2013 DVOA Projections
by Aaron Schatz
The time has come for our annual preseason DVOA projections, updated from the projections that gave us the season forecasts in Football Outsiders Almanac 2013. Our DVOA projections suggest a Packers-Patriots Super Bowl, the same matchup forecast by last year's DVOA projections. The mean win projections suggest a Seahawks-Broncos Super Bowl, the same matchup forecast by last year's actual in-season DVOA ratings.
We must start with the requisite link to an explanation of DVOA. For anyone new to our site, DVOA stands for Defense-adjusted Value Over Average and measures a team's performance on every play of the season compared to league average in the same situation, adjusted for opponent. I know a lot of people may be coming here from various message boards and this is just going to look like a jumble of pointless numbers. Trust me, there is a method to the madness, and over the past dozen seasons past DVOA ratings -- as well as these multivariable-based DVOA projections -- have been a far more accurate predictor of future performance than wins or points.
Offense, defense, and special teams DVOA are all projected separately using a system based on looking at trends for teams over the past decade. The equations include a number of variables based on performance over the past two seasons in different splits (by down, passing vs. rushing, red zone vs. whole field) plus variables based on recent draft history, injury history, offensive and defensive pace, coaching experience, quarterback experience, and even weather. Strength of schedule was then figured based on the average projected total DVOA of all 16 opponents for 2013, rather than based on last year's performance.
The major difference between this year's preseason forecast and the forecast produced three months ago for Football Outsiders Almanac 2013 is unfortunately caused by an error. As one step towards producing team offensive projections, I create a quarterback projection that does not take into account any team variables. (This allows us to account, to give one example, for the fact that Carson Palmer should be expected to improve the Arizona offense.) The variable I used for Russell Wilson in the Seattle projection was pasted incorrectly, leading to a Seattle projection that was a little bit lower than it should have been. Fixing this moves Seattle a bit higher, making the Seahawks one of the top projected teams in the league. Don't worry San Francisco fans, we're still projecting you into the wild card.
The quarterback variable is also an important one explaining why a couple of teams are lower than previously projected. Despite the recent success of rookie quarterbacks, our "quarterback projection without team variables" numbers projected E.J. Manuel a little bit below Kevin Kolb and Terrelle Pryor significantly below Matt Flynn. Geno Smith and Mark Sanchez are pretty much a wash.
If your response to that paragraph was "the idea that Kevin Kolb might have been better than E.J. Manuel is insane," then please remember the following: The numbers we are presenting here are exactly what the projection system spit out. As we say every year: "A few of them will look strange to you. A few of them look strange to us." As always, the offensive projections come out in a wider range than defensive projections because offense performance tends to be easier to predict (and more consistent from year to year) than defensive performance. If you are looking for subjective projections, tomorrow we will be running our usual staff predictions article where we all talk about where we think the numbers are wrong.
The first postseason odds report of the 2013 season is also online, and I've added the playoff odds and Super Bowl championship odds to the table below. The mean wins forecast produced by our season simulation seems very conservative. In fact, this forecast is even more conservative than the forecast you found in Football Outsiders Almanac 2013 because due to time constraints, I did not have the opportunity to adjust our win projections to better reflect the recent historical distribution of wins in the NFL (fewer teams between 7-9 and 9-7 than you would expect from a normal distribution, and more teams around 12-4 or 4-12). However, this kind of conservative forecast generally leads to smaller errors than a forecast that looks more like a real set of final standings with the best team around 14-2 and the worst team around 2-14. Obviously, the best team in the league will likely have more than 11 wins, and the worst team will have more than 10 losses.
Surprisingly, despite slightly more conservative projections for mean wins, the new forecast raises the playoff odds of the top teams; we now have four teams with over 80 percent chance of making the playoffs instead of just the two we listed in Football Outsiders Almanac 2013.
Projected division champions are colored in light yellow. Projected wild card teams are colored in light blue.
| S.B. WIN
| S.B. WIN
167 comments, Last at 20 Oct 2013, 12:05pm
#5 by Scotty B. (not verified) // Sep 04, 2013 - 4:10pm
Yeah, I don't get it. I know Atlanta has a tough schedule, but the offense alone should account for 8 wins. If the defense steps up at all I see 10-11.
Still, this is a little scary since FO is more right than I prefer to admit.
Play the games!!
#26 by Bay Area Bengal (not verified) // Sep 04, 2013 - 6:16pm
Barnwell has been doing a team-by-team analysis over on Grantland, and his argument for Carolina's resurgence hinges on the Panthers' "bad luck" in close games.
As defined by FO, close games are games decided by 7 points or fewer, and this is a metric that typically has high season-to-season variance. Simply put, if a team plays 100 games against evenly matched teams, it's likely to go 50-50. If the game score is close at the end, it' was probably an even match-up, so when a team goes on a streak (good or bad) in close games one season, they are likely to regress to the mean the following season.
Last year, the Panthers were 1-7 in close games, so simple regression to the mean will probably boost their record by a couple of wins.
Carolina also had a top-10 weighted DVOA last year, including a top 10 offense, and they've upgraded their defense through the draft.
I'm still not sold, though. I like Cam Newton and I think Steve Smith is always overdue for a big season and a deep playoff run, but I'm not sure this is the year.
#32 by Yaguar // Sep 04, 2013 - 7:26pm
Panthers opponents were 35/37 on FG attempts. That's just nuts.
Typical sports message board idiots see stuff like this, don't realize what's happening, and concoct elaborate narratives about how Cam Newton isn't a "leader."
DVOA just shrugs and says "well, that won't happen again."
#57 by Aaron Brooks G… // Sep 05, 2013 - 9:16am
Not as nuts as you might think. Opponents only attempted 5 FGs longer than 50 yards against the Panthers, which is about where kicks become 50/50. Inside of 50, kickers, league-wide, are 87%. Carolina's opponents were 93% on these -- but man there were a lot of <30 yard attempts in there.
I think the problem is less kick luck, and more that Carolina's D didn't do anything until opponents were into the red zone.
#46 by SisyphusRocks (not verified) // Sep 05, 2013 - 2:10am
I assume you meant division. Although I expect the Rams defense to be improved (I'm surprised DVOA puts it at 23), the Rams are likely going to be behind SF and Seattle, barring big injuries. And I doubt their record will be better than Green Bay or New Orleans.
#4 by dan harmon (not verified) // Sep 04, 2013 - 4:10pm
Pittsburgh? They finished last year at -2.9% defensively and considering they're starting a nearly identical team on that side, one year older and one year more injured, I don't see a jump forward this year. I get that the calc's take into account all sorts of trends and such but I think that's definitely a case of predictions not meeting the eye test.
#6 by dan harmon (not verified) // Sep 04, 2013 - 4:13pm
I should also note that, as a Bears fan, I also see their DVOA prediction as... generous. The difference is that their ~-28% at the end of last year adds justification. The Steelers' numbers don't...
#25 by thok // Sep 04, 2013 - 6:06pm
If you give Pittsburgh a defensive DVOA of 0, they still project to be the sixth best team in the AFC, and thus a wild card. Unless some team gels, the AFC probably will have at most 5 good teams, and that's generously claiming Cincinnati, Baltimore, and Houston to be good instead of adequate.
#7 by Aaron Schatz // Sep 04, 2013 - 4:20pm
Just to give a general answer to most people's questions about why team X is higher or lower than you would expect... Hey! We wrote a book! Have you considered buying a copy of Football Outsiders Almanac 2013? It only costs $12.50 in PDF form! Still awesome even after the season starts! The Jets chapter explains their projected rebound! The Rams chapter explains their projected regression! Panthers chapter explains their projection and points out that Ron Rivera will probably screw it up! BUY NOW! All your questions answered! Click "Online Store" on the top of the page.
#161 by Lelouch vi Britannia (not verified) // Sep 10, 2013 - 1:32pm
How is that in any way a similar situation? The 49ers overperformed because they maintained exceptional health, developed an incredible running attack and replaced their barely-adequate quarterback with a vastly superior playmaker. You could also argue that the Plexiglas principle failed to take into account how big of an improvement it was to hire Harbaugh. These projections are based on the statistical performance of a team with the same coach and same quarterback for the past five years, who will almost certainly not be replaced by someone better at any point this season.
TL;DR, not even remotely comparable.
#110 by max // Sep 06, 2013 - 6:52am
I bought your book and tried to read it but I found it offensive and inappropriate. That whole passage on Bradford looking like a black QB was absolutely tasteless. I want my money back. I sent you an email requesting a refund, still awaiting your response.
#141 by Karl Cuba // Sep 07, 2013 - 9:54am
I was intrigued so I read the Bradford capsule. It isn't racist, they aren't attempting to 'determine' if he's black or white. The looked for his closest statistical comparisons and the top three were black, backup journeymen. It's a stats joke.
You are being a bit of a pudding.
#142 by Will Allen // Sep 07, 2013 - 12:00pm
Maybe you are just off your meds.
(edit) To add on, the paragraph on Bradford is also a joke abut the inanity of chattering about the melanin levels of qbs, which was s stupid feature of NFL punditry for decades, culminating in the Redskins/Broncos Super Bowl, where Doug Williams was asked, if I remember correctly, "How's it feel to be a black quarterback?" Of course the inanity in the punditry was a result of the inanity of NFL management and major college football powers, in being slow to put players with more melanin than average at the position. FO is also making a joke about the stupidity of paying attention to melanin levels.
Thus ends my instruction in the oh-so-subtle employment of irony. Please fill out your instructor evaluation prior to leaving.
#148 by Will Allen // Sep 07, 2013 - 4:07pm
Yes, if you are ignorant of the history of the NFL, the irony may be beyond you. FO is apparently guilty of, in your case, and whomever you showed it to, of giving too much credit for knowledge of the subject matter that their Almanac addresses, NFL football. This is a somewhat specialized publication.
Ignorance leads to anger. Anger leads to hate. This is the Path to the Dark Side! Turn away young Padawan!
#162 by Lelouch vi Britannia (not verified) // Sep 10, 2013 - 1:37pm
You mean you took a satirical passage about racism in the NFL completely out of context, showed it to your friends (doubtless with a good amount of pressure to confirm your opinions) and they thought it was racist?
Shocking, absolutely shocking. That's damning evidence if I ever saw it.
#150 by Marko // Sep 07, 2013 - 5:01pm
That Doug Williams story has been repeated often, albeit inaccurately. As the story goes, the question supposedly was, "How long have been a black quarterback?" But that question never was asked. The actual story, which is rather long, is detailed here: http://www.snopes.com/sports/football/williams.asp.
#158 by max // Sep 08, 2013 - 9:39am
I appreciate your perspective. Unfortunately, there was no reference to the Doug Williams story in the FOA article. Instead what we have are a few insulting posters attempting to sound more intelligent than they actually are.
The fact remains that the FOA article is distasteful and racist to many people. And insulting them probably isn't the best course to debate the case.
#11 by Will Allen // Sep 04, 2013 - 4:55pm
I think the mean wins on the Vikings is a little low, simply because I think they'll be good enough on the line of scrimmage to be better than that. However, they have a pretty good chance of starting the season 0-2, against two division opponents on the road, and given week 4 is in London (and who knows what effect that locale has), they don't play their 2nd true home game until week 6. Then, from week 7 on, they don't play a team that ranks lower than 19 by DVOA.
I take it back. Absent some more running on water by 28, or Ponder achieving a middling rank as an NFL starting qb (which may be the greater miracle), 5.5 mean wins sounds about right.
(in case you don't know it yet, Aaron, the login is a little buggy today)
#14 by KK Probs (not verified) // Sep 04, 2013 - 5:07pm
Aaron, I love your work and buy your books. I have a question that I have never seen addressed: to what extent is coaching already factored into the DVOA ratings? Is this an "assume all coaches are equal" rating? Or, can we assume that if the Panthers traded Ron Rivera for Bill Belichick DVOA would immediately notice and bump up the Panther win projection and drop the Patriot win projection with no other input?
#19 by Will Allen // Sep 04, 2013 - 5:33pm
It isn't as crazy as it sounds. The Rams have virtually zero chance of winning, or finishing higher than third in the division, given the rank of Seattle and San Francisco. The Vikings have a lot of things working against them.
Now, if the Vikings get a split of their first two games, their odds will improve significantly, I would think. If they mange to go 2-0, which would surprise me quite a bit, their chances will make a huge jump, I suspect.
#18 by Duff Soviet Union // Sep 04, 2013 - 5:29pm
THe one that stands out to me is KC being worse than Oakland.
I think you underestimate how bad a coach Romeo Crennel was. Reid will get them competitive, if not exactly good. I also don't see Oakland having a league average defense. 32nd ranked offence seems about right though.
#36 by craig in NS (not verified) // Sep 04, 2013 - 8:09pm
Hrm. Oakland, projected: -19.8 O, .6 D, 3.6% ST (3rd)
2012: -9.6%, 12.5%, -5.8%
2011: 2.6%, 9.6%, -1%
I see roughly zero chance that Oakland is as good as either the ST or Defense numbers. They might surprise on offense (to the tune of 'as bad as last year, but in different ways'). But even that isn't likely.
#96 by Eddo // Sep 05, 2013 - 5:31pm
Why would you assume coaching skill - unlike any other skill that exists in human beings - is constant over time?
People don't necessarily "forget how to coach", but they certain can get better and worse over time.
Two notable examples from 2000s: Joe Gibbs and Marv Levy.
#136 by Sifter // Sep 06, 2013 - 10:11pm
Reid's coaching skill may have declined - but we don't really know, as his lack of recent success may have been due to the Eagles players as much as his coaching. I would assume it was the players fault a lot more actually, whether it be on field execution (which was god-awful...thanks Eagles safeties!) or negativity in the locker room etc. Going to a new team and having his new players buy into his way of coaching could be all that Reid needs to be a successful coach again.
#155 by Bowl Game Anomaly // Sep 07, 2013 - 11:11pm
Joe Gibbs made the playoffs 2x in 4 years during the 2000s. All other Washington coaches combined have made the playoffs 2x in the 16 years Gibbs didn't coach since his first retirement, including last season. Gibbs may have gotten worse, but he was also better than you realize.
#23 by Mehllageman56 (not verified) // Sep 04, 2013 - 5:54pm
I realize I need to buy the book, but I really doubt New England ends up with the 2nd rated offensive DVOA for the year. They have a great offensive line, and Brady, but the losses at receiver and tight end will knock them farther than that. It's great that they got a bunch of guys in the draft, but rookie receivers aren't reliable. The book on them will be to force Brady to beat you deep with his 36 year old non-cannon of an arm. I still think they win the division, but they're not going to look that great doing it.
By the way, the Jets projected rebound would only be one game, not enough to keep the New York media from firing Ryan.
Whether that will get the guys who actually run the team to fire him is another matter.
#88 by Anonymousse (not verified) // Sep 05, 2013 - 4:21pm
What losses at WR?
They swapped Welker for Ammendola, and then improved the rest of their WRs. Deion Branch and Brandon Lloyd aren't losses, they were cut because better players were readily available.
Frankly, I think their WR group is better than it was last year.
TE is in rough shape, but Gronk played 11 games and Hernandez 10. Gronk will probably play just as much this year, and Hernandez, he's a loss, but not a huge one.
#91 by Will Allen // Sep 05, 2013 - 4:51pm
I suppose we could get into a semantic battle over the meaning of the word "huge", but I find it a puzzling statement about Belichik's roster evaluation talents to claim that the Patriots would offer the contract to Hernandez that they did, if suddenly removing him from the roster would be a mere "loss".
#123 by mehllageman56 (not verified) // Sep 06, 2013 - 1:03pm
He's still a rookie; whether he was drafted or not does not matter. Rookie wide receivers usually have seasons like the one Stephen Hill just had, where they show flashes but also make mistakes that cost games. Even players who were productive in their rookie years have bad moments; Wayne Cherbet and Jerry Rice both had huge fumbles that cost their teams games.
Perhaps I am overlooking Amendola, but I doubt he becomes Welker 2.0. He'll play well when he's healthy, which isn't as often as Welker. When both Amendola and Gronk are out, the Pats will have problems. So, yeah I doubt they end up with the 2nd rated Offensive DVOA. Probably top ten though, but not good enough to carry them to 12 wins. They'll win the division though, unless one of the young quarterbacks turn out to be the real deal this year, as opposed to next year, which is more likely.
#31 by Karl Cuba // Sep 04, 2013 - 7:11pm
I had a look at how the read option affected the offensive DVOA of the two teams that switched to using a lot of it during the season.
I looked at Seattle's ODVOA for the last five weeks of the season when they switched to using the read option and compared it to the preceding five weeks so as to avoid the early games when Wilson was operating with training wheels. The efficiency jumped from just above 20% to over 40% with the read option.
The other team was the 49ers, they ran less than twenty read option plays in the regular season, including the designed runs out of that look before Kaepernick took over. Their regular season ODVOA was 17% but when they wheeled out the read option in the playoffs their offensive efficiency jumped to over 50% during the three game pool playoff run.
While I acknowledge that the sample sizes aren't very big and simply averaging DVOA for individual games is not exact, this does look at two different teams working with largely the same personnel as they were using beforehand. The improvement is so large that I don't think it can be entirely discounted. Seattle's run game was particularly fecund.
Yes, I think that defenses will be better prepared but colleges haven't found a complete solution, that's why it keeps spreading. I also doubt that the niners offense will be 35% better than it was in last year's regular season. However, I do think the read option will continue to be highly productive if the qb can remain healthy.
I do have one axe to grind with regards to this projection (and yes, it's a niners fan complaining about the Seahawk's projection). The read option massively boosted the Seattle offense at the end of the year, does the projection account for the 49ers boost with the same wrinkle during the playoffs?
#98 by Bacchus44 // Sep 05, 2013 - 6:30pm
I don't think SEA ran read option more than 10% of the time after they put it in. But they really did take the training wheels off for Wilson and allowed him to throw the ball down the field a lot more. Now maybe the prospect of the read option opened up the pass game? But its actual use didn't affect the numbers all that much, because they didn't use it all that much.
#101 by Karl Cuba // Sep 05, 2013 - 7:29pm
Field Gulls, hardly an anti-Seattle site has put up numerous articles describing them as running out of the read option more than ten times per game, plus passing plays out of the look. That's where I got the number from.
Vic Fangio also described them as running 'ten plus' read option plays per game in the lead up to their clash near the end of the season. I wasn't making the number up.
#115 by KK Probs (not verified) // Sep 06, 2013 - 10:10am
Fairly dramatic year-over-year improvement from the Redskins as well: -7.0% in 2011 to 15.3% in 2012.
One thing all of these (Seattle, San Francisco, Washington) had in common was major QB changes. Had Alex Smith, Tavaris Jackson, and Rex Grossman been running the read option, I don't think we would have seen as large of increases. A lot needs to be done on separating the skill from the offense.
I am interested in the general concept this brings up, however, of grouping DVOA by offense type (or possibly defense for that matter). It seems like there's an enormous amount of untapped research to be done on this. Hopefully, it can be discussed in another thread where it's not buried. Good though, Karl Cuba.
#130 by Karl Cuba // Sep 06, 2013 - 3:03pm
I'd looked at Washington's figures but didn't include them because of the large turnover in their personnel. If you want another example look at the Panthers improvement after taking Newton and beginning to use some read option, 32nd in ODVOA to 5th, but again too many other factors changed.
What I liked about the SF and Seattle comparison was that both teams had provided a non read option portion of the season to act as a control.
#34 by jonnyblazin // Sep 04, 2013 - 7:52pm
The disparity between the teams in the AFC is interesting. 3 teams being the class of their division (NE, DEN, HOU) 3 good teams in the AFC North (BAL, CIN, PIT), and then a whole lot of garbage, with IND and NYJ being the next best projected teams at -7.1%.
#37 by justanothersteve // Sep 04, 2013 - 8:36pm
As much as I like the idea of the Packers having the second best team, I don't see it on the field. I think both Seattle and SF will be better in the NFC, both based on my impressions of the GB offensive line (and the harder to measure effects of OL to DVOA) and what Karl stated concerning the increased offensive efficiency of Sea and SF in the second half of the season.
#39 by merlinofchaos // Sep 04, 2013 - 10:49pm
I have no particular point, I just find this interesting.
Elite QBs: Brady, Brees, Manning, Rogers
Young possibly Elite QBs: Griffin III, Kaepernick, Newton, Russell
Good, maybe great but not Elite: Dalton, Flacco, Roethlisberger, Schaub
Roethlisberger in my eyes is always a couple ticks away from being elite. He's definitely great, to me. Flacco...well, he won a superbowl and he picked on Champ Bailey to do it. Schaub and Dalton are both competent, at least.
It says a lot about the power of an elite quarterback. The 4 young guns in there don't have enough games under their belts to be elite yet (and Kaepernick is the farthest from the title) but I'd be shocked if history doesn't record 2, maybe even 3 of them as being elite in 10 years.
#53 by rfh1001 // Sep 05, 2013 - 8:17am
I think 'great' is better than 'elite'.
Given that elite has weirdly become a sort of technical term for the best current QBs - and people get to argue whether Eli or Flacco might be it - I think great is the word you reserve for Brady and Manning. Brees probably. Rodgers has a good chance if he carries on just a few more years. Er. That's it.
I slightly worry that, by crapping on about the semantics of QB description, which is a world where the choice of words is mainly a tool to create meaningless, space-filling articles, I am being a tit.
Still: Post comment.
#79 by Bobman // Sep 05, 2013 - 3:18pm
Yeah, I noticed that too, but his stats were not quite there. He passes the eye test for sure, and based on what he was asked to do last year with no D and minimal run game (and no checkdowns, Bruce Arians, you psychotic monster!).... but based on the numbers, he's not there yet. Hamilton's offense should make him look like a different (and much better) man on paper at least.
#97 by merlinofchaos // Sep 05, 2013 - 6:17pm
I was listing the 12 QBs predicted by DVOA to be in the playoffs, which left out Luck, so that is the only reason he isn't on the list.
He also has the potential for greatness, though the season of Luck has been eclipsed rather a lot by the surprise emergence of RG3 and Wilson. But one season does not a career make! Ask Vince Young. :)
#40 by DoYouSeeWhatHa… (not verified) // Sep 04, 2013 - 11:01pm
I suppose my greatest solace as a Falcons fan is the spotty results of last year's projections.
Seattle and SF were supposed to be bottom 10 teams by DVOA
Minnesota was a bottom 5 team
St Louis was supposed to suck
Buffalo was supposed to be good
At least I know that DVOA projections don't hit at 85% or anything....
I do think the model is overselling Carolina given how the current NFL works. If the drive of modern offenses is to throw the ball more and defend the pass, and Carolina has poor receiver depth and possibly the worst secondary in the league, it stands to reason that they'll have a hell of a time pushing 9-10 wins...especially in a division where they face Top 8ish DVOA/DYAR QBs 4 times and an out of division schedule that will expose them to Kaepernick, Brady, and Wilson.
#42 by greybeard // Sep 05, 2013 - 12:08am
When a projection system says there will be only 4 win differential between the best team and worst team, all it is doing is minimizing the mean error of predictions and not really providing any insight on anything.
It is also saying that the best team is a total of 9 points better than the worst team. Last year New England scored 14 points more per game than their opponents on the average. All opponents ,not the weakest one. Shows how ridiculous to predict only 9 pts differential between the best and worst team.
These predictions are maybe slightly better than predicting all teams as 8-8. I never undstood why people get excited about these predictions.
At my workplace after every big project we do a post mortem, compare our original estimates to what really took place. The goal is to understand how successful we are in projecting the tasks and improve upon it. I can guarantee that we are light years better in predicting than FO (to be fair our job about predicting the schedule is much easier, since we have direct control over the outcome) yet we try to get better. I cannot remember I ever saw a review from FO of how successful the predictions for the previous year were. I guess that would kill the excitement about these pre season predictions.
#43 by jonnyblazin // Sep 05, 2013 - 12:39am
It's not that hard to extrapolate your own projections from the numbers though. Since the best defensive team has usually a -20% or more DVOA, and you think Chicago will have the best defense, you can input the numbers as you wish. All these projections are saying is that if you simulate the season a zillion times, some of the teams that you think are going to be great won't be great, and some of the teams you think will stink won't stink. So naturally the range of numbers are going to be compressed. But of course there are going to be extreme teams one way or the other, it's just foolish to predict which teams they will be with complete confidence. Except the Jets. I think they'll stink.
#44 by BaronFoobarstein // Sep 05, 2013 - 12:41am
I have a simple algorithm that I'm able to use to get pretty good scheduling estimates at work. First do a "rough draft" by estimating what the major items will be and how long they will take. This should be rough; if you have more than 4 items you're being too detailed. Now sum those estimates for your rough draft. Now throw all this crap away, make a gut-feel estimate and double it. Viola!
#106 by greybeard // Sep 06, 2013 - 1:33am
I am going from memory based on what FO writers posted in the past. I may have misremembered, that is possible.
I don't see why it would not scale linearl though.. They write comments like "The second ranked team is closer to tenth than first ranked team". You cannot say things like that if it is not linear.
NURBS is out of fashion by the way. It is triangles all the way.
#114 by Revenge of the NURBS (not verified) // Sep 06, 2013 - 8:43am
NURBS will return, and make them all pay!!!
"You cannot say things like that if it is not linear."
Sure you can. All it means is that the 2nd through 10th teams are all about the same, and that the #1 team is far ahead of the rest. DVOA doesn't scale linearly with points. The calculation is much more complicated than that. And if you think about it, it makes sense, because football is much more complicated than that. The object isn't to score the most points possible; the object is to score more points than the other team, and there are a variety of ways to achieve that end.
#51 by Darragh (not verified) // Sep 05, 2013 - 8:09am
Outcomes will obviously be a comination of skill and luck. Some of the 6 win teams will have very bad luck and finish around 2-4 wins and some of the 10 win teams will have very good luck and finish around 12-14 wins.
Cant see how the Aaron / FO should somehow incorporate that luck. Like if we had 32 coins and each was tossed 16 times the average number of heads projection for each would be 8. Obviously that wont be the outcome but projecting 8 for each is not some effort to minimise mean error - it is an actual estimate of the number of heads a coin will generate in 16 games.
PS as a rams fan I think the rankings are totally ridiculous
#56 by willisisgod (not verified) // Sep 05, 2013 - 9:00am
What they should do is take the typical variance range for any given year and apply it accross the teamsat each rank. Moving the pats and pack to 13 and Raiders to 4 and so on. But historically these previews have been average at best as projections. Vegas does a much better job. Don't get me wrong I like DVOA more than almost anything 4 weeks in because it is great at evaluating recent past performance. But these are the same ptojections that thought the Jets would be Ianthe playoffs last year
#95 by Anonymousse (not verified) // Sep 05, 2013 - 5:04pm
It's not good luck though, it's excessive regression leading to smaller overall error bars.
For example, lets look at NE:
YEAR PRED ACTUAL DIF
2012 11.9 12.0 .1
2011 10.2 13.0 2.8
2010 10.0 14.0 4.0
2009 11.1 10.0 -1
2008 12.0 11.0 -1
2007 11.8 16 4.2
2006 9.6 12.0 2.4
2005 9.7 10.0 .3
Thats as far back as the predictions go, but its a pretty stable pattern. The Patriots are about 2 wins better than the projection if you assume +/- 2 games for luck. Even when Brady missed the whole season, they still almost hit the projection.
They're ~12 games low over 7 seasons. That's not luck, that's overly conservative projections, or regressing things to 8-8 too heavily.
NE is predicted at 27.4 DVOA. The two teams in that range last year won 11.25 games. In 2011 there were 4 teams between 20 and 28 DVOA. They averaged 13.25 wins. In 2010, there were 3 teams between 20 and 22 (none closer) and they averaged 10.67. 2009? 6 teams in the 20s, 11 wins. If my math is right, thats over 11.5 games averaged for teams in the mid 20s... So why is our highest prediction 10.6 games, despite having 3 teams with DVOA higher than 20?
To save face. Its the same thing they accuse coaches of, better to be a little bit wrong all the time, than really wrong once or twice. Nobody gets fired for punting on 4th and 3.
If you look at the perennially bad teams, you'll find the opposite, FO continually predicting 6 or 7 wins while the team wins 3-5 and occasionally everything goes right and they hit their projection.
#104 by Perfundle // Sep 05, 2013 - 7:53pm
What a coincidence that you picked the one team with the most conservative projections. How about Philadelphia as a counterexample?
YEAR PRED ACTUAL DIF
2012 8.3 4 -4.3
2011 10.3 8 -2.3
2010 9.2 10 0.8
2009 9.1 11 1.9
2008 11.1 9.5 -1.6
2007 10.9 8 -2.9
2006 9.9 10 0.1
2005 11.7 6 -5.7
Or, San Francisco as a counterexample in the other direction?
YEAR PRED ACTUAL DIF
2012 7.5 11.5 4
2011 7.7 13 5.3
2010 6.2 6 -0.2
2009 5.4 8 2.6
2008 5.4 7 1.6
2007 8.3 5 -3.3
2006 4.8 7 2.2
2005 4.7 4 -0.7
In fact, over the last 8 years, the lowest mean-squared error versus the actual records would've involved regressing it even closer to 8-8, by 30%. I have to imagine the FO staff knows this stat, but they also know that if they regress it further, the teams would be even more bunched together and you'd have more of this kind of outcry. The fact that that extreme level of regression is the optimum means that the kind of predictions you'd prefer have a far greater chance of massively missing a lot of teams than these safe DVOA ones.
It's also why Vegas' team win predictions look exactly like DVOA's as far as regression toward 8-8 goes. If they had come out with stuff like Oakland 3 wins and Denver 13 wins, then betting that every team would be closer to 8-8 is practically guaranteed money.
#109 by greybeard // Sep 06, 2013 - 2:07am
You picked two teams that have been inconsistent. Also showed that DVOA is not good at tracking inconsistent teams.
It is not important whether they are just cutting the distance to 8-8 or doing something else. What matters is their predictions do not have any dynamic range at all. Everybody is clustered. The best team is slightly better than mediocre. The worst team is slightly worse than mediocre. That is never happened in NFL, there is always a guarantee that a team sucks so bad that the fans become experts about the draft prospects by November.
BTW, as far as I know Vegas is not in the business of predicting what NFL teams will do. They are in the business of predicting what the gamblers think NFL teams would do. There is a big difference.
#122 by Perfundle // Sep 06, 2013 - 12:58pm
How is Philadelphia inconsistent? Until last year they'd been hovering between 6 and 11. Why didn't you mention that Anonymousse picked a team that is extremely consistent, and thus isn't a good example either?
So you'd prefer them to pick 12-4 teams and 4-12 teams, and watch them miss massively for many of them? Because that is what is going to happen every time. Their job is not to please you and pick "reasonable" looking records that have dynamic range. I assume their job is to minimize their error. That is neither done by picking all 8-8 records nor is it done by picking a bunch of 12-4 and 4-14 records.
As for your comment about Vegas, if there was a sure-fire strategy that would win money every time, then all the gamblers would pick it and bankrupt Vegas. Vegas is in the business of not giving away money, and picking team wins of 3 or 13 is giving away money.
And, since you asked to see how well these predictions have been, I went and checked the last seven years of predictions against the lines set by the Las Vegas Hilton Race & Sports Book, and you would've earned $1080 (betting either the over or the under amount) over those seven years, which is pretty good.
#125 by Anonymousse (not verified) // Sep 06, 2013 - 2:48pm
"Anonymousse picked a team that is extremely consistent, and thus isn't a good example either?"
The fact that they're so consistent is exactly why they're a good example of the problem. They consistently finish with DVOA +20%-+40%, yet the system continually predicts them to win as many games as a +10% team.
Yes, I'd prefer them to pick 12-4 teams and 4-12 teams and miss massively when something drastically changes. If the Broncos should be a 14-2 team, I want them to say the Broncos should win 14. Not say the Broncos should win 10 beacause Peyton Manning might be eaten by a lion while visiting the denver zoo.
They're trying to hedge issues like Peyton Manning's Neck, or Brady's Knee into the predictions, when they should just be ignoring those, as they're not predictable, and they're not predictive. Yes, injuries happen, but season ending injuries to starting QBs are something altogether different, and don't happen all that often.
#132 by Anonymousse (not verified) // Sep 06, 2013 - 3:08pm
To use a baseball analogy, what their system is doing would be similar to continuing to predict that Mike Trout is going to hit .280/.350/.445 (10-15% better than league average) every year while he continues to hit .330/.400/.550 every season.
#163 by Perfundle // Sep 12, 2013 - 3:03pm
Good lord, if you play the 2012 season 10000 times the Broncoes are not going to be a 14-2 season every time. They're probably going to be averaging 10 wins, which is what the algorithm here spits out. Giving you only 1 out of 10000 seasons worth of simulations provides no knowledge whatsoever.
As for new England, you really don't understand how stats work, do you? Given 32 teams there will be some teams who look like they're consistent, just as if you flip a coin 6 times there will be about 1/32 of those where the coin hits heads or tails all 6. It doesn't actually mean they have consistency built into them.
#127 by greybeard // Sep 06, 2013 - 2:49pm
"How is Philadelphia inconsistent? Until last year they'd been hovering between 6 and 11. Why didn't you mention that Anonymousse picked a team that is extremely consistent, and thus isn't a good example either?"
It does not matter which example you choose. In everyone of them you will see the FO prediction will be in 6-11 range. That is my point. Regardless of which team you choose it will be in that range. "You suck, you are the worst team in NFL: 6 wins, You are great, you may be breaking all NFL records: 11 win,s you are neither how about 7-10 wins?"
"So you'd prefer them to pick 12-4 teams and 4-12 teams, and watch them miss massively for many of them? Because that is what is going to happen every time. Their job is not to please you and pick "reasonable" looking records that have dynamic range. I assume their job is to minimize their error. That is neither done by picking all 8-8 records nor is it done by picking a bunch of 12-4 and 4-14 records."
Their job (not really though, none of this is their real job, these are just products on the way to earning them advertising and consulting money) is to provide insight, not minimizing some error metric.
"As for your comment about Vegas, if there was a sure-fire strategy that would win money every time, then all the gamblers would pick it and bankrupt Vegas. Vegas is in the business of not giving away money, and picking team wins of 3 or 13 is giving away money."
No idea how this relates to my comments. All I pointed out that was what Vegas predicts is different than what FO predicts. So comparing them is not really valid.
"And, since you asked to see how well these predictions have been, I went and checked the last seven years of predictions against the lines set by the Las Vegas Hilton Race & Sports Book, and you would've earned $1080 (betting either the over or the under amount) over those seven years, which is pretty good."
I am not a gambler, so I would not care how they compare to Vegas. I am curios though, since Vegas lines moves from the first day they are available to the season opening day, how does the book capture how much money one would have made? I would guess that it would be different based on which day you put the gamble.
Also what is the initial investment here? I made $1080 by risking how much?
#164 by Perfundle // Sep 12, 2013 - 3:07pm
This provided plenty of insight to most people here. I'm sorry it did not provide insight to you.
I mentioned the success of these predictions because you asked for them when you said
"How about proving the "haters" wrong by showing us how successful these predictions are?"
I'm not sure what you'd prefer me compare them with, if not Vegas.
#124 by mehllageman56 (not verified) // Sep 06, 2013 - 1:15pm
Having a quarterback the level of Brady will let a team consistently beat the projected win total. I believe Barnwell wrote a column about it, comparing Peyton's win totals and the projected win total by DVOA for the last decade.
#52 by LKN (not verified) // Sep 05, 2013 - 8:14am
Anyone who watched Atlanta & Carolina saw a tale of 2 teams who were definitely on opposites ends of "lucky" spectrum. Atlanta played a lot of teams (like Denver) early in the season before they got significantly better. Atlanta's division record was 3-3 (only losses to division foes and they lost to every one and should have lost to the Panthers twice had it not been for a fluke pass at the end of one game. Atlanta does have a collection of perhaps the best skill position players in the game on offense, but their offensive line is a huge liability and they have one of the worst Front 7's on defense in the NFL. DVOA certainly recognizes this and projects this.
#54 by Will Allen // Sep 05, 2013 - 8:19am
The Bears keep hiring coaches with Minnesota ties, that people, who have direct knowledge of the coaches as human beings, tell me are guys who deserve to be rooted for. I still think Tice is a better coach than circumstances have allowed him to show, which has resulted in a lot of ridicule, some of it deserved, a lot of it less so, to directed at him.
I've been hearing great things about Trestman for more than 35 years, because I know guys who were his teammates at the University of Minnesota. These are the same guys who spoke of Dungy in reverential tones, before Dungy was even on the Steelers, and they were only slightly less glowing in their description of Trestman as a leader. I also think some odd circumstances occurred which resulted in Trestman waiting a long time for this opportunity, and that his track record suggests that if anyone can get Cutler to better achieve his potential, it is Trestman. I'm kinda' at the point in my football fandom where I'd rather root for people than laundry, when I know enough about the people, especially given how stadium politics turns me sour. I can't completely stop a habit that started when I was seven or so, so I will still root for the Vikings, but I really want to see the Bears do well, too, and I think they have a pretty good chance.
#72 by JoeyHarringtonsPiano // Sep 05, 2013 - 12:57pm
I don't really have a dog in this fight. I read this site for entertainment, and it has more insight into why teams and players are/aren't successful than other football/entertainment sites out there. I think looking at these predictions are for fun, nothing more. I don't ever remember FO claiming that their predictions are infallible.
#77 by BroncFan07 // Sep 05, 2013 - 2:51pm
Aaron, haters gonna hate you, and everyone that does season predictions, until you can accurately and precisely predict everything that will happen in the future. And then they will hate because you ruined any need to watch the season. To save time, maybe you can just pick everyone's favorite team to to go the Super Bowl.
#69 by Deemo15 // Sep 05, 2013 - 12:50pm
KC has a top 10 caliber defense this year, so to see them worst in the league is interesting to say the least. And any projection that has them below Oakland and significantly behind the Jets is extraordinarily interesting.
I suspect this will be a team that will drastically outperform these projections.
#73 by Mehllageman56 (not verified) // Sep 05, 2013 - 1:03pm
Yes, the ranking suprised me as well, until I looked at last years dvoa ranking for them: 31st. With as much talent as they have, I would expect them to have performed better last year too. But to be suprised at FO projecting KC's defense to be worse than the Jets is silly; the Jets have ranked in the top ten in defensive DVOA the last four years under Ryan. They also used two first round draft picks on defensive players, and their huge loss from last year, Revis, only played a game and a half.
Now, projecting the Jets to have the 2nd best DVOA for defense may be a little much, but they'll probably be in the top ten again.
#76 by greybeard // Sep 05, 2013 - 2:17pm
When I looked at it I saw that they missed on 6 playoff teams. They had SF as 22nd best team , SEA as 28th best team. KC as the 16th and PHI as 12th and NO as 5th best team. They had the SuperBowl winner as not qualifying for the playoffs.
And what is it with the "attacks on the computer"? Are you being sarcastic?
#83 by Ryan D. // Sep 05, 2013 - 3:46pm
They admitted they were wrong about SF.
They expected Seattle to suck, because no one saw Russell Wilson coming before last season started. NO ONE. Even Seattle didn't see it coming, because they paid Matt Flynn a lot of money to handle a clipboard.
KC and Philly were decent teams that had bad luck and then had the players quit on the coaches. There was certainly enough talent for those to be around average performing teams.
New Orleans should have been a highly projected team, since there is no way to project what happens when a team loses its head coach. Add in a change in defensive coordinators, and the single largest defensive implosion ever, and you get a team that still went 7-9. No one could have predicted that the Saints would allow more yards than any other team in history.
Let's be honest here, Baltimore shouldn't have even won their division. They had to convert 4th and forever with a miracle dumpoff to Ray Rice against the Chargers just to do it. If that play happened 100 times, they might convert it 5 times, at best. Then, they would still have to win the game after converting it. At 9-7, they still would have made the playoffs (as the 6th seed), but they would have went on the road to Houston, instead of hosting the Colts. Do you still think they would have won (in order) at Houston, at Denver, at New England/Cincinatti or hosting the Colts, and then in the Super Bowl?
Who would have guessed that a 4th and forever conversion for Ray Rice could be so valuable?
#90 by Aaron Brooks G… // Sep 05, 2013 - 4:39pm
"They admitted they were wrong about SF."
Sort of. Grudgingly. Despite missing by 4 wins and 37% DVOA, they also attempted to argue that 11-4-1 actually did represent a substantial regression to the mean.
The computer seems to have caught up to the Harbaugh era this year, though.
#103 by Karl Cuba // Sep 05, 2013 - 7:33pm
Special teams, defensive turnovers, close game performance all regressed. This was the crux of their argument.
If you insist on only looking at the mean projection then that's really on you. Nobody is making you read their stuff.
#105 by greybeard // Sep 06, 2013 - 1:14am
Predictions is silly business. All your excuses about why they missed their predictions is just another proof of that. Maybe they should stop spending their time on silly stuff and use it for something more productive.
#128 by greybeard // Sep 06, 2013 - 2:54pm
I am not grumpy at all. On the contrary I have been in a pretty good mood. That is why I bothered to pick up an "internet debate" -something that rarely adds anything to one's well being. I do not do that when I already feel down, internet debates usually make me feel more miserable.
#78 by cjfarls- (not verified) // Sep 05, 2013 - 2:55pm
Folks get way too uptight about projections "missing" in the past and somehow claiming that makes this year's projections dubious.
for example, DVOA didn't like SEA, WAS and Indy last year because rookie QBs tend to suck. The projection isn't wrong because Wilson, RG3 and Luck are the exceptions to the rule... they are the EXCEPTIONS! Projections are built on the typical, not the exception.
Similarly for teams that didn't regress due to injury luck (i.e., projecting average injury isn't wrong, even though it is very possible for a team to be healthy or hurt 2 years in a row).
There are many things that can't be projected objectively. Other things can't even be projected subjectively (e.g. specific injury luck). Hindsight is 20-20, but no model has it. I can make a outmybutt projection that Peyton Manning gets hurt in game 1 and Denver goes back to the cellar of the AFCW with the Raiders... but even if I'm correct and that happens, that doesn't mean that is actually a good projection. Its a wildass guess, and should be judged as such.
So rather than getting fixated on what is wrong with the model (I assure you there is lots, and they will address some of those things tomorrow), pay attention to what the model is saying that is useful.
#82 by Chemdrj (not verified) // Sep 05, 2013 - 3:33pm
I also believe that they normally have a disclaimer about normally being wrong due to the favorite being 35% lock to win which means that they are wrong 65% of the time. I think being 6 of 12 is better than they expect.