thousands and…

When you run thousands and thousands of simulations, it's a very rare team indeed that has an average DVOA over 20.0% or performs better than 12-5. But each NFL season only happens one time, so outlier results not only happen, but are expected—someone keeps rolling sevens all season long, it's just difficult to predict who.

Purely methodological question:

I assume you're using some form of Monte Carlo here, but how are you handling the target distribution?

Someone keeps rolling sevens all season long

This is suspiciously similar to the modeler's trap, where modelers have a tendency of assuming that when their model does not match reality that the error is with reality. How do you sort whether a team was lucky vs when your model had a flawed assumption of their baseline strength?

7 Well, luck is another name…

Well, luck is another name for "factors outside of our knowledge/view" since very few things are truly random (especially at this level) more accurately projected talent would explain alot of these

14 Isn't that basically what…

Isn't that basically what this article is about? These 5 teams all performed significantly better than the even their lucky modeled versions did, and thus it's likely that there were effects not present in the model.

17 The line about rolling 7s…

The line about rolling 7s and whether or not success is sustainable is asserting a model where results are driven by luck and we just live in the reality where the Bills won a lot.

It goes back to the question about gambling odds and when to start suspecting a die isn't fair.

18 where results are driven by…

where results are driven by luck

"Luck" doesn't have to be week-to-week, it can also be season-long. Betting on Josh Gordon to make it through a season, for instance. A lot of the projections are also regression types, as in "first-time head coaches tend to be bad," so lacking information you go with your best guess. Then the season comes and he's aggressive and innovative, and so of course your projection is off and it wasn't luck per se (it would've happened the same with the same person) but you had no way to predict it since you had no other information.

Systematic effects vs. week-to-week statistical.

19 How do you sort whether a…

How do you sort whether a team was lucky vs when your model had a flawed assumption of their baseline strength?

This is a really good question. My understanding is that there is no methodological answer, and it's what Bryan is trying to parse in the commentary.

There are some elements which FO are pretty sure are unduly influenced by randomness. For example sacks, which are more highly subject to randomness than pressures. I don't think that there's a methodology applied to this analysis, so it's looking and guessing.

Of the elements discussed (and there are lots) a couple stand out to me as possible sources of randomness - Diggs' interception count and the Bills sack rate (even though it is adjusted). I would therefore parse that all other improvements are real things that the model didn't capture.

Which makes sense, and I think what the commentary is saying. The model missed the second year improvements for Hurts and Herbert, and made an assumption about Jones as a rookie that was reasonable, but wrong. The model also didn't predict the huge leap forwards for the Eagles running game, or the defensive improvement for the Patriots - both somewhat based on changes in personnel.

Big picture, the model is fine, but sometimes players are better than expected as rookies, or make leaps in the second year. Those improvements are sustainable, so good news for the Eagles, Patriots and Chargers, but you wouldn't change the model because most rookie QBs are not as good as Jones, and most QBs don't see that second year leap.

Apart from putting a higher range into the simulation for teams that have a big personnel turnover, I'm not sure what else you would do. Those improvements are not luck based, but it's very hard to determine if a lot of changes in the player has are inherently good or bad. Probably neither (on balance) so the only thing to do is increase the range of outcomes where you can identify those cases.

3 It's almost unthinkable that…

It's almost unthinkable that a player on Herbert's trajectory could go three straight years without making the postseason, right?

It happened to Hadl, Fouts, Brees, and Rivers, so...

21 This a weird comment. Rivers…

This a weird comment. Rivers was in the postseason the first 4 years he was a starter. Brees did not make the playoffs his first three years but he also wasn't very good until his 4th season, so in no way was he on the same trajectory as Herbert. I don't know anything about the old-timers though.

4 Top underperformer is…

Top underperformer is probably BAL, although that can be explained by not having accurate roster information at projection time

5 but no one saw Philadelphia…

but no one saw Philadelphia managing to generate an above-average offense in 2021.

The collapse in 2020 was due to them going into the year thin at OL to start with, and then getting blasted by injuries. Which completely imploded Carson Wentz. Remember the interception he threw with the Colts where he was backed up, practically sacked, and just tossed the ball up because... god knows why? That's Wentz in a nutshell - he's mediocre in structure, and out of structure if the rest of the team's good, he's great, but if the rest of the team is bad... he's a total disaster.

2021 had the OL rebound dramatically, with Mailata, Kelce, and Johnson playing practically the entire year. It wasn't likely, but it was pretty easy to foresee.

The key to Philly improving in the future isn't about them getting more receiving threats (it's not like they won the Super Bowl with receiving superstars or anything), although that'd help for this year. It's about them replacing the key guys on the OL (Kelce and Johnson) and rebuilding the defense as well.

6 Rankings

I do not understand your rankings.  The Bills had the second largest increase, yet they are ranked fifth.

9 It's explained in the last paragraph of the article

In reply to by Raiderfan

Our rankings are based on how many standard deviations each team beat or fell short of its projection, rather than just looking at the raw numbers. 

So I think it's supposed to be like this, except that New England doesn't look as if it's in the right slot:

Buffalo Bills:

Projected: +0.97 standard deviations
Actual: +2.19 standard deviations
Difference: +1.22 standard deviations

Philadelphia Eagles:

Projected: -0.96 standard deviations
Actual: +0.30 standard deviations
Difference: +1.26 standard deviations

New England Patriots:

Projected: +0.43 standard deviations
Actual: +1.42 standard deviations
Difference: +0.99 standard deviations

Los Angeles Chargers:

Projected: -0.73 standard deviations
Actual: +0.69 standard deviations
Difference: +1.42 standard deviations

Dallas Cowboys:

Projected: +0.51 standard deviations
Actual: +2.45 standard deviations
Difference: +1.94 standard deviations

But I think that's how FO is ranking the teams' performances.

10 I thought it might be by…

I thought it might be by DVOA difference.  But that doesen't work either. Wrong in different ways however. 

5. Buffalo Bills

Projected DVOA: 10.7%; +0.97 standard deviations
Actual DVOA: 17.0%; +2.19 standard deviations

Difference: 6.3% DVOA, +1.22 standard deviations

4. Philadelphia Eagles

Projected DVOA: -10.5%; -0.96 standard deviations
Actual DVOA: 3.8%; +0.30 standard deviations

Difference: 14.3% DVOA,+1.26 standard deviations
 

3. New England Patriots

Projected DVOA: 4.7%, +0.43 standard deviations
Actual DVOA: 23.3%, +1.42 standard deviations

Difference: 18.6% DVOA,+0.99 standard deviations

2. Los Angeles Chargers

Projected DVOA: -8.0%; -0.73 standard deviations
Actual DVOA: 8.7%; +0.69 standard deviations

Difference: 16.7% DVOA,+1.42 standard deviations

1. Dallas Cowboys

Projected DVOA: 5.6%; +0.51 standard deviations
Actual DVOA: 30.9%; +2.45 standard deviations

Difference: 25.3% DVOA,  +1.94 standard deviations

 

11 The Patriots were the…

The Patriots were the problem, as I put the wrong numbers there.  They went from +0.43 standard deviations to +1.84, not +1.42 -- +1.42 is how much they improved by, just a few thousandths below the Chargers.

That's been corrected, which should hopefully make that make more sense.

8 I think it's just "5 team…

I think it's just "5 team that beat projections the most" and they arnt really ordered by anything except maybe some sort of "surprise factor"

13 Something to consider. Buff…

Something to consider. Buff was 32nd in variance despite leading the league in DVOA and the Pats were in the high 20s in variance as well.

That suggests to me that while Buffalos mean defense was great, its not something you could count on every week. It may also be an artifact of a weak schedule which is warping the numbers. Maybe DVOA overstates blowouts over bad teams and understates poor performances against good teams.

Either way, perhaps a sharpe ratio like stat would make sense for defenses.

15 Buffalo's defense was good,…

Buffalo's defense was good, but not tremendous to the eye last year. They got up big on the weak teams on the schedule and then the pass D/pass rush was tremendous when teams got one-dimensional. The run defense was awful until Harrison became a starter, which is why they overhauled the D-Line in Free Agency.

16 Prescott injury

I don't think that the calf injury limited Prescott in a physical way. I watched breakdowns of the Cowboys offense after the injury and his throwing motion didn't seem hampered at all. He still escaped the pocket when need as well.

He also had a left shoulder injury, so it seems like a mental thing that made him play scared, and so he became less accurate and had processing issues. Didn't help that the run blocking suddenly became awful and the team was intent on playing an injured Zeke.

25 While I haven't been able to…

In reply to by Romodini

While I haven't been able to watch the tape super closely, to me the biggest change in the passing game was the absence of deeper strikes. Prescott from 2019 through the first half of 2021 was very strong with the deep ball, and the passing attack was generally about taking what defenses give it underneath until the right opportunity was available deep.

Prescott in the second half didn't even attempt all that many deep throws, at least relative to earlier in the year. And his ball placement was much worse than prior. Either he had a mechanical issue (which could originate from a physical issue or a mental hangup) that then forced the offense to change its decision-making, or he lost comfort with reading deeper. I figure that the former is much more likely - he knew he couldn't trust himself to put the ball in the right place on a number of throws he would have attempted just prior, and thus he focused on throws he could trust. That narrowed what opposing defenses had to defend and also removed some of the most efficient throws from the offense.

20 I applaud you for a…

I applaud you for a statistical metric for outperforming/underperforming expectations, but I have a couple suggestions for presenting the data in future years:

  1.  I would like to see the histogram of simulated DVOA, with a line or bar or point indicating where the team ended up in reality. That way I can see whether the actual season was an outlier because of something about the shape of the distribution of simulations, and I can compare distributions between teams and ask all sorts of annoying questions about why they look different.
  2. Maybe this is just me, but I'd prefer seeing the projected DVOA written as X +/- Y (where Y is the standard deviation) and the real DVOA written as just a single number. I don't find the # of SD above zero to be a particularly useful number to report.

And, of course, feel free to ignore these suggestions, as my co-authors in real life often do.

27 +1 to both of these…

+1 to both of these suggestions, in particular the histogram of projected DVOA, would be cool to see teams anticipated for higher/lower distribution of outcomes

28 +1 to both of these…

+1 to both of these suggestions, in particular the histogram of projected DVOA, would be cool to see teams anticipated for higher/lower distribution of outcomes

23 “and the odds of Prescott…

“and the odds of Prescott getting hurt three years in a row has to be low.”

do the odds have to be low, though? At what point does it become an issue with the player more than a luck thing? Because while luck obviously plays a huge role, there is clearly a difference between someone like Brett Favre and Chad Pennington in terms of health. Jason Verrett is a great example of a “cursed” guy who just can’t stay on the field through no real fault of his own.

24 I mean

He started every game his first 4 years (67 games). 

<34% of missing full games seems about right. 

26 Dallas's defense and regression

While it's a safe bet that Dallas will experience negative regression on defense from a corresponding likely regression of forced turnovers, it's worth pointing out that the defense experienced a lot of injuries early on, the unit was adjusting to its second straight new DC, and the group was in the youngest (top fifth of the past few seasons) tier for snap-weighted age. Comfort in scheme, added experience, and more even injury effects should help buoy the defense from dropping too much simply from turnover regression. I say "should", as of course these things often don't play out as we would intuitively expect!

29 Odds of reinjury

the odds of Prescott getting hurt three years in a row has to be low. 

This is untrue.  I believe FO presented data a few years ago showing that the odds of injury or reinjury increase with the number of prior injuries a player incurs. So perhaps the odds are lower than 50%, but Prescott’s chances of injury are higher than someone who has had no or fewer prior injuries.