Week 7 DVOA Ratings
Tampa Bay stays on top of the Football Outsiders DVOA ratings after Week 7. In fact, Tampa's 45-20 win over Las Vegas came out as their second-best game of the year so far (59.7%) and increased the gap between the Buccaneers and the rest of the NFL.
Last week, I wrote that Tampa Bay ranked 28th in DVOA history through Week 6. Well, the Buccaneers moved up a bit this week, which means it is time to break out this table again:
|BEST DVOA THROUGH 7 GAMES, 1985-2020|
Baltimore and Pittsburgh are next behind Tampa Bay, getting ready for their big matchup coming this Sunday. The Ravens move up a little bit on their week off thanks to changes in opponent adjustments, while Pittsburgh moves down a little bit from a close game with Tennessee. Kansas City moves up to No. 4 after beating Denver, primarily because of a boost in special teams from Byron Pringle's kick return touchdown. The Chiefs defense didn't actually improve in DVOA because that kind of great defensive performance is what we expect teams to have against the Broncos this season. After Kansas City, we've got Indianapolis as surprising fifth and then New Orleans sixth.
Then comes the best division in football, the NFC West. The Los Angeles Rams are up to seventh this week, the San Francisco 49ers are up to eighth, and the Seattle Seahawks drop to ninth. Arizona is now 11th, with Green Bay in between. That's all four NFC West teams in the top 11.
There's a lot of interest in whether the entire NFC West can make the playoffs. The biggest barrier is going to be the New Orleans Saints, and either the Bucs or the Saints end up with a wild card in roughly two-thirds of our simulations. The entire NFC West makes the playoffs in only 4.8% of our simulations. Other possibilities:
- Entire NFC West has a winning record: 38.4%
- Entire NFC East has a losing record: 63.1%
- Entire NFC East is 6-9-1 or worse: 26.4%
- Entire NFC East is 5-10-1 or worse: 3.8%
- Entire NFC West has a winning record and entire NFC East has a losing record: 25.9%
A quick digression on the NFC East: Washington now has the best DVOA in the NFC East at No. 21 after a 25-3 spanking of Dallas. San Francisco and Washington got big ratings for their wins over New England and Dallas, respectively, but their positive ratings weren't as strong as the negative ratings for the losing teams. New England's loss to San Francisco is now rated as the worst single game of the year (-106.9%) and Dallas' loss to Washington is rated as the second-worst game (-102.8%), with both games surpassing Cleveland's huge losses to Baltimore and Pittsburgh.
Returning to the NFC West: Not only is the entire division in the top 11 overall, but the entire division is in the top 11 for offense specifically. And three of the four teams are in the top 11 on defense as well. Seattle is the clear exception, second in offense and 28th in defense.
Just how imbalanced are the Seahawks? Normally, we get a team's total DVOA by subtracting defense from offense. Because defense is better when it is negative, that gives us the best teams overall. But instead, let's look at each team's balance between offense and defense by adding offense and defense. This will give us a "game offense" rating where the highest teams will be playing games with a lot of offense (they are good at offense, bad at defense). So Seattle has a "game offense" rating of 29.9%, Las Vegas of 27.6%, and so on. Here are the top five:
|Top 5 "Game Offense" DVOA, 2020|
Now, here's the thing: Seattle's game offense rating of 29.9% isn't very high at all. In the new version of DVOA we debuted this year, going back to 1999, there are 27 teams that ended the year with a "game offense" rating above 30%. That's more than one team per year. This was more common at the beginning of the decade, but in recent years both the 2017 Patriots and the 2018 Chiefs were more imbalanced in favor of offense than this year's Seahawks. And being this imbalanced doesn't stop you from winning games as long as your offense is strong enough. Twenty of these 27 teams made the playoffs, and five of them made it to the Super Bowl, although only the 2006 Colts won a championship. Notice that four of the five teams on this table for 2020 have winning records.
What about the flip of this, teams with a very strong defense but not a very strong offense? It turns out we have those teams this year as well, and with a stronger effect than the all-offense teams. Washington is currently 29th on offense and fifth on defense, adding up to a "game offense" rating of -40.7%. Denver is also pretty extreme. Here are the bottom five teams in game offense:
|Bottom 5 "Game Offense" DVOA, 2020|
Unlike Seattle and other offense-first teams, Washington does turn out to be historically imbalanced right now. Only seven teams since 1999 have a game offense rating lower than this year's Washington Football Team. Three of those teams are recent, including last year's Pittsburgh Steelers and both Arizona and Buffalo from 2018.
We saw most of the offense-first teams made the playoffs, so what about these defense-first teams? Turns out they are mostly awful. Last year's Steelers (8-8) are the only one of the teams with game offense rating of -40% or lower to go better than 6-10. It's an interesting mix: most of these defense-first teams were bad but the few that were really good overall tended to win the Super Bowl. There are 34 teams from 1999 to 2019 with a game offense index of -30% or lower. Only eight of those 34 teams made it to the postseason. Yet half of those teams that made the playoffs went on to win the Super Bowl: 2000 Ravens, 2002 Buccaneers, 2008 Steelers, and 2015 Broncos. All four of those teams were either No. 1 in defense or close to it. The moral of the story may be that if you put together a team built mostly around defense, that defense better be really, really good.
As for the most balanced team in the league so far, that would be the Dallas Cowboys, with the closest game offense to zero at 1.2%. The Cowboys are terrible on both offense and defense! In honor of the Cowboys, let's finish up today with two remarkable historical comparisons for the two great historical franchises that had horrible games in Week 7.
The Dallas Cowboys are now at -26.0% DVOA after seven games. The last time they had a DVOA this bad after more than six games was 1990. The last time they had a DVOA this bad after even just two games was 2001, when they were at -32.1% after Week 5.
The New England Patriots are now at -20.8% DVOA after six games. The last time they had a DVOA this bad after more than six games was 1995. The last time they had a DVOA this bad after even just two games was 2008, when they were at -21.8% after Week 6.
* * * * *
Football Outsiders playoff odds, snap counts, and the FO+ database are now all updated through Week 7.
A couple more housekeeping notes this week. First, we've introduced a new, often requested view in the FO+ database. The "DVOA as of a Specific Week" view now can be filtered by either offense or defense to see pass/run splits from the past! These new pass/run splits show what DVOA would have looked like at a specific point in history with the current version of DVOA (so scrambles are pass plays) and opponent adjustments as they would have looked at the time. Right now, these tables go back to 2005, but eventually they will be filled all the way back to 1985.
Here's an example: 2007 offense as of Week 8, when the undefeated Patriots were at their apex before declining in the second half of the year.
We've also made a change this week in how we figure the playoff odds report. We've always known that our results were a bit conservative, but frankly they seemed a bit too conservative. Analyzing the last ten years, it looks like we can get more accurate results by rescaling DAVE/DVOA each week so that the standard deviation matches the standard deviation of the full-season ratings. This adjustment gets smaller with each week because as the preseason projection becomes a smaller part of DAVE, the scale of the ratings gets closer to the scale of full-season ratings.
The difference isn't huge, but it does make a difference. The gap between the best teams and the worst teams is now larger. The top teams end up with about 0.5 more mean wins over the course of the simulations. The New York Jets end up with fewer wins and a higher chance of earning the No. 1 pick by about 6%. However, the Jets only end up 0-16 in 6.7% of simulations. It's still much more likely that they will somehow stumble their way into winning a game or two.
* * * * *
These are the Football Outsiders team efficiency ratings through seven weeks of 2020, measured by our proprietary Defense-adjusted Value Over Average (DVOA) system that breaks down every single play and compares a team's performance to the league average based on situation in order to determine value over average. (Explained further here.)
OFFENSE and DEFENSE DVOA are adjusted to consider all fumbles, kept or lost, as equal value. SPECIAL TEAMS DVOA is adjusted for type of stadium (warm, cold, dome, Denver) and week of season. As always, positive numbers represent more points so DEFENSE is better when it is NEGATIVE.
Because it is early in the season, opponent adjustments are only at 70% strength; they will increase 10% every week through Week 10.
DAVE is a formula which combines our preseason forecast with current DVOA to get a more accurate projection of how a team will play the rest of the season. DAVE is currently 50% preseason forecast for teams with seven games played and 55% preseason forecast for teams with six games played.
To save people some time, please use the following format for all complaints:
<team> is clearly ranked <too high/too low> because <reason unrelated to DVOA>. <subjective ranking system> is way better than this. <unrelated team-supporting or -denigrating comment, preferably with poor spelling and/or chat-acceptable spelling>
- NON-ADJUSTED TOTAL DVOA does not include the adjustments for opponent strength or the adjustments for weather and altitude in special teams, and only penalizes offenses for lost fumbles rather than all fumbles.
- ESTIMATED WINS uses a statistic known as "Forest Index" that emphasizes consistency as well as DVOA in the most important specific situations: red zone defense, first quarter offense, and performance in the second half when the score is close. It then projects a number of wins adjusted to a league-average schedule and a league-average rate of recovering fumbles. Teams that have had their bye week are projected as if they had played one game per week.
- PAST SCHEDULE lists average DVOA of opponents played this season, ranked from hardest schedule (#1, most positive) to easiest schedule (#32, most negative). It is not adjusted for which games are home or road.
- FUTURE SCHEDULE lists average DVOA of opponents still left to play this season, ranked from hardest schedule (#1, most positive) to easiest schedule (#32, most negative). It is not adjusted for which games are home or road.
- VARIANCE measures the statistical variance of the team's weekly DVOA performance. Teams are ranked from most consistent (#1, lowest variance) to least consistent (#32, highest variance).
88 comments, Last at 31 Oct 2020, 6:06pm
#1 by CHIP72 // Oct 27, 2020 - 5:34pm
I would love to see a list of the teams that rank the highest all-time in both the "Game Offense" and "Game Defense" ratings. One has to figure teams like the early 2000s Chiefs, the 2000 Rams and Vikings, and late 1980s Dolphins would rank high in Game Offense, and the 1991 Eagles would probably lap the field in Game Defense. (As an Eagles fan, I can tell you that it isn't a good idea to mention the names "Brad Goebel", "Pat Ryan", and "Bryce Paup" around Eagles fans who are old enough to remember the 1991 team.)
On a related note, I hope Football Outsiders is able to eventually calculate DVOA for seasons going back to at least 1981, partly because that was the first season I followed the NFL but mostly because I'd love to see what kind of DVOA offensive ratings and "Game Offense" ratings the Air Coryell Chargers would have, especially in 1981 and 1982 when their offense was particularly potent and their defense was not good. The Dan Fouts-led Chargers were sort of like the Dan Marino Dolphins a few years later, but even further ahead of their time.
I'd also like to see exactly how bad the 1981 Baltimore Colts were in historical terms. From reading about that team (both in the Super Bowl 16 game program*, the first Super Bowl game program I received as a kid, and much more recently), it appears a lot of teams called off the dogs on the Colts long before the end of the game. The Colts did play a lot of good teams that season, but its hard to hold the record for most yards allowed for 30+ years (finally broken by the 2015 Saints) and STILL hold the record for most points allowed in a 16 game season (533 points, or 33.3 points per game), especially considering the fact offenses were not as explosive and with the exception of the Air Coryell Chargers, passed for a lot fewer yards back then than they have over most of the last 20 years.
*Super Bowl programs include a week by week season recap, and I remember seeing all the lopsided scores and increasingly terrible win-loss record the Colts posted from reading those reviews.
#5 by Aaron Brooks G… // Oct 27, 2020 - 6:03pm
the 1991 Eagles would probably lap the field in Game Defense
By comparison, the 1986 Bears are only -40%. Very few great defensive teams seem to exceed -35%.
It's easier to get here via historically-inept offense. 2018 Arizona is at -44%. 1992 Seattle is at -57%!
#7 by theTDC // Oct 27, 2020 - 6:13pm
Game defense, where having a terrible offense and great defense means you win.
It's quite interesting that terrible offenses dominate that list over the great defenses. Why would we expect great defenses to be "ruined" with good/passable offenses more than we would expect truly awful offenses to be "ruined" with also bad defenses?
Maybe it's a lead thing. Great defenses give their offenses the lead, so they call more low DVOA runs to waste clock. Terrible offense would theoretically do the same thing though, for the other team, so that doesn't make sense, either.
Maybe it's the Jeff Fisher brand of idiocy, where you've got some "hard nosed," "old fashioned," coach, who doesn't really understand football but can get the guys to play hard. So as a result the defense can play okay and not be too embarrassing. But the offense, which requires a little more finesse, just sucks.
#19 by CHIP72 // Oct 28, 2020 - 12:06am
I suspect the reason why great defense/bad offense teams tend to be worse than great offense/bad defense teams is due to the number of points scored in a game and each team's margin of error related to that. With a high scoring offensive team that also gives up a lot of points, there's no upper limit to how many points the good offensive team can score; those great offensive teams are out on the right side tail of the bell curve in terms of scoring points. The high number of points also means that individual mistakes by the team's offense or defense can be more easily negated because the offense can come back and score more points.
In contrast to great offense/bad defense teams, great defense/bad offense teams have a much lower margin of error. There are few points scored in their games, and individual mistakes are magnified if a touchdown or even a field goal is enough to swing the outcome of a game if points are at a premium. Additionally, unlike great offensive teams that at least theoretically have no upper bound for the number of points they score, great defensive teams DO have a lower bound in terms of points they give up; they cannot give up fewer than zero points in a game. A team can be merely good or even only decent defensively in a given game, but still shut out a team or hold them to single digits due to their offense holding the ball for much of the game, or good special teams play repeatedly giving the other team poor field position, or because their opponent's offense turned the ball over multiple times. Finally, excellent defensive teams also usually do not earn many points (i.e. score touchdowns or safeties) for their team for their play. The difference in the number of points scored by a great defense vs the number of points scored by a bad defense over the course of a season is much smaller than the difference between the number of points scored by a great offense vs. a bad offense.
#25 by Aaron Brooks G… // Oct 28, 2020 - 9:02am
Interesting. I wonder if a log analysis isn't appropriate.
Looking backward, the 1977 Falcons defense allowed 14 TDs on the year, versus scoring three of their own. They net allowed less than a TD per game. Unfortunately, their own offense made everyone else look like the 1977 Falcons defense.
#45 by theTDC // Oct 28, 2020 - 2:17pm
Maybe, but I think that only explains point differential, not DVOA. If you have one defense with 90% of better than average plays, and 10% of worse than average plays, you would expect that they would have the same DVOA as an offense with similar proportions. The margin of error isn't really any higher with having a great defense, because it's all about cumulative plays and strength over average.
It's possible that there are some factors that make the mistakes on defense more painful though, such as a breakdown in coverage potentially leading to an 80 yard score. On offense a similar mistake might mean the QB sees the receiver runs the wrong route and throws the ball out of bounds. So there may be some systemic factors that overly punish defenses for small mistakes versus offenses, artificially evening the field.
#46 by Aaron Brooks G… // Oct 28, 2020 - 2:54pm
It's easier to rack up a larger count of positive plays for a great offense versus that of a great defense.
A great drive for a great offense is 9 plays of first downs or touchdowns. Those 9 plays outweigh a lot of negative ones.
But a great drive for a defense is a 3-and-out. You accumulated 1/3rd the number of positive plays to use to offset negative ones.
Consider just historical beatdowns on both sides of the ball.
Pittsburgh ran 62 more plays than Cleveland did. Their offense got 62 more plays to generate positive DVOA than their defense got. I think you see a lot of this, just on less stark of a magnitude.
#51 by CHIP72 // Oct 28, 2020 - 5:37pm
RE: Great defensive/bad offensive teams and relatively poor records compared to great offensive/bad defensive teams, the issue identified in the article isn't defensive DVOA, it is win-loss record. Most great defensive teams, regardless of their exact defensive DVOA rating, give up relatively few yards and points compared to average defenses, so we can also focus on their yardage and points allowed. Excellent defensive plays usually result in a small number of yards lost, relative to the number of yards gained on excellent offensive plays. I mean, a 3 yard loss on a run play or a 10 yard loss on a pass play (i.e. a sack) are really good defensive plays, right? But if you turn it around and think about 3 yard runs and 10 yard passes on offense, those are NOT unusually good offensive plays; offensive plays often gain that many yards. The impact of "great" defensive plays is generally less than the impact of "great" offensive plays. That's been especially true over the last 20-25 years as turnovers have declined significantly.
Related to the above, the break even point between a good offensive play and a good defensive play is not zero yards; it is roughly the average yardage gained per play across the league. (More accurately, it is probably the median yards gained per play across the league; the median yards gained per play is always a little less than the mean yards gained per play.) Even a good or great defensive team is going to give up a relatively large number of yards (usually around or a little more than 250 yards) in an average game, and the opposing team's offense will usually have at least 1-2 good scoring opportunities. Additionally, it is usually harder for defenses to increase the yardage lost on a play than it is for offenses to increase the yardage gained on a play. Teams that score a lot of points usually have more big offensive plays; big offensive plays make it easier for teams to score. It is hard for offenses to sustain 18 play, 80 yard drives. But the opposite, the correlation between big defensive plays and points scored by their own team, is not necessarily true with great defenses, or at least the correlation is much smaller. There are very, very few defensive plays that result in 30 yard losses. Great defenses don't score a lot of points for their teams, and their big plays often do not directly make it easier for their teams to score, due to the smaller number of yards lost compared to yards gained on big plays on offense. Great defensive teams that are paired with bad offensive teams have a smaller margin of error than their great offense/bad defense counterparts because of the nature of big offensive plays vs big defensive plays and scoring points for your team as a result of those big plays.
#55 by theTDC // Oct 28, 2020 - 8:00pm
I appreciate both of the responses above, but I think mostly, but not entirely, missed the point. DVOA is not a counting stat. If you have some offense that manages to stay on the field a lot, the DVOA doesn't change as long as the barest minimum number of plays have happened to get a statistically valid number.
I think it is possibly somewhat valid that great defenses and terrible offenses are more random, because they play less snaps and therefore the sample size is smaller. That doesn't explain why the DVOA can be higher for the offenses.
What I agree with is the argument that the maximum "good," play for an offense is worth more, pick sixes aside. Since turnovers are more rare, a great offense can have more impact on a game, and can seperate itself further from other offenses in the league. That part makes sense.
#59 by Aaron Brooks G… // Oct 29, 2020 - 8:35am
A good offense has more opportunity to generate positive plays than it does negative ones, and has more positive plays banked to balance them out. A great defense only has <= 3 play drives, so it has many fewer banked plays with which to cancel the occasional bust.
Think of it as a weighted mean, where great offense have a much larger potential weighting function.
#9 by Vincent Verhei // Oct 27, 2020 - 6:26pm
Mike Tanier covered those 1981 Colts for us way back in 2007 (my first year with the site ... man I'm old). Like everything Mike writes, it's well worth your time to read.
#18 by CHIP72 // Oct 27, 2020 - 11:40pm
I've read that article previously.
The Colts' defensive stats that season, especially against the pass, are so much worse than everyone else that it is mind bending. In 1981, the Colts gave up 37 touchdown passes. The record number of touchdown passes for any quarterback in a season at that point in time was 36 touchdowns (IIRC Y.A. Tittle held the record; Dan Marino shattered it in 1984 with 48 touchdown passes). Baltimore also gave up more than 1 yard more per pass attempt than any other team in the league that season.
#20 by Richie // Oct 28, 2020 - 1:40am
"Baltimore also gave up more than 1 yard more per pass attempt than any other team in the league that season."
I think you might be looking at NY/A or AY/A. The Colts surrendered 8.6 Y/A in 1981 while the Chargers surrendered 8.2. The Saints and Lions followed with both at 7.6 Y/A.
#40 by CHIP72 // Oct 28, 2020 - 12:44pm
I wasn't clear enough in my comment above - the 1981 Colts gave up more than 1 more net yard per pass attempt (8.2) than any other team in the league. (The 1981 Chargers were the second-worst team at 7.0 net yards per pass attempt.)
The difference between the Colts' and Chargers' defenses in yards per pass attempt and net yards per pass attempt was due to a significant disparity in sacks. Though sacks were not an official statistic until the following season, according to Pro Football Reference the Colts only had 13 sacks (the fewest in the NFL by 7 sacks) for 100 net yards lost. By contrast, though the Chargers were poor defensively, they did have a good pass rush, finishing tied for 3rd in the league in sacks with 47, which resulted in 384 yards lost by the opposing offenses.
#3 by Aaron Brooks G… // Oct 27, 2020 - 5:51pm
Offense seems to have more range in a given year than defense does, so even the Better-D teams are going to be dominated by their shitty offense.
Detroit and the Rams were utterly hopeless on any scrimmage play. Yet their immobility twice-outweighed their lack of resistance.
#75 by turbohappy // Oct 29, 2020 - 3:37pm
This season has had some weird home/away splits early so I wouldn't draw many conclusions yet. Will be interesting to see if home field is less of a thing in a season when there are few or no fans in the stands. Or if home field is more about not having to travel and the fans don't really matter too much.
#88 by dbostedo // Oct 31, 2020 - 6:06pm
There was a long-ish article in Sports Illustrated years ago (early 2000's I think) about what really is the source of home field advantage. (I can't seem to find it online anywhere...) They tried to tease it out of a variety of data across the major sports. Crowds and travel defintiely played a factor... but the biggest factor was found to be refereeing/umpiring, comprising maybe half of home field advantage. The home team tended to get a benefit (and crowds of course could factor into that too, in addition to direct impact on the team).
I wonder if a similar study would find the same with all the additional data we have today?
#29 by Pat // Oct 28, 2020 - 10:31am
It's not optimized for predictivity - it's a balance between being descriptive (telling you how well the teams *actually* did, rather than the score) with improved predictivity (correlating with itself a year later).
Also worth noting that it correlates with *point margin*, rather than winning percentage : as in, a +10% offense tends to score around 24-ish ppg versus a 0% defense. This matters because there are certain factors which correlate with *wins* better - a more consistent team (low variance) will, on average, have won more games than a less consistent one (assuming they're good).
Predicting who wins a game is always a bit of a silly exercise. Natively, there's not *that* much spread in team strengths in football. Consider a game between the top team last year (Baltimore) and the #10 ranked team (Tennessee). That's a 33.6% separation in DVOA: or around 7 points on a neutral field. Which is one touchdown. Effectively you're saying "I expect Baltimore to win 28-21", but *really*, that's like saying "I expect Baltimore to win 4-3," in which case Tennessee would obviously still win a huge portion of the time.
In other words, even if you developed some super-perfect predictive metric that was more precise than Vegas odds, you still wouldn't outdo Vegas odds by very much, because the "game winner" is mostly random.
#32 by theslothook // Oct 28, 2020 - 11:29am
"It's not optimized for predictivity - it's a balance between being descriptive (telling you how well the teams *actually* did, rather than the score) with improved predictivity (correlating with itself a year later)."
I believe it's optimized based on year to year correlation, not game to game or descriptive.
"because the "game winner" is mostly random."
I don't think this is true in all or most cases. If a great team at home faces a poor team on the road, I suspect the win probability is quite high. This might also be true if the venues were switched. Vegas odds may not pin down the scoring margin correctly, but the correlation between the magnitude of the margin and win probability is quite high.
#35 by Pat // Oct 28, 2020 - 11:53am
"I believe it's optimized based on year to year correlation, not game to game or descriptive."
It's balanced between year-to-year to itself and correlation to (points/wins): it's in the "does it work" article. Note that the article used to say points (at least... I'd swear it did, maybe I'm nuts) - it now says wins, so I'm not super-sure. Kinda don't understand the point of EW if it's tuned to wins now.
"but the correlation between the magnitude of the margin and win probability is quite high. "
I think you're misunderstanding what I'm saying? The correlation between Vegas margin and win probability is really, really good: however, the outcome of any game - on average - is something like 40%-ish random (so 'mostly' isn't correct here, to be fair, but it's within spitting distance). What I mean is that even if you knew with absolute certainty which team was better (in some 1D platonic ideal), you still would end up only predicting winners straight up around 60-some percent of the time.
This isn't exactly surprising: if you have 2 teams that are exactly equal (the Vegas margin is 0), you'd just say "??" and be right 50%, wrong 50%. So the degree to which you can predict winners is determined by how much the strength of a team spreads away from "even," and because football's very low scoring, the answer is 'not that much.'
To be fair the 60% bit is not quite true, you can beat that a little bit because of home-field: basically, the simple calculation that pops out "NFL games = 40% random" ignores the fact that there's a pull towards 8-8 because of HFA (because the 'true' separation of teams isn't that big, and the home-field randomness means you'd expect some goofiness in the standings since #3 might beat #2 and lose to #4 based on home field).
So, OK, better said, the "game winner" has significant randomness in it.
#37 by theslothook // Oct 28, 2020 - 12:00pm
All good points I agree with. I may have been engaging in pedantry.
This discussion made me think of win probability models. People I think treat win probability as a kind of static measure. IE the same things that determine win probability 20 years ago are the same today, but this is wrong.
#69 by Pat // Oct 29, 2020 - 12:29pm
It's week 7. Of course it doesn't. It literally says it right above: they're only at 70% strength right now.
The reason for this is that if you did try to do that at this point, the ratings would jump around like crazy every week because your idea of the strength of a team is changing so rapidly.
That being said you could easily imagine a full-strength adjustment based on DAVE, which is already attempting to smooth some of that out. But it'd *still* be weak because year-by-year teams regress to the mean (which is part of the preseason projection) but in-year, they don't. So you'd have DAVE being like "maybe the Jets won't be *that* bad again this year" whereas everyone watching them is like "nono, they still are."
Last year if you look at New England in Week 7, they were getting *massive* adjustments down due to opponent adjustments. Like, 15% or something. Full-strength adjustments in Week 7 would've pushed them basically to San Francisco's level.
The other thing to take into consideration is that DVOA is scaled *by season*. Which means that when you say "oh, NE was like the best team ever last year" you actually mean "NE was the best team *relative to the league* that we've seen at this point."
Why does this matter? Think about it. Imagine if you've got 5 teams, and their "era-independent" DVOA is +20%, +10%, 0, -10%, -20%. Average of 0, so the rescaled value's the same. Now make the bottom team *godawful*: like, -80%. Now the average is -12%. So we rescale based on that, and the teams become +32%, +22%, +12%, 2%, -68%. See what happened? All we did is take one team and make them horrible... and now we've got a bunch of much better teams.
Look at last year's week 7. See a similar pattern? NE/SF are historically good, right? But Miami's historically bad too. Now I'm not trying to say that NE/SF weren't *good* last year (same thing happened in 2005, but no historically great teams then) - they were, but the weak opponent adjustments and Miami's horribleness likely inflated that value somewhat.
And regarding their playoff loss: New England's "historically great" performance was their pass defense. Their run defense was okay. Their offense was meh. Who'd they lose to in the playoffs? Oh yeah. The Titans. Who attempted twenty passes (ish) out of around sixty. That's the benefit of being an offense - you get to choose.
#70 by Aaron Brooks G… // Oct 29, 2020 - 1:31pm
The rescale is a problem when multiple teams are openly tanking and skewing the mean, but...
If you have +20%, +10%, 0, -10%, -20%, but because of suckitude, we really have +20%, +10%, 0, -40%, -100%, and that becomes +42%, +32%, +22, -18%, -78% due to normalization. I'm not arguing with the +42. I'm saying the opponent strength adjustment doesn't properly handle when your schedule consisted only of the -18% and the -78%, and often those teams' 2nd string.
Something like the college system's expected wins for an average team seems to capture this idea more directly. That's a system more familiar with large variance in team quality.
#71 by Pat // Oct 29, 2020 - 1:55pm
"Something like the college system's expected wins for an average team seems to capture this idea more directly."
Yeah, but all that would do is tell you "the Patriots should win those games." It doesn't tell you what to do with the data that you get. A system like DVOA doesn't actually care who wins the games. It's just trying to get as much information as it can from each game, based on relative performance.
I do agree with you that I don't think opponent adjustments between cases like that are right, because I don't think they're possible - because there essentially is a floor in how good a team can do. Imagine if a team just, didn't have an OL for some weird reason. What happens? QB gets sacked every time they attempt to pass. So how do you evaluate 2 teams against them, then? They both get the same results. There's not really anything a defense can do better than "shut an offense down completely."
Obviously that's an extreme, but... it's mostly what happened with the Jets last year. The Eagles held them to 61 pass yards. The Patriots held them to 69 pass yards. Functionally, those numbers are both "zero." Was the Eagles pass defense performance better than the Patriots? You have no idea! The only thing you know is the Eagles and Patriots defense both passed some threshold in performance to reduce the Jets to zero. That's pretty much it. You can't measure how bad the Jets passing offense was, and you can't measure how good the defensive performance was. It was unmeasurable.
This is really just an offense problem, though, because there's no limit as to how bad an offense can be (there is a limit to how bad a defense can be - offenses still have to execute the play). So my guess would be that when you've got a super-bad offense you'll need to overcorrect: not because you can measure anything (you can't), but just because that performance was almost certainly not all because of the defense.
#78 by Lost Ti-Cats Fan // Oct 29, 2020 - 5:15pm
Isn't this, in the end, the same issue you get with college football, and trying to extract information from games against cupcakes? I recall this was discussed last year, as DVOA implicitly expects a base line competence from even the worst units, and if you think you've fallen below that threshold, you need to adopt new techniques. The counter argument was that the gap between even the worst NFL O and the best was still a far tighter range than college football, and DVOA's opponent adjustment formula can deal with the NFL level of variance.
To, which, per your comment "You can't measure how bad the Jets passing offense was", I can only say: true that. A man would go mad if he even tried.
#80 by Pat // Oct 29, 2020 - 6:48pm
Yup, exactly. Honestly stuff like this is usually temporary: there's just too much pressure for teams to *not* be horrible. So it really just ends up being just like injury noise. Yeah, correcting for a half season of crap Jets QBs when Darnold was back is silly. But trying to deal with that is *way* harder.
#83 by Lost Ti-Cats Fan // Oct 30, 2020 - 8:57am
I wonder if the true state of Adam Gase's offense is "incomprehensively awful", and it's only been the quality of Tannehill and Darnold that have raised it to the level of "awful"?
Perhaps Darnold's biggest weakness is "Gase is my coach", and like Tannehill he might revive his career should he find himself in different circumstances?
That whole pressure for teams not to be horrible thing I agree with. Presumably NYJ will succomb to it at the end of this season, after they've wrapped up a high draft pick.
#85 by dryheat // Oct 30, 2020 - 10:33am
Perhaps Darnold's biggest weakness is "Gase is my coach", and like Tannehill he might revive his career should he find himself in different circumstances?
I would bet on exactly this happening. There's nothing wrong with Darnold that a competent coach can't fix.
#38 by p_cj // Oct 28, 2020 - 12:28pm
I think DVOA essentially tracks success on a per play basis. So, what would be the difference in the following two scenarios:
1) QB throws three consecutive completions off 10 yards for a total of 30 yards in 3 plays.
2) QB throws two incompletions followed by a 30 yard completion for a total of 30 yards in 3 plays.
Or substitute RB for QB, whatever. End result is exactly the same for both scenarios but I assume one player would be considered vastly superior to the other?.
#11 by Pat // Oct 27, 2020 - 7:31pm
Where did that 10 point jump for Philly come from?? I mean, yeah, they beat the Giants, but it wasn't exactly a good performance. I know it's by play and not game, but you'd figure their 1 game rating would have to be like, +40% or so to jump that high.
#15 by horn // Oct 27, 2020 - 9:52pm
It's amazing that with no help on O, and missing some DBs/DLs, PHL was a missed FG from leading undefeated Pitt with a few mins left, and a 2pt away from tying the Ravens. Plus the 17-0 collapse vs the Skins in week 1, of course.
Team will be very different after the bye with the return of: Reagor, Goedert, Sanders, Peters, Seumalo, Alshon, Ertz, Driscoll/Johnson, Malik Jackson, Avonte Maddox.
#12 by jheidelberg // Oct 27, 2020 - 7:46pm
With the extreme teams in history in general, is there are correlation between having a great offense and bad defense?
Conversely, is there a correlation between having a great defense and a bad offense?
It seems to me that the game plan changes when you play an extreme team. I am off to the races trying to score when playing Seattle or Green Bay and throw caution to the wind as a turnover may be as bad as a punt. Would aggressive play by opponents create a worse Seattle/Green Bay defense?
Conversely, when playing Washington or Chicago, I know going in that I do not need a ton of points, and will play more close to the vest, making sure I do not turn the ball. By playing conservatively, am I making these defenses even better?
Also regarding your point that if you are going to win on defense it better be great. When I open things up trailing by
1 - 8 points at the two minute warning of the 4th quarter, will Washington or Chicago stop most offenses?
#66 by beargoggles // Oct 29, 2020 - 11:51am
This is my thought, that many good D/bad O teams probably aren’t really as good on D as their ratings. Opposing offenses play more conservatively to decrease variance and pad leads. Same could conversely be true. Good offenses leading to prevent defenses and garbage time.
One could study the former maybe: take these good D/bad O teams and look at the next season—and see what happens to teams that improve on offense the following year vs. the rest. Do those teams get worse on defense? Obviously this is imperfect but would be interesting. I’d probably exclude truly elite defenses as you probably can’t fake that.
#50 by jheidelberg // Oct 28, 2020 - 5:14pm
At least I would know that they would never be on the field.
I am not expecting production from a receiver coming off a torn achilles that has not played for 2 1/2 years and had no other offers. This reminds me of the Wizards signing Michael Jordan (ticket seller move). Am I missing something? Is there any NFL player to take off so much time and then after age 30 become a serviceable player?
As a Ravens fan, the impossible question to answer is, "Who is the best WR in Ravens history?" The Ravens have a history of picking up receivers past their prime, Derrick Mason, Steve Smith Sr., Anquan Bouldin. All grade B-C receivers at the time. Torrey Smith? Ravens GOAT? But maybe these are their best receivers ever. This, while whiffing on first round picks Travis Taylor and Breshad Perriman. Amazing how the Ravens have kept up with the Steelers over the past 2 decades, despite the enormous disparity in the QB/WR category, especially in the more recent pass dominant NFL.
Last year the Ravens (best offense in team history) had a #2 receiver that was my personal Blake Bortles Garbage Time Award winner Willie Snead IV with 31 receptions. The IV should stand for 4th quarter, when the TD's were plentiful with the Ravens already leading by more than 17 points.
#53 by jheidelberg // Oct 28, 2020 - 6:24pm
That answer is a winner for me.
Although only throwing 59 passes at age 44, I must give you credit for coming up with this name. He last played in 1993, then came back in 1998 started 1 game and came in relief in 7 others. DeBerg was never a very good QB looking back at this stats yet managed to play from 1978-1993 (yes must grade on curve due to the era). By quarterback rating, his age 44 season was actually the 3rd best of his career.
How on earth did you come up with Steve DeBerg? This is from the obscure, hardly known comebacks.
#56 by Richie // Oct 28, 2020 - 8:19pm
I just started thinking about various old players who I thought had a year gap (DeBerg, Flutie, Riggins, Moss, Marshawn Lynch).
If you look at PFR's "Adjusted Passing" section, DeBerg was probably a slightly above average QB. He just had the misfortune of losing his job to Montana, Elway, Testaverde/Young and then Montana again.
He had 9 seasons of 300+ pass attempts. 5 them were above average in passer rating. 7 of them were above average in ANY/A.
#57 by jheidelberg // Oct 29, 2020 - 12:00am
I feel that 2 1/2 years is a long time without playing, didn't really want to compare to a one year player as so many are simply injured (even Tom Brady).
I love your DeBerg info. His W-L record was awful, but his stats are from a different era, and I am so used to looking at modern day statistics that it never occurred to me that he may actual be good or even mediocre. I figured he was the Ryan Fitzpatrick of his time, playing for so many years, with a poor winning percentage and hopping from team to team.. The replacement information is great, he never had a chance to keep his job. Now that I feel that the Deberg to Fitzpatrick comparison is unfair to DeBerg, I challenge you or anyone to come up with statistics that tell me that Ryan Fitzpatrick is/was an above average QB.
#60 by Aaron Brooks G… // Oct 29, 2020 - 8:41am
Fitz has been above-average for a while now. He's been a 500-1000 DYAR guy over a projected season since about 2013. That's firmly in the above-average starter category. He's just notoriously high variance.
DeBerg was sort of a 'break in case of emergency' type guy. You didn't want to have to rely on him, as he was unreliable, but he could bail you out of a short-term crisis.
#67 by Richie // Oct 29, 2020 - 11:53am
The other thing about Fitzpatrick is he has basically only played on terrible teams. He started 12 games for the 2014 Texans who finished 9-7. And he started all 16 games for the 2015 Jets, who finished 10-6 (and missed the playoffs).
In 2015, the Jets were 10-5 and favored by 3 points in week 17. I assume they would have made the playoffs had they won. I don't remember the game, but it looks like high variance Fitz showed up. The Jets trailed 19-17 entering the fourth quarter, and Fitz threw interceptions on all 3 drives in the fourth quarter.
#72 by jheidelberg // Oct 29, 2020 - 2:27pm
Hmm... Steve DeBerg and Ryan Fitzpatrick (formerly know on this site as Fitz six picks) are above average. This was a six pick game not six pick season (making your point on high variance). Makes me realize why its so hard to win a Super Bowl without a top QB (yes we have exceptions like the recent Nick Foles, Peyton Manning final year and Joe (Montana for 4 postseason games) Flacco.
Next thing I know someone will come up with how Nick Foles and/or Joe Flacco are above average. Dare anyone to try on Peyton Manning's final year.
#65 by Richie // Oct 29, 2020 - 11:47am
My main memory of DeBerg is from the 1990(?) season when he broke a finger on his non-throwing hand. He played the rest of the season with a giant cast on that hand and did well.
#84 by CHIP72 // Oct 30, 2020 - 10:16am
The Steve DeBerg/Ryan Fitzpatrick comparison, without looking at their stats and adjusting for era, is a good one because both guys were/are viewed as journeymen quarterbacks who weren't good enough to be a quarterback on a Super Bowl contender but WERE good enough to make a team competitive in games and in seasons, so various teams were willing to play them regularly until they got someone else who was better. In DeBerg's case, twice his team eventually got someone better who became a Hall of Famer (Joe Montana with the 49ers and John Elway with the Broncos), so DeBerg is sometimes looked at unfavorably because he wasn't comparable to Montana or Elway.
DeBerg was probably about an average NFL quarterback in his career; I wouldn't argue strenuously if you wanted to say he was somewhat below average or somewhat above average. I think because DeBerg came into the league as an unheralded player (10th round draft pick) not kept even in his rookie season by his original team (Dallas Cowboys), and played regularly early in his career on a team that was not good (the 1978-80 San Francisco 49ers), he didn't have as long of a leash as average or near average quarterbacks who were high draft picks and expected to get better. Teams looked at DeBerg, similar to how they look at Fitzpatrick, as a "good enough" short term solution, rather than as a possible long term solution.
One other, unrelated item I want to mention about Steve DeBerg - I remember reading many years ago that Joe Montana said he personally liked DeBerg and often competed with him in a friendly way at games, like pinball and video games, outside of football when they were teammates. From what I remember reading, Montana was disappointed on a personal level when the 49ers got rid of DeBerg, who Montana considered a friend, even if he was pleased at a professional level because he (Montana) became the 49ers' undisputed starting quarterback. I've always thought it was cool that Montana and DeBerg were very competitive in a friendly way with one another.
#58 by jheidelberg // Oct 29, 2020 - 12:07am
Yes, after retiring for 3 years Deion Sanders did come back. It is hard to judge a defensive player, but the fact that he played two season for the Ravens showed that after the first year, he had shown enough to play a second year for the Ravens. Plus he played all 16 games in year two. He even had 5 INT's in the two years. Will Dez Bryant catch 5 passes?
Regarding best all time Ravens receiver can I have Ed Reed as the best Ravens receiver of all time, even though he was a safety?
#22 by Arson55 // Oct 28, 2020 - 7:19am
Dallas is clearly ranked too high because they have the worst defense I've ever seen, an offensive line that couldn't block a trio of four year olds, and a coaching staff that has no clue what it is doing. Gouging out my own eyes so I don't have to see what my team has become is way better than this. If I see anymore of this team, I'm going to cry, and yet I'll watch them every week regardless because I hate myself and want to suffer.
#27 by Pat // Oct 28, 2020 - 10:03am
Nope - if you search through the archives for "Forest Index" you'll see its origins. Estimated wins just translates certain portions of DVOA (red zone, performance in close games, variance) into a projected winning percentage, which is then just multiplied by number of weeks.
Why are there two measures? Because DVOA doesn't correlate to wins - it's kindof a "half-and-half" metric balanced between correlating with point differential in-season (which is a descriptive measure - as in, "here's what happened in the game") and correlating with *itself* the next season (which is a predictive measure - as in, "here's what will happen next"). But note even the "descriptive" portion *isn't trying* to correlate to wins, just *points*.
But that led to the idea of a "Forest Index" ('missing the forest for the trees') - if you just care about likelihood of *winning games* (which is *not* really a good measure of a team's strength) then certain portions are more important than others, even if they aren't predictive, or a good measure of point-scoring strength of a team. Variance is a good example: if a team scores 50 points one game, and 0 points the next, they'll average at 25 ppg, which is ~OK ish (+13% DVOA) - but they'll also likely be 1-1. A team with a lower DVOA and lower variance will score fewer points per game on average, but might win more. (For *really* bad teams this flips around - high variance leads to an *increase* in estimated wins. This prevented EW from going negative back in like, 2004, if memory serves, when faced with the godawful NFC West of the time).
Anyway, for the tl;dr crowd, EW generally increases week-to-week because it's multiplied by week number. But if a team's estimated winning percentage jumps around a ton (due to their performance) it can decrease, too.
#28 by Aaron Brooks G… // Oct 28, 2020 - 10:21am
There's no theoretical basis on which EW could go lower than 0. If the calculation can, there's a flaw in the calculation.
if you just care about likelihood of *winning games* (which is *not* really a good measure of a team's strength)
This, however, suggests the problem is in roster construction.
#30 by Pat // Oct 28, 2020 - 10:46am
"There's no theoretical basis on which EW could go lower than 0. If the calculation can, there's a flaw in the calculation."
That just takes all of the fun out of it! It's a rating which tends to correlate with winning percentage. Obviously, regardless of a team's intrinsic ability, they're capped at 0% and 100%, so nominally you want to transform that rating to a sigmoid/logistic-type function.
But in order to do that, you need lots of teams in those nonlinear areas (actually way more than you need in the linear portions) which you don't have. Early on there were no teams so far out of the linear region that you needed to worry at all. *Then* you got a team that was just so godawful bad you're like "seriously... wtf?" It turned out that part of the reason the team was *actually* winning was because the *average* was high, but the variance was high. So sometimes it would play like a much better team, and obviously, then it could win. That change fixed that.
But in the end I'm pretty sure there's just a hard cap/floor on the Forest Index projection to winning percentage now (world's simplest sigmoid transform) just so people don't go "how do you have negative estimated wins?!" I think that's lame, though, negative estimated wins is a lot more fun than comparing 0.01 vs 0.1 estimated wins.
#36 by Pat // Oct 28, 2020 - 11:56am
Oh, that's awesome. My kids keep asking me "if the Patriots played the Cowboys, who would you want to win?" (standard "impossible question") and my answer has always been "I'd want there to be some way that they both lose," and now I can use that as an example (although my dislike of the Patriots is more muted this season).
#41 by Aaron Brooks G… // Oct 28, 2020 - 1:38pm
The situation was this:
The APFA in 1921 was the wild west. Just a hilarious mess of assholes who couldn't conduct themselves in a professional manner.
It was a snowy mess, and Rochester was guaranteed $800 to play. Because the weather sucked, only a few fans turned out, and Washington could only pay $200. Under the then-rules, the home team decided whether or not to play due to weather reasons, and Washington was going to play. Rochester went home. Initially this was ruled as a forfeit by Rochester. Rochester pointed out Washington hadn't met their contract, and this was changed to a Washington forfeit, as they refused to pay. (And, in fact, never did.) Eventually the NFL decided this was stupid and declared the game a cancellation and pretended the whole stupid mess never happened.
#42 by t.d. // Oct 28, 2020 - 1:53pm
three bottom-half offensive teams in the top 5 overall is surprising, as is that two fo them are Baltimore and Pittsburgh. Also, that 13.1 special teams dvoa for Baltimore seems crazy-high, like historically-speaking
#76 by turbohappy // Oct 29, 2020 - 3:48pm
I can't say exactly, but I've noticed that it's actually oddly common that at least one team ends up with huge splits in past schedule and future schedule. You'll find some team with a split this big or close to it most weeks actually.
#79 by jheidelberg // Oct 29, 2020 - 6:43pm
I noticed that 3 of the top 5 teams DVOA (Baltimore, Pittsburgh, and Indianapolis) get virtually no help in being a top team from their offense. So much discussion here about offense vs defense. I noticed a comment previously from Aaron something to the effect that when doing DAVE projections, overall effectiveness was more predictive than separating offense, defense and special teams.
Does this mean that football outsiders position is that DVOA is the ultimate predictor of future success regardless of the splits between offense, defense and special teams?
I assume that with the large variance of offenses vs defenses and special teams, that this year is an anomaly having three of the top five DVOA teams having relatively mediocre offenses.
#86 by PTORaven // Oct 30, 2020 - 1:07pm
Hey, i wanted to let you know there's a bug with sorting DPI on that page. DPI is listed as number of DPI penalties "slash" yards of penalties, but if you try to sort it you'll see that Tom Brady ("11/215") gets listed between Drew Lock ("1/4") and Dwayne Haskins ("2/19") instead of being listed after Ryan Fitzpatrick ("8/108").