Week 10 DVOA Ratings
by Aaron Schatz
Now that we're halfway through the season, the Football Outsiders DVOA ratings are working with a lot of data. Our ratings tend to look at more of the past than subjective power ratings and even sometimes other "advanced stats," because the research we've done has shown us that you get a more accurate picture looking at the whole season than just the last few weeks. There are a couple of important implications for this.
First, one game doesn't move the needle as much as you might think. This is particularly true at the top and bottom of the rankings where teams are spread a little bit further apart than they are in the 10-20 range. So Green Bay only moves up two spots this week to No. 3 after blowing out Chicago. Philadelphia moves up only one spot to No. 5 after routing Carolina.
Second, some teams will be ranked higher or lower than expected because they played very well or poorly early in the season. Even the weighted DVOA ratings at this point still count every game with at least 67 percent strength, even all the way back to Week 1. So, for example, I had a Twitter user suggest that our ratings are undervaluing the Patriots, who are still only eighth in total DVOA and ninth in weighted DVOA. The issue here is their performance early in the season, but more specifically, one game weighing them down: the Week 4 blowout by Kansas City. The single-game DVOA ratings for the Week 1 loss to Miami and the Week 3 narrow-escape win over Oakland are very close to zero, but that Kansas City game is at -72.7%. If the Patriots continue to play well in the next couple weeks, it will show that their performance over the past few weeks is a more accurate barometer of how good they are, and they'll move up. The way the weighted DVOA system works, the first big change in a game's strength occurs eight weeks after it is played, so unless the Pats have a big loss in the next couple weeks, they'll move up a bit after Week 12.
New Orleans is in a similar situation, except that it isn't just one game dragging the Saints down. The Saints have three games with a single-game DVOA under -30% in the first five weeks. But DVOA loves their big Week 8 win over Green Bay, and surprisingly, it also loves their narrow Week 7 loss to Detroit. As a result, there's a huge gap between the Saints' DVOA in Weeks 1-5 and their rating in Weeks 6-10. If you wanted to look at just the last few weeks, here are the teams with the biggest gap in DVOA between Weeks 1-5 and Weeks 6-10:
| Biggest DVOA Rise, Weeks 6-10 vs. Weeks 1-5 | Biggest DVOA Drop, Weeks 6-10 vs. Weeks 1-5 | |||||||||||
| Team | DVOA Wk 1-5 | Rk | DVOA Wk 6-10 | Rk | Change | x | Team | DVOA Wk 1-5 | Rk | DVOA Wk 6-10 | Rk | Change |
| NO | -15.0% | 25 | 36.6% | 2 | -51.7% | x | SEA | 43.5% | 2 | -3.9% | 18 | 47.4% |
| JAC | -52.6% | 32 | -4.6% | 19 | -47.9% | x | SD | 14.1% | 6 | -29.7% | 28 | 43.8% |
| PHI | 3.3% | 16 | 41.5% | 1 | -38.2% | x | CIN | 16.7% | 5 | -24.1% | 27 | 40.9% |
| MIA | -1.7% | 20 | 32.3% | 3 | -34.0% | x | NYG | 8.6% | 9 | -30.0% | 29 | 38.6% |
| NYJ | -31.2% | 30 | 2.6% | 15 | -33.8% | x | ATL | 11.9% | 7 | -22.2% | 26 | 34.1% |
You'll notice that the top three teams of Weeks 6-10 are on that table, and you may be wondering where the No. 1 overall Denver Broncos are. The Broncos are the No. 4 team of Weeks 6-10, dropping from 53.1% DVOA in Weeks 1-5 to 30.1% DVOA in Weeks 6-10.
Two interesting teams which are nowhere near either of these tables are the Buffalo Bills and Kansas City Chiefs. Both of those teams have been about the same over the last five weeks as they were over the first five weeks, although I should note that "how Kansas City was over the first five weeks" includes both their horrible Week 1 loss to Tennessee and that big blowout of New England in Week 4. Anyway, looking at the DVOA ratings you may notice that the Bills move up and the Chiefs move down even though Kansas City won on Sunday. (The Chiefs are ranked No. 10 same as last week, but their DVOA drops by two percentage points.) This was a classic example of a game where DVOA reflects that the "wrong team won." Of course, the wrong team didn't really win; the Bills were the ones who had bad timing, made bad decisions, and couldn't hold onto the ball. But when it comes to elements of the game that are predictive for future performance, Buffalo was superior on Sunday. The biggest issue, of course, was that Buffalo fumbled the ball three times and the Chiefs recovered each one. This game is dramatically different if Scott Chandler had been able to hold onto the bouncing ball after Bryce Brown fumbled while running towards the end zone. The DVOA system also gives the Bills credit for getting the ball down to the 15 on that final attempt at a game-winning drive, even though they couldn't get it into the end zone once they made it down there. Overall, the Bills gained 5.1 yards per play, and the Chiefs only 4.7 yards per play.
* * * * *
Since this is the first week with opponent adjustments at full strength, I wanted to also point out which players and teams have suffered from or been gifted by particularly difficult schedules this season.
Quarterbacks: Carson Palmer was having a reasonable season before his ACL tear on Sunday, but honestly, it wasn't a great season. Palmer ranked 12th in unadjusted YAR but 19th in DYAR because of the Arizona schedule. Palmer had faced only one defense in the defensive DVOA top ten (Philadelphia) while Drew Stanton faced two of them in his three starts (Denver and San Francisco). With the AFC West and NFC East on the schedule, and the decline of the Rams defense, Russell Wilson and Colin Kaepernick also get penalized in DYAR once we add opponent adjustments.
The two quarterbacks with the toughest schedules so far are two quarterbacks on opposite ends of the performance spectrum. Andrew Luck is an MVP candidate in part because he's done what he's done against a tough schedule; he ranks fourth in YAR but second in DYAR. In particular, the Colts have faced teams that are much better against the pass than the run, such as Cincinnati (No. 8 vs. pass, No. 32 vs. run) and Houston (No. 11 vs. pass, No. 27 vs. run). On the other hand, you have Derek Carr. At times he's looked good as a rookie, and at times he's looked totally lost. But one reason he sometimes looks totally lost is that the Raiders have played a tough schedule. Carr is 31st in YAR but 26th in YAR. Tom Brady and Philip Rivers also have a gap of over 100 between YAR and DYAR due to tough schedules. The Chargers have faced the toughest overall schedule of opposing defenses, when you combine passing and rushing. Their average defensive opponent has had -7.3%, making them the only team that has faced what is on average a top-ten defense.
Running Backs: Today's Quick Reads details how the Seahawks are threatening the all-time record for team rushing DVOA. One reason is that while they have played an easy schedule of opposing pass defenses, they've played a tough schedule of opposing run defenses. Marshawn Lynch has 231 rushing DYAR right now but only 197 rushing YAR. Andre Ellington and Frank Gore also get boosted in DYAR by opponent adjustments. So do both Giants running backs, Rashad Jennings and Andre Williams.
The biggest gap on the other side belongs to Justin Forsett, who is fifth with 136 rushing YAR before adjustments but drops to 12th with 91 DYAR after adjustments. And in case you are wondering, DeMarco Murray is getting pretty much no schedule benefit, with 278 YAR and 277 DYAR.
Wide Receivers: The biggest boosts in DYAR from opponent adjustments generally belong to the wide receivers of the AFC South, and the top four gaps between DYAR and YAR belong to Allen Robinson (who just broke his leg, ending his rookie season), Kendall Wright, Justin Hunter, and T.Y. Hilton. The Baltimore receivers have also played a particualrly difficult schedule. Tampa Bay's Vincent Jackson has the biggest gap in the other direction by far, with -48 YAR before adjustments but -92 DYAR afterwards. That's a gap of -44; no other receiver has a gap over -30.
Tight Ends: There are a lot of tight ends who get small penalties in DYAR for playing easier schedules but nobody really gets a big penalty. However, there are three tight ends who have a gap of at least 20 between YAR and DYAR due to playing tough schedules. Jimmy Graham goes from fourth without opponent adjustments to first among tight ends after opponent adjustments. Jared Cook goes from 18th to 12th, and Delanie Walker from 17th to 15th.
[ad placeholder 3]
Defenses: Yes, the New York Jets secondary is basically a tire fire at this point, but the Jets defense hasn't been quite as bad as it has looked. The Jets have played the toughest schedule of offenses so far this year, with their average opponent having 5.1% DVOA. And that schedule has been dramatically slanted towards the passing game. The average Jets opponent has had 26.9% passing DVOA, but only -13.6% rushing DVOA. The only defense to face an easier schedule of opposition running games is Miami (-15.1%). Both the Jets and Dolphins have played the Patriots (No. 2 passing, No. 31 running), Chargers (No. 3 passing, No 28 running), and Packers (No. 5 passing, No. 12 running). Interestingly, there are a number of teams who have faced the opposite: lots of offenses tilted towards the run. Washington, Green Bay, Dallas, and St. Louis have faced four of the five easiest schedules of pass offenses (Atlanta is the other) but the four hardest schedules of run offenses. These teams have shared opponents such as Minnesota (No. 31 passing, No. 3 rushing even without Adrian Peterson) and Seattle (No. 22 passing but No. 1 rushing).
Returning to overall offensive DVOA of opponents, other teams to face particularly tough schedules this season include Jacksonville, Carolina, and Oakland. Philadelphia, Atlanta, and Pittsburgh have faced the easiest schedules of opposing offenses.
* * * * *
 |
Once again in 2014, we have teamed up with EA Sports to bring Football Outsiders-branded player content to Madden 15 Ultimate Team. Each week, we'll be picking out a handful of players who starred in that week's games. Some of them will be well-known players who stood out in DVOA and DYAR. Others will be under-the-radar players who only stood out with advanced stats. We'll announce the players each Tuesday in the DVOA commentary article, and the players will be available in Madden Ultimate Team packs the following weekend. We will also tweet out images of these players from the @fboutsiders Twitter account on most Fridays. One player each week will only be available for 24 hours from the point these players enter packs on Friday.
The Football Outsiders stars for Week 10 are:
- QB Aaron Rodgers, GB (24-HOUR HERO): Led all Week 10 QB with 208 DYAR (18-for-27, 315 yards, 6 TD).
- LOLB Connor Barwin, PHI: 8 combined tackles, 3.5 sacks.
- SS James Ihedigbo, DET: 4 Defeats (2 TFL plus an interception and a pass defensed that both prevented third-down conversions).
- P Brett Kern, TEN: Averaged 51.8 gross yards and 48.8 net yards per punt on eight punts.
- C Max Unger, SEA: Helped Seahawks RB gain 133 yards on 17 carries with 88 percent Success Rate on runs up the middle.
A quick apology to Kansas City Chiefs fans. We considered doing both Allen Bailey and Ron Parker this week. Bailey had eight combined tackles including three tackles for a loss on runs. Parker had three passes defensed on that final Buffalo series where the Bills tried to score from the Kansas City 15, and he also stripped the ball from Bryce Brown that resulted in a touchback. However, we wanted to make sure Detroit and Tennessee fans got a Football Outsiders player in MUT, as we had not done anyone from those teams yet. We'll make sure to get to Bailey or Parker next time one of them has a nice big stat line. (Doing Kern and Ihedigbo leaves Chicago and Carolina as the last teams without FO players in MUT, but they certainly aren't getting anything this week. Yuck.)
* * * * *
All stats pages are now updated with Week 10 information -- or will be in the next few minutes -- including FO Premium, snap counts and playoff odds. This week also brought the return of our weekly playoff odds report for ESPN Insider. There's additional commentary over there on the current playoff race, including analysis of how the odds of getting a bye week will change for Indianapolis and New England depending on who wins the big Sunday night game. The playoff odds report for ESPN Insider also means that playoff odds will be updated earlier on Tuesdays here on Football Outsiders.
* * * * *
[ad placeholder 4]
These are the Football Outsiders team efficiency ratings through ten weeks of 2014, 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 for strength of schedule and 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. WEIGHTED DVOA represents an attempt to figure out how a team is playing right now, as opposed to over the season as a whole, by making recent games more important than earlier games. As always, positive numbers represent more points so DEFENSE is better when it is NEGATIVE.
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>
| TEAM | TOTAL DVOA |
LAST WEEK |
WEIGHTED DVOA |
RANK | W-L | OFFENSE DVOA |
OFF. RANK |
DEFENSE DVOA |
DEF. RANK |
S.T. DVOA |
S.T. RANK |
|
| 1 | DEN | 37.3% | 1 | 36.2% | 1 | 7-2 | 23.4% | 1 | -17.7% | 2 | -3.8% | 25 |
| 2 | BAL | 23.0% | 2 | 23.7% | 2 | 6-4 | 8.2% | 12 | -6.4% | 9 | 8.4% | 2 |
| 3 | GB | 21.3% | 5 | 21.3% | 3 | 6-3 | 20.6% | 2 | -0.7% | 11 | 0.0% | 15 |
| 4 | SEA | 20.7% | 4 | 19.4% | 5 | 6-3 | 13.1% | 5 | -9.9% | 7 | -2.3% | 23 |
| 5 | PHI | 19.9% | 6 | 21.0% | 4 | 7-2 | -1.2% | 17 | -10.6% | 6 | 10.5% | 1 |
| 6 | MIA | 15.5% | 3 | 16.6% | 6 | 5-4 | 4.1% | 13 | -16.0% | 3 | -4.6% | 27 |
| 7 | IND | 13.0% | 8 | 14.5% | 7 | 6-3 | 11.0% | 8 | 3.6% | 18 | 5.6% | 5 |
| 8 | NE | 13.0% | 7 | 12.7% | 9 | 7-2 | 10.5% | 9 | 4.7% | 22 | 7.1% | 3 |
| 9 | DET | 10.6% | 12 | 10.5% | 10 | 7-2 | -6.5% | 22 | -24.6% | 1 | -7.5% | 32 |
| 10 | KC | 10.4% | 10 | 13.5% | 8 | 6-3 | 8.3% | 11 | 1.4% | 16 | 3.5% | 8 |
| 11 | PIT | 8.6% | 9 | 9.7% | 11 | 6-4 | 17.3% | 3 | 7.6% | 25 | -1.1% | 19 |
| 12 | NO | 6.8% | 11 | 8.8% | 12 | 4-5 | 13.6% | 4 | 8.3% | 26 | 1.4% | 12 |
| 13 | BUF | 6.3% | 14 | 5.2% | 14 | 5-4 | -14.5% | 27 | -14.6% | 4 | 6.2% | 4 |
| 14 | DAL | 6.2% | 15 | 6.7% | 13 | 7-3 | 11.4% | 6 | 3.8% | 19 | -1.4% | 21 |
| 15 | ARI | 3.3% | 16 | 2.8% | 15 | 8-1 | -10.2% | 25 | -13.2% | 5 | 0.4% | 13 |
| 16 | SF | 0.4% | 20 | -0.1% | 16 | 5-4 | -4.3% | 20 | -9.7% | 8 | -5.0% | 28 |
| TEAM | TOTAL DVOA |
LAST WEEK |
WEIGHTED DVOA |
RANK | W-L | OFFENSE DVOA |
OFF. RANK |
DEFENSE DVOA |
DEF. RANK |
S.T. DVOA |
S.T. RANK |
|
| 17 | CLE | -0.2% | 19 | -1.0% | 17 | 6-3 | -1.0% | 16 | -0.7% | 12 | 0.2% | 14 |
| 18 | SD | -2.7% | 17 | -3.0% | 18 | 5-4 | 11.3% | 7 | 13.7% | 31 | -0.3% | 16 |
| 19 | ATL | -3.7% | 18 | -4.5% | 19 | 3-6 | 10.0% | 10 | 17.5% | 32 | 3.8% | 7 |
| 20 | CIN | -4.1% | 13 | -7.4% | 21 | 5-3-1 | -4.1% | 19 | 4.2% | 21 | 4.3% | 6 |
| 21 | NYG | -7.3% | 23 | -5.5% | 20 | 3-6 | -2.1% | 18 | 4.0% | 20 | -1.3% | 20 |
| 22 | WAS | -7.9% | 22 | -9.6% | 22 | 3-6 | -0.5% | 15 | 3.1% | 17 | -4.3% | 26 |
| 23 | HOU | -11.6% | 25 | -12.0% | 23 | 4-5 | -4.9% | 21 | 0.9% | 15 | -5.7% | 31 |
| 24 | NYJ | -12.3% | 28 | -12.3% | 24 | 2-8 | -15.0% | 28 | -0.4% | 13 | 2.3% | 9 |
| 25 | MIN | -14.8% | 27 | -14.1% | 25 | 4-5 | -17.4% | 29 | -0.8% | 10 | 1.8% | 11 |
| 26 | CHI | -15.4% | 21 | -18.6% | 27 | 3-6 | -0.3% | 14 | 10.0% | 30 | -5.1% | 29 |
| 27 | TEN | -17.1% | 26 | -19.0% | 28 | 2-7 | -8.0% | 23 | 8.5% | 27 | -0.7% | 17 |
| 28 | STL | -19.8% | 30 | -17.4% | 26 | 3-6 | -11.2% | 26 | 7.5% | 24 | -1.1% | 18 |
| 29 | CAR | -21.0% | 24 | -23.8% | 30 | 3-6-1 | -9.9% | 24 | 9.5% | 29 | -1.7% | 22 |
| 30 | OAK | -22.0% | 29 | -21.1% | 29 | 0-9 | -19.4% | 30 | 4.7% | 23 | 2.2% | 10 |
| 31 | JAC | -26.2% | 31 | -24.3% | 31 | 1-9 | -22.1% | 32 | 0.4% | 14 | -3.8% | 24 |
| 32 | TB | -35.8% | 32 | -35.5% | 32 | 1-8 | -21.4% | 31 | 9.2% | 28 | -5.3% | 30 |
- 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).
| TEAM | TOTAL DVOA |
W-L | NON-ADJ TOT VOA |
ESTIM. WINS |
RANK | PAST SCHED |
RANK | FUTURE SCHED |
RANK | VAR. | RANK | |
| 1 | DEN | 37.3% | 7-2 | 39.1% | 8.7 | 1 | 2.6% | 9 | -2.4% | 22 | 13.1% | 11 |
| 2 | BAL | 23.0% | 6-4 | 29.6% | 6.7 | 4 | -5.6% | 28 | -3.1% | 24 | 10.9% | 8 |
| 3 | GB | 21.3% | 6-3 | 20.9% | 6.3 | 7 | -2.8% | 24 | -0.7% | 17 | 15.9% | 15 |
| 4 | SEA | 20.7% | 6-3 | 17.6% | 6.7 | 5 | -1.8% | 22 | 2.6% | 12 | 15.9% | 14 |
| 5 | PHI | 19.9% | 7-2 | 20.9% | 7.1 | 2 | -8.6% | 32 | 3.2% | 9 | 9.6% | 5 |
| 6 | MIA | 15.5% | 5-4 | 22.5% | 6.5 | 6 | -0.5% | 18 | 5.7% | 5 | 20.7% | 27 |
| 7 | IND | 13.0% | 6-3 | 15.6% | 6.9 | 3 | 2.5% | 10 | -6.3% | 29 | 10.4% | 7 |
| 8 | NE | 13.0% | 7-2 | 15.8% | 5.4 | 11 | 0.1% | 16 | 7.4% | 2 | 16.8% | 18 |
| 9 | DET | 10.6% | 7-2 | 15.9% | 6.1 | 8 | -1.0% | 20 | -6.2% | 28 | 9.6% | 6 |
| 10 | KC | 10.4% | 6-3 | 12.0% | 6.0 | 9 | 2.3% | 11 | 3.3% | 8 | 22.6% | 30 |
| 11 | PIT | 8.6% | 6-4 | 8.6% | 5.2 | 14 | -4.8% | 27 | -2.0% | 20 | 18.3% | 21 |
| 12 | NO | 6.8% | 4-5 | 10.4% | 5.2 | 15 | -4.1% | 26 | -6.9% | 31 | 19.8% | 24 |
| 13 | BUF | 6.3% | 5-4 | 7.0% | 5.3 | 13 | -0.8% | 19 | 7.5% | 1 | 7.1% | 4 |
| 14 | DAL | 6.2% | 7-3 | 7.7% | 5.4 | 12 | -5.9% | 29 | 3.7% | 7 | 15.3% | 12 |
| 15 | ARI | 3.3% | 8-1 | 6.7% | 5.4 | 10 | 0.4% | 13 | 5.6% | 6 | 3.0% | 2 |
| 16 | SF | 0.4% | 5-4 | -2.3% | 4.8 | 18 | 3.2% | 7 | 0.7% | 14 | 12.6% | 10 |
| TEAM | TOTAL DVOA |
W-L | NON-ADJ TOT VOA |
ESTIM. WINS |
RANK | PAST SCHED |
RANK | FUTURE SCHED |
RANK | VAR. | RANK | |
| 17 | CLE | -0.2% | 6-3 | 9.8% | 5.1 | 16 | -6.5% | 30 | 0.3% | 15 | 16.3% | 16 |
| 18 | SD | -2.7% | 5-4 | 0.1% | 4.8 | 20 | 3.7% | 6 | 6.0% | 3 | 20.1% | 25 |
| 19 | ATL | -3.7% | 3-6 | 4.0% | 4.8 | 19 | -8.1% | 31 | -0.3% | 16 | 18.8% | 22 |
| 20 | CIN | -4.1% | 5-3-1 | -2.1% | 5.0 | 17 | 0.4% | 14 | 1.9% | 13 | 25.7% | 31 |
| 21 | NYG | -7.3% | 3-6 | -17.9% | 4.3 | 21 | 5.6% | 3 | -6.4% | 30 | 21.7% | 29 |
| 22 | WAS | -7.9% | 3-6 | -7.9% | 4.1 | 23 | -3.0% | 25 | -3.4% | 25 | 17.7% | 20 |
| 23 | HOU | -11.6% | 4-5 | -7.0% | 3.1 | 28 | 0.0% | 17 | -5.4% | 27 | 4.4% | 3 |
| 24 | NYJ | -12.3% | 2-8 | -16.3% | 4.0 | 25 | 6.7% | 2 | 3.0% | 11 | 11.8% | 9 |
| 25 | MIN | -14.8% | 4-5 | -12.3% | 4.3 | 22 | -1.0% | 21 | -2.4% | 23 | 19.3% | 23 |
| 26 | CHI | -15.4% | 3-6 | -19.9% | 4.1 | 24 | 4.5% | 5 | -4.5% | 26 | 15.4% | 13 |
| 27 | TEN | -17.1% | 2-7 | -15.0% | 3.4 | 27 | 0.3% | 15 | -2.3% | 21 | 20.6% | 26 |
| 28 | STL | -19.8% | 3-6 | -19.8% | 3.1 | 29 | 1.2% | 12 | 3.1% | 10 | 17.1% | 19 |
| 29 | CAR | -21.0% | 3-6-1 | -27.5% | 3.7 | 26 | 5.6% | 4 | -8.6% | 32 | 16.6% | 17 |
| 30 | OAK | -22.0% | 0-9 | -31.5% | 2.1 | 32 | 7.0% | 1 | 6.0% | 4 | 2.8% | 1 |
| 31 | JAC | -26.2% | 1-9 | -27.2% | 2.4 | 31 | 3.1% | 8 | -1.9% | 19 | 21.3% | 28 |
| 32 | TB | -35.8% | 1-8 | -29.7% | 2.6 | 30 | -2.8% | 23 | -1.4% | 18 | 28.0% | 32 |
Comments
208 comments, Last at 17 Nov 2014, 4:50pm
#1 by Will Allen // Nov 11, 2014 - 6:22pm
I still don't have any strong opinion as to who gets the NFC title. Nobody jumps put at me as having markedly fewer weaknesses, or being so strong in one area that the weaknesses are masked.
If I had to pick a team down the DVOA list a ways, I guess I'd guess the Lions, simply because their defensive line smacks people around, and their qb has the talent to get hot and put together 3 really good games. Heaven help them if they need to kick a forty yarder to tie a game, however.
#8 by swgallagher // Nov 11, 2014 - 6:39pm
in my view, the top 3 teams in the NFC are GB, Seattle and Detroit.
Seattle sure hasn't looked the part, but they are the only team in that bunch that has had a legitimate shot to win every game they've played which is generally a very telling statistic.
Detroit has the D and the play-making O to win anywhere; they seem play-off built to me.
I feel bad that AZ isn't getting more love, but I think Clayton said it best - they make the play-offs and are one and done. These next two weeks for them will be incredibly enlightening...
I fill similarly about Philly. There is too much data on Sanchez to suggest he will be terrible. Right?
#122 by Otis Taylor89 // Nov 12, 2014 - 8:11am
I can't believe I'm saying this, but I think after viewing Sanchez in the preseason and in two regular season games, I have more confidence in him than going into the playoffs in 2014 than I would have in a 2nd year Russell Wilson circa 2013.
I can't believe I'm saying this in part due to the fact I also witnessed the Buttlefumble.
#127 by Will Allen // Nov 12, 2014 - 10:31am
The NFC East is giving us an example of the value of coaching in the NFL. Earlier this year, I was pretty high on the Cowboys, and then I watched them mismanage their way to a loss at home against an inferior Washington roster. Yes, Romo's back is one reason I'm no longer thinking of the Cowboys as a strong contender, but just as big a factor for me is the Cowboys being a management/coaching fiasco; somehow, some way, I just believe they are going to screw things up, even if their defense continues to surprise.
I believe the guy who coaches the Eagles,by contrast, does not have an owner as a major impediment, is organized as hell, and has real insight as to how to efficiently optimize player and team performance. I'm really looking forward to see if the play of Sanchez for the next 3 months confirms my belief. This Sunday at Lambeau will be an excellent early test of the theory.
#156 by TecmoBoso // Nov 12, 2014 - 2:12pm
Agree. If the Eagles are coached by just an average coach, they're 5-4 probably, maybe even 4-5. My concern with the Eagles is the schedule. Easiest in the league so far and 1-2 in games against top 15 of DVOA, and should be 0-3 as they got lucky in the Colts game. That leaves six wins against teams 21 or lower in DVOA.
Green Bay sure does seem like the best team in the NFC, but they might not even win the North which would make their path to the Super Bowl really tough.
#162 by Will Allen // Nov 12, 2014 - 2:58pm
Another one of the reasons I'm really interested to see the Eagles play in Lambeau this weekend is because I think the Eagles defense matches up well with Rodgers Inc, even if the metrics indicate that the Eagles are better against the run than the pass. I haven't seen them a ton, but it seems to me that they rush the passer well, which is critical against a hugely talented qb who likes to wait as long as possible to throw the ball.
It'll also be interesting to see Kelly's scheme against a Packers defense that now has Matthews playing in the middle.
#166 by ElJefe // Nov 12, 2014 - 3:37pm
If the Eagles win in Lambeau this Sunday I expect they will have home-field throughout the NFC playoffs. What makes this game intriguing to me is that the Eagles have the highest-variance defense I have ever seen. Some defenses are "bend but don't break"; the Eagles practice the earthen dam: Hold resolute and then fail spectacularly.
DVOA overrates them (#6) because they don't give up sustained drives, but they have to be among the league "leaders" in giving up long runs and long passes. It's almost as if the DC (Billy Davis) recognizes that 3/4 of his secondary is usually in trouble if they have to sustain coverage so he excessively blitzes to force early throws while Williams/Fletcher/Allen are still in the same zip code of the men they are supposed to cover.
I think the game on Sunday can go two ways: The Eagles overrun the Packers OL and blow Green Bay out, or they don't get to the passer and Rodgers throws for at least 4 TDs. It seems to me that a close game will only happen by accident.
Overeducated Layabout
#202 by chemical burn // Nov 15, 2014 - 4:36pm
"Hold resolute and fail spectacularly" is an amazing description and easily the best I've seen of the Eagles D this year. Green Bay I think might be the absolute worst match-up for them in the league because their biggest weakness (giving up gigantic plays to the outside wr's) is GB's biggest strength (huge plays to their outside wr's.) Also, the fact Rodgers is such a smart, match-up conscious QB means he's obviously noticed that "find Nate Allen and throw the ball at him" would be a great strategy if "find Nate Allen and throw the ball five to ten yards away from him to the man he is supposed to be providing over the top help on" weren't even better. Once teams realize that bracket coverage with Williams and Allen means a wr is actually not covered at all, they begin to throw the deep ball at will. If the blitz doesn't get home, the Eagles D is just helpless. Fortunately, the blitz gets home or is at least disruptive a healthily amount of time and Jenkins in particular has been really smart about baiting QB's into short, meaningless outlet throws. That Barwin, who has been amazing this year and deserves serious DPOY consideration, has been their most effective pass defender this apart from Jenkins (in combined terms of coverage and batted balls) says a lot.
Sometimes the Williams/All failure is an obvious disaster, like the end of Arizona game which... gah. What the good-goddamn hell - just play with 10 men on the field and don't put Allen out there. At least no one will be under the impression he'll understand when a wr is being passed off to him in zone or expect him to be providing his assigned over the top protection. But look at the end of the Panthers game - if you didn't watch it, you might assume that the Eagles D gave up those late touchdowns because they switched to a prevent defense or eased up on the blitzing, but that's not true: they kept coming after Newton but once he realized that the double-coverage on Benjamin was effectively meaningless, he began to move the ball with alarming ease. On the other side of the field, Fletcher is better than Williams, but again he's just awful against the deep ball when he has to sustain coverage - in 49er's game he was a disaster whenever Kaepernick broke out of the pocket. If you gave me an O/U of 400 yards for Rodgers this weekend and forced me to bet everything I have on it, there's just no way I could take the under. If you dropped it down to 350, I wouldn't even blink.
I know some very good teams like all those Lovie Smith Bears teams lived off of a stout run-defense, turnovers and return TD's, but all that stuff just feels like smoke and mirrors to me. I mean, is the plan really to rely on Sproles punt returns, fumble recovery luck and pick-sixes deep into the playoffs? I just can't buy it...
#206 by thebuch // Nov 16, 2014 - 7:33pm
Good thing you didn't take that bet, Packer fans know Rodgers doesn't get huge yardage totals because in games like this he gets to stop passing around halfway through the third quarter (give or take half a quarter) and he probably won't go over 350 yards, even if they put up over 50 points.
#165 by ElJefe // Nov 12, 2014 - 3:28pm
Well ... if you want to call the Eagles "lucky" to beat the Colts, you would have to consider them "unlucky" to lose to the Cardinals. They pretty thoroughly outplayed Arizona and still lost. Even after having watched the game I can't rationalize how 520 yards of offense became 20 points; the Eagles didn't turn the ball over that much.
And the other game they lost (@ SF) was when they were starting offensive lineman #7, #8, and #9. No team can be competitve with that line-up; I suspect the breaking point for OL quality in the NFL is between #7 and #8, since most teams won't dress three backup lineman on gameday. Teams are fine up until #8 has to play and then it's open season on the QB.
Overeducated Layabout
#21 by Rick_and_Roll // Nov 11, 2014 - 6:58pm
In the NFC, I think Seattle is the team to beat and will come out on top, as a Bronco fan I hope I'm wrong... Don't be surprised to see them win out and at least get a bye, if not home field. I'm not sold on Detroit (QB, running game), Dallas (Defense, QB) or Philly (QB, too many wins/points from ST & TOs). Arizona is too banged up. GB could give them a game in GB, but not in Seattle.
While their defense isn't all time great like last year, it is still very good. Below are my pro/cons for Seattle, and I think their cons are very manageable.
Pros
1.Wilson seems to have gotten better than he was last year.
2.The running game is very solid.
3.Experienced, well coached and best home field in NFL.
Cons
1.Their OL seems shaky in the non-play action passing game and Seattle isn't the type of team to win a shootout if they aren't getting turnovers.
2.Their pass rush has declined from incredibly good in 2013, to just good in 2014
3. They are getting average to below average play by the CBs opposite of Sherman, which is probably getting exposed because of Con #2.
#23 by Will Allen // Nov 11, 2014 - 7:08pm
Now that I think about it, and look at the schedule, if the Eagles win in Green Bay on Sunday, I like their chances in the playoffs a lot, assuming they don't have to play their 3rd string qb. They have the home field when Seattle plays them on December7, and I think coaching will be a real difference when they play Dallas twice. I thought the game in lambeau was the most interesting contest this Sunday, but now I'm really intrigued.
#25 by dmstorm22 // Nov 11, 2014 - 7:13pm
I have to argue with the pros, to me Wilson hasn't gotten any better. Don't want to start any #HotTake Wars, but to me he's stagnated from where he was as a rookie. Now, he was very good as a rookie, but his passing is below replacement level this year. His dropoff in passing is worse than his improvement in rushing metrics.
Also, I think their pass rush has dropped further than 'just good'. They're 5th worst in ASR, and have a hard time getting sustained pressure.
#30 by jacobk // Nov 11, 2014 - 7:27pm
I would hesitate to put Wilson in the Pros column so far this year (other than for rushing), but I do think there's a decent chance he's rounding into form. The Seahawks installed a Harvin focused lateral passing offense that they ran for the first six weeks of the season to gradually less and less effectiveness. After the Harvin trade they shifted back to last year's formula of runs, play action deep shots, runs, and pick plays.
There have been some growing pains with the rookie receivers and Wilson has been a shade off on his deep balls. Last year he was relatively accurate on deep shots so I would tend to attribute the misses more to rust than to inability. If he gets those issues tuned up I don't see any reason he can't go back to putting up the typical Russell Wilson stat line of 18/25, 200-250 yards, 2 TD, 0 int... if he can do that on a consistent basis, I like Seattle's chances.
There's a whole separate conversation to be had about why you would drastically overhaul your offense right after winning the Super Bowl, but at least it didn't torpedo the whole season.
#31 by Perfundle // Nov 11, 2014 - 7:35pm
There's a whole separate conversation to be had about why you would drastically overhaul your offense right after winning the Super Bowl, but at least it didn't torpedo the whole season.
I don't know what the alternative is when your best receiver leaves and is replaced by a receiver you broke the bank for who required a special scheme from him to be effective, and a scheme that looks to be working spectacularly in his limited appearances in 2013 to boot. It's pretty clear now that the best option would've been not to trade for Harvin in the first place so Tate could've been extended, although I'm not sure if he wouldn't still have left considering his stated desire to get more targets. Not being able to attract good receivers/having receivers being unhappy with their limited role in the offense is the biggest problem with any run-heavy offense, as Harvin was unhappy for the same reason.
#34 by jacobk // Nov 11, 2014 - 7:49pm
I kind of assumed they would run the same offense with Harvin in the Tate role and then sprinkle in the occasional bubble screen and jet sweep as constraint plays. The decision to make bubble screens and jet sweeps the core of the offense is still a little baffling.
I've seen claims that Harvin just isn't very good at running routes down the field and/or doesn't like doing so, but I have a hard time completely believing it. He made some nice downfield catches last year (the one in the Minnesota game stands out in my memory), so I don't get why you would decide to only throw him the ball behind the line of scrimmage.
Obviously I'm not an offensive coordinator so I'm sure they had their reasons, but I'm happy to see them go back to a formula that worked before.
#39 by Perfundle // Nov 11, 2014 - 8:12pm
I think you underestimate how many bubble screens Seattle was running even before Harvin. Just look at Wilson's passing chart from last year:
http://www.footballoutsiders.com/images/FilmRoom/FilmRoom-082114-05.jpg
It wasn't so much that they ran too many but that Harvin wasn't as effective as Tate in running them.
Also, I would disagree that the Harvin-led offense suffered a steady decline in effectiveness. They were very effective in their first game and very ineffective in their last game, sure, but they were still great against Washington if you include all those TDs that got called back on penalties that didn't affect the play. They hardly used Harvin at all against San Diego, and Denver has proved to have a great defense, so basically you're left with a one-game stretch of offensive inefficiency, and you can't have a sample size smaller than that. There was the locker room fighting and insubordination stuff, but that's separate to his impact on the offense.
#123 by gomer_rs // Nov 12, 2014 - 9:46am
I may be totally off base but I think the poor O-line play started to effect Russell Wilson's throwing mechanics sometime last year and is still doing so. At the top of his drop on long passes and sometimes medium he looks like he's taken a little hop step, almost as if to doge a blindside rusher, that is putting his feet in the wrong spot for his throws and messing his timing and accuracy.
This is also why I think he does so much better from the shot gun where he can see if his tackles whiff on block and his footwork gets better since he doesn't have to assume they do.
_______
I remember when they were the Sea-chickens.
#29 by Perfundle // Nov 11, 2014 - 7:26pm
1.Wilson seems to have gotten better than he was last year.
Uh... what? Most people would come to the exact opposite conclusion. Sure, his protection is bad and his receivers are worse than last year, but he's noticeably less accurate downfield (last year he might've missed Kearse on that throw against New York, but not by as much as he did) and more hesitant to pull the trigger (such as on that pass to Lockette near the end of the game), although that could be from not trusting his receivers to run the right routes.
2.The running game is very solid.
A large whack of that is from Harvin and Wilson. Harvin is gone, and most teams don't defend the read-option at "Green Bay against Kaepernick" levels of incompetence like New York did on Sunday.
3.Experienced, well coached and best home field in NFL.
The home field narrative should get retired now. Dallas, Oakland and New York didn't seem to be affected at all, especially on offense.
1.Their OL seems shaky in the non-play action passing game and Seattle isn't the type of team to win a shootout if they aren't getting turnovers.
Hopefully this will improve as Wilson gets more familiarity with his receivers and the OL gets healthier. Also, Seattle's track record under Wilson suggests that A. they're the type of team who can win any kind of game, and B. they aren't the type of team would would end up in a shootout in the first place.
2.Their pass rush has declined from incredibly good in 2013, to just good in 2014
I don't even see it as good. Even when they do apply pressure like as not the QB will escape it, as Rivers and Romo did time and again. Losing Mebane isn't going to help either.
3. They are getting average to below average play by the CBs opposite of Sherman, which is probably getting exposed because of Con #2.
I think the biggest issue is not the CBs but the LBs and the shallow coverage in general. With injuries to Wagner, Smith and Chancellor, their replacements continually allow easy completions to receivers finding holes in the zones. It wasn't exactly a strength even in their historic season last year, but now they're giving up 3rd-down conversions at a shocking 23rd-in-the-league clip, with none of the turnovers that made up for it.
#28 by RickD // Nov 11, 2014 - 7:18pm
Green Bay and Seattle might well be the best two teams in the NFC. They might also end up being the wild cards.
Philly and Arizona both have very good coaching (IMO) but both have major QB issues. It's hard to make the Super Bowl without an elite QB. It can be done, but it's hard.
And then there's Detroit. Great defense, Megatron, a QB I wouldn't trust in the playoffs, and a franchise with a tradition of failing in the playoffs. (To be fair, the tradition is to miss the playoffs.)
I'm far less impressed with the Saints than DVOA is.
#43 by rmonihan // Nov 11, 2014 - 9:56pm
Let's see.
Giants win with Hostetler.
Oakland wins twice with Plunkett.
Trent Dilfer. Brad Johnson. Doug Williams.
McMahon and Rypien had good seasons, less than stellar careers.
We'll see how Flacco plays out, I'm still not convinced he's a "great" QB - he's more a good game manager like Dilfer was. You might toss Russell Wilson in there, someday...but he's new so I'll toss him out for now.
These teams all benefited from great Defense, not a great QB. On the other hand, it's true that's a small subset of Super Bowl winning QBs. 7 out of 49 (1/7th of all SB winning QBs were mediocre or worse).
But this ignores that of the other SB winning QBs, almost all had superior Defenses.
I'd go out on a limb and say that you've a better chance of winning it all if your Defense is exceedingly strong and you've got a meh QB. Sure, having a great QB helps. But then again, Jim Kelly, Warren Moon, Dan Fouts, early John Elway, Fran Tarkenton, Donovan McNabb, Dan Marino, Ken Anderson and quite a few others might say that being a very good to excellent QB doesn't amount to a hill of beans when it comes to winning a Super Bowl.
Hell, Eli won two. While he has crazy, nutty January numbers, his career consists of two solid seasons and a whole bunch of garbage. He won his first Super Bowl at the end of one of his worst seasons, didn't get past game one of the playoffs in his best season.
Meanwhile, his much better brother only managed to win one Super Bowl, while appearing in 3.
#62 by RickD // Nov 11, 2014 - 10:47pm
This isn't the 1980s. Bob Griese won two Super Bowls handing off all the time. But this is a pass-happy era. So either you have to have a great defense and a good enough QB or have your own QB that can light things up himself.
Flacco had a great playoff run on the way to winning the Super Bowl. Go look at the stats if you don't believe me.
Brad Johnson and Trent Dilfer played with great defenses. Does Philly have that? Arizona does, and they might well have home field advantage, but I am still hesitant.
When was the last time a team made the Super Bowl with a backup QB starting in the playoffs? And no, I wouldn't count Brady. I think you'd have to go back to Earl Morrall. Is Mark Sanchez going to lead a team to the Super Bowl? Is Drew Stanton?
"But then again, Jim Kelly, Warren Moon, Dan Fouts, early John Elway, Fran Tarkenton, Donovan McNabb, Dan Marino, Ken Anderson and quite a few others might say that being a very good to excellent QB doesn't amount to a hill of beans when it comes to winning a Super Bowl."
You've got the logic backwards. At no point do I argue that having a great QB is sufficient to make the Super Bowl. I'm arguing, with a slight caveat, that these days it's necessary. And part of that is that there are a dozen or so QBs that can light up a weak defense these days.
#84 by ElJefe // Nov 11, 2014 - 11:24pm
To reply to the "backup QB" part of this question:
Doug Williams was not the season-opening starter at QB for the 1987 Redskins (Jay Schroeder), Jeff Hostetler was not the season-opening starting QB for the 1990 NY Giants (Phil Simms), Trent Dilfer began the season as a back-up for the 2000 Ravens (Tony Banks), Jim Plunkett was a mid-season replacement for Dan Pastorini(!) on the 1980 Raiders. You could even throw Kurt Warner (1999) onto this pile.
All off the top of my head. Now I know how RaiderJoe feels, every day.
Overeducated Layabout
#169 by Grae // Nov 12, 2014 - 3:50pm
"When was the last time a team made the Super Bowl with a backup QB starting in the playoffs?"
Would you count Kurt Warner? He's kind of a strange case, because he was a backup at the start of the season, but due to Trent Green's injury in the pre-season, ended up starting the regular season.
#137 by MilkmanDanimal // Nov 12, 2014 - 11:38am
Time to get on one of my traditional high horses . . .
Brad Johnson was quite good in 2002; he was 10th in DYAR and 11th in DVOA. He was absolutely an asset and, while he wasn't slinging it around like 2013 Peyton Manning, he was a QB who was helping his team win. 22 TDs and 6 INTs is not a detriment in any way, and Brad Johnson was generally a good QB throughout his career (until his late-career dropoff), and he was particularly good in 2002.
Trent Dilfer in 2000 was a giant ball of suck the Baltimore defense dragged to a championship. Brad Johnson in 2002 was very good and contributed to Tampa winning that Super Bowl.
#139 by Will Allen // Nov 12, 2014 - 11:59am
Anybody with a lick of sense understands Brad Johnson had a good career, but in my time machine, Johnson doesn't get hurt early in 1998, and the Vikings pretty much have the same year they had with Cunningham (maybe better, absent Cunningham allowing a strip sack at the end of the first half in the NFCCG) and Johnson's career is seen very differently. Then again, if he gets stuck with the Vikings defense post 1998, who knows what would have happened?
#152 by Sakic // Nov 12, 2014 - 1:39pm
I won't argue that Brad Johnson had a good career but I find it hard to believe that he would've had the same year Randall Cunningham had in 1998. Things are little foggy but I remember thinking that when Cunningham took over the Vikings went from a pretty good team to a really, really, scary team. They already had Cris Carter and Jake Reed but and then throw in Cunningham throwing looping bombs seemingly every other play to the rookie Randy Moss...yikes. I still have nightmares about the week 5 MNF game against the Packers. To me, Cunningham just seemed perfectly suited for that offense.
#153 by Will Allen // Nov 12, 2014 - 1:50pm
It's an interesting question as to how much that jump was due to Johnson being replaced by Cunningham halfway into the 2nd game, and how much was due to Moss now being 5 games into his rookie season, which was when things really took off. I admit having a little anti-Cunningham bias, since I've always been disinclined to like a qb with a big wind-up, but he was super-productive that year, for sure.
#168 by ElJefe // Nov 12, 2014 - 3:45pm
"Cunningham just seemed perfectly suited for that offense"
Flip that around: That offense was perfectly suited for Randall Cunningham.
Randall (with his loooong release) could only excel when throwing patterns where the ball being delivered late wasn't an issue. He could throw outs (the receiver can't run past the point where Randall aims), button-hooks/comebacks (seriously, the staple passing play of the Eagles offense was posting up Keith Jackson on a linebacker, and yes, that is the proper terminology for what they were doing), and "go-deep" because the ball being late only made the play into a jump ball between the WR and a DB. Randy Moss on the deep ball, Cris Carter to outmuscle and outmanuever DBs, Robert Smith and stout OL so you aren't passing on every play and don't have to throw into much heavy coverage; Cunningham and that offense were perfect together.
Overeducated Layabout
#2 by swgallagher // Nov 11, 2014 - 6:24pm
Just curious... if there is a past article that justifies counting special teams performance as 1/3 of a teams overall DVOA?
I certainly understand the value of the field position game, but there are just so few ST snaps in comparison to offense and defense - especially once you consider how many are completely benign...
Fair catch, touch back, PAT...
#19 by Perfundle // Nov 11, 2014 - 6:51pm
Oh, you mean in a purely numerical sense. But I don't see why it shouldn't be that way. After all, you could have a team with 0.0% offensive and defensive DVOA, and then ST DVOA would represent 100% of their overall ratings.
#44 by thebuch // Nov 11, 2014 - 10:00pm
That's an overly simplified way of looking at it.. Because there are far fewer snaps, ST DVOA will not deviate as much as offensive or defensive DVOA. That said, a 0.0% offensive and defensive DVOA IS there rating, that they're exactly league average, and that factors in just as much as being good or bad at offense/defense. Their formula is still offense+defense+special teams, its just average+average+variable instead of variable+variable+variable, and being average says just as much about the team as does any other arbitrary number. I hope that makes sense.
#48 by Perfundle // Nov 11, 2014 - 10:09pm
I was just pointing out that looking at what percentage the special team's portion of DVOA takes up isn't a good way of accessing its correct weight. Baltimore and Philadelphia's special teams' DVOA take up a large percentage partly because they're the two best teams at it, but also because they don't have amazing offensive + defensive DVOA totals. If special teams is being weighed too highly, it would still be too high even if Baltimore had 20% offensive DVOA and -20% defensive DVOA, whereupon the ST percentage would drop to 17.3%.
#50 by Perfundle // Nov 11, 2014 - 10:25pm
I don't think there is an exact 1/3 threshold, is there? I'm guessing it's simply that the best/worst special teams each year just happens to fall around 1/3 of the value of the best/worst offense and defense. After all, if a team returned every single kickoff for a touchdown and forced a fumble that turned into a TD every kickoff, they would have 0% offensive and defensive DVOA and something like 500% special teams DVOA.
#72 by Tim F. // Nov 11, 2014 - 11:08pm
My understanding is that DVOA values for Offense, Defense, and ST all stem from the likelihood of scoring from X position on the field after Y result on the play. The values for ST are relative to O/D; thus, they are always in flux and being recalculated but tend to regress to that 1/3 proportion. I could be wrong though.
#3 by herewegobrowni… // Nov 11, 2014 - 6:28pm
Bengals are only projected to go ~2-5 the rest of the way! Ouch! Fits the growing common wisdom that the AFCN race is going to be a 3-team affair as they fall off, with the Cle @ Bal Week 17 game possibly being for all the marbles.
#12 by intel_chris // Nov 11, 2014 - 6:46pm
So, Denver is back to having a 1 & 2 offense and defense again. Is this the 4th or 5th week in that position? My feeling is that this is still an artifact of lots of mediocre teams, injury luck (the teams beaten by Denver all seemed to be missing key player(s) just at the point of the game), but it is still a curious fact.
#129 by Bernie // Nov 12, 2014 - 10:41am
And this is why you have to take all team ratings (of any sort) with a grain of salt, because does anybody really think the Colts and Pats are exactly even?
I love the Colts, but I can't shake the feeling Brady s going to tear us apart SUnday night. Gronkowski is going to run wild in the middle - the only way for Indy to defend him successfully is to put 3 guys on him, and that'll just make them even worse at defending the swing pass out of the backfield, which Vereen will murder them with.
The Colts only hope to win that game is for Andrew Luck to go uber cyborg and hit on every pass all over the field, and rack up 50 points.
#15 by Sporran // Nov 11, 2014 - 6:48pm
Interesting: PHI, DAL, GB, ARI and SEA all won -- yet saw their chances of making the playoffs decrease. It appears that a lot of weight is being given to the wins by Detroit, San Francisco, and Atlanta(?) -- but still, this seems like a very odd occurrence -- so much so that I'd wonder if the simulation were run correctly (or was just randomly off).
#22 by Perfundle // Nov 11, 2014 - 6:59pm
It's not that all of them should go up, but for all of them to go down is what's surprising. Still, you can't expect entirely common sense results when you run Monte Carlo simulations. It could simply be that last week's results overestimated their true chances given their DVOA ratings and underestimated them this week.
#33 by Perfundle // Nov 11, 2014 - 7:46pm
More important than that is the fact that the five winning teams had DVOA rating advantages of 25.8%, 26.8%, 23.1%, 26.2% and 23.4% going in and four of the teams were at home, which meant that other than Dallas they were all expected to comfortably win their games. Meanwhile, San Francisco was an underdog and Detroit was about even with Miami, so winning those games significantly boosted their expected win total, and Atlanta went up because the teams in front of them both lost.
#35 by RickD // Nov 11, 2014 - 7:52pm
I'd say the real effect we're seeing here is that San Francisco and Detroit both scored DVOA upsets. In prior simulations, both were probably considered likely losers of those games (esp. SF @ NO). The other NFC contenders won games they were expected to win. Resolving those games helps them less compared to prior simulations than the Lions and 49ers are helped.
#18 by Will Allen // Nov 11, 2014 - 6:50pm
Is there any data to suggest that high variance teams through the first half of the season have a greater tendency to improve their weighted DVOA ranking by year's end. That is, does a high variance team through week 8 have a better chance of improvement, compared to a tea with similar DVOA, but markedly lower variance? Or is the opposite true? Or is there no discernable trend?
#91 by intel_chris // Nov 11, 2014 - 11:38pm
I think Aaron did analyze this (at least) once a few years back, but I don't recall what the outcome was. My guess is that variance is randomly distributed and so some high variance teams improve, but others decline netting out to no discernible effect. However, that's just a guess.
#27 by big10freak // Nov 11, 2014 - 7:16pm
Boy is the Detroit defense carrying the load for the team. Subpar offense, dreadful special teams and sporting a 7-2 record thanks the defense and a splash of previous bad luck in close games being reversed this season.
#138 by JoeyHarringtonsPiano // Nov 12, 2014 - 11:46am
Keep in mind DVOA doesn't know Calvin Johnson was out or ineffective for 5 out of 9 games, but your point is a good one. The defense is carrying this team on its back. The entire field goal unit needs to be taken out back and flogged.
#38 by Vandal // Nov 11, 2014 - 7:59pm
Here's the thing about the Patriots DVOA versus Denvers DVOA.
They have:
Identical records,
Identical points Scored
Identical Points allowed.
Virtually identical opponent strength (2% different)
And yet DVOA has New England as 24% worse than Denver, Which is the same difference as between New England the the Jets.
You can say expert power rankings are short sighted, but check out Nate Silvers work. It doesn't suffer from recency bias, and actually has New England as #1.
I think the problem with DVOA, is the linear nature of the rankings. I think your game DVOA scores should show diminishing returns. So that a team that has 1 VERY bad game (-100%), and 5 good games (+20%) aren't being shown as a 0% DVOA team. You can only lose once per game... Anyway, I think that's the main flaw, the lack of diminishing returns...
#40 by Perfundle // Nov 11, 2014 - 8:50pm
Nate Silver's ratings have the opposite of recency bias, which is inertia, and it's an even worse flaw when rankings teams. Cleveland could win out and they probably still wouldn't break the top ten, for instance. ELO ratings are useful in chess where players' talent levels hardly change, but it's next to useless when evaluating NFL teams that see major upheavals in talent. The majority of New England's #1 ranking came from the fact that they started the season ranked 3rd; they only dropped to 7th even after getting blown out by Kansas City and falling to 2-2.
As for your last paragraph, the lack of diminishing returns wouldn't explain the current difference in DVOA, because even if New England had a very bad game, the fact that they have the same point differential should mean that they should have had better games on average than Denver in the other 8 games. The disparity between simple stats and DVOA suggest that New England has been poorer in certain aspects of the game that simple stats don't show up, and also that the 2.5% difference in opponent strengths is bigger than you make it out to be. Besides, it's only 2.5% because New England played the best DVOA team in Denver while Denver played a merely good team in New England. If you had it your way and Denver and England had equal DVOA ratings, the difference would shoot up to 5.2%.
#41 by Tim F. // Nov 11, 2014 - 8:55pm
No idea what you think you mean by "diminishing returns" but you seem to be trying to assert your own intended result on the value of DVOA. (538's ELO attempts to measure a team's predicted success of winning for an entire season; FO's DVOA attempts to measure which teams is most successful on all plays NOW — the two will likely grow closer as the season progresses. Anyone looking at DVOA should understand its intent, what is being measured, consider opponent adjustments, the likelihood for change, variance, etc.) That YOU want NE to be #1 NOW is not a good reason to alter DVOA.
#42 by RickD // Nov 11, 2014 - 9:50pm
I don't think it's entirely legitimate to make this all about what Vandal "wants". The points he raises are noteworthy. And it's not just that Denver is #1 while NE isn't - it's that based on the most obvious stats, they appear to be equal, yet NE is way, way behind according to DVOA.
Meanwhile, Baltimore hasn't beaten a winning team (by current record) since Week 3, have a worse record, a worse score differential, and have played a weaker schedule than New England. But they seem to be locked in at #2. It is a curious thing. They even were on the losing side of a major ass-kicking by Pittsburgh last week (and Pittsburgh is rated lower than Kansas City), but they cannot seem to leave the penthouse.
DVOA has its own inertia. It's not something I completely understand, but we've known that for years. There's always a team or two that's inexplicably high, or inexplicably low. DVOA isn't perfect.
Nor is 538's ELO system. Both are just statistical systems, and by their very nature, statistics entail compressing information, and there will always be information loss in the process.
FWIW, I don't have an issue with Denver's rating. And I can live with NE's rating. Baltimore, OTOH, seems way too high.
#46 by Perfundle // Nov 11, 2014 - 10:03pm
Yeah, Baltimore's ranking makes no sense to me either. But regarding inertia, at least DVOA's is based on data from the current season, instead of carrying over data from previous seasons. Right now ELO is basically the equivalent of giving preseason DVOA 70% weight, if not more.
#51 by Tim F. // Nov 11, 2014 - 10:26pm
"And it's not just that Denver is #1 while NE isn't - it's that based on the most obvious stats, they appear to be equal, yet NE is way, way behind according to DVOA."
It's unclear to me how this is remotely surprising or problematic. DVOA has never correlated to raw W-L, yardage, points stats provided by the NFL. I don't know why anyone would look at the NFL basic stats as a good comparison between teams. I don't think many would even consider Denver and NE "equal" by trusting their own eyes of what they've seen thus far, even if the head-to-head game caused them to rethink how a rematch would go— they may have an opinion about who is better head-to-head but I doubt many would say that the result of any of their games would be exactly the same if you substituted one for the other.
Basically, it seems some either want DVOA changed in defiance of its fundamental principles and lessons learned (why would DVOA discount high variance wins/losses when it's attempting to measure success on every play because the data reveals that teams that are more successful on every play rather than boom/bust are the more successful teams?) or they want any advanced analytics to reflect the same result, or they want a perfect Power Ranking and/or Super Bowl Winner predictor. All of which seem quite silly to me. I'd prefer every type of analysis exist on its own merits, improve its own efficiency and accuracy, and that their results self-reveal what they are good at measuring and how "predictive" those results are. In that regard, I find DVOA (and all of the associated data provided by FO) far more telling that 538's ELO, despite the fact that DVOA has only predicted the Super Bowl winner a handful of times over the last 15 years (although it was highly successful the decade prior). So, yeah, I might have been pretty glib and dismissive of Vandal, but I'm not troubled or even puzzled by the current DVOA values for Denver, NE, Baltimore, Miami, or any other team at the moment.
#68 by RickD // Nov 11, 2014 - 11:00pm
"It's unclear to me how this is remotely surprising or problematic."
Surprising? No. I know that DVOA behaves this way.
"DVOA has never correlated to raw W-L, yardage, points stats provided by the NFL."
That's certainly not true. If DVOA didn't correlate to any of those things, this website would not get any traffic. Who's interested in a statistic that doesn't correlate to W-L?
...
"I don't think many would even consider Denver and NE "equal" by trusting their own eyes of what they've seen thus far,"
I think most NFL fans would consider them to be roughly "equal" about now. Not that FO's system need follow the popular opinion. But let's be honest about what the popular opinion is.
"even if the head-to-head game caused them to rethink how a rematch would go— they may have an opinion about who is better head-to-head but I doubt many would say that the result of any of their games would be exactly the same if you substituted one for the other."
Well, no. I agree that one cannot simply plug and play one football team for another. But you go a bit far with how much you are willing to dismiss all other evidence in favor of DVOA.
"Basically, it seems some either want DVOA changed in defiance of its fundamental principles"
Well, I don't need it to be changed, so I'll leave that point to Vandal.
#117 by Tim F. // Nov 12, 2014 - 6:01am
"Who's interested in a statistic that doesn't correlate to W-L?"
Ummm, FO? Isn't the primary goal of FO to find hidden insight that's not revealed in the conventional wisdom? I've never perceived their mission as the same as those guys who will sell you their game picks.
"But let's be honest about what the popular opinion is."
I would say the popular wisdom is that Denver is heavily favored to make it to the Super Bowl as the AFC favorite even if Peyton can't carry them and the whole team can't prepare for and overcome New England. That, even though NE is coming on strong and is very dangerous, they can be exploited defensively, don't have much of a run game, have a weak run defense, and can be beat by not only the few best teams but by some mid-tier teams that match up with them well. Whereas, only NE and two or three other of the best teams are likely to beat Denver.
"But you go a bit far with how much you are willing to dismiss all other evidence in favor of DVOA."
Actually, most advanced stats are closer to agreeing with DVOA on the rankings of Denver and NE. It would be more appropriate to say that I'm only being dismissive of ELO, but even in that regard, I don't think I'm dismissing it, merely saying that they are trying to measure two very different things.
#125 by RickD // Nov 12, 2014 - 10:24am
"Who's interested in a statistic that doesn't correlate to W-L?"
Ummm, FO? Isn't the primary goal of FO to find hidden insight that's not revealed in the conventional wisdom?
Do you even know what "correlate" means?
I can assure you nobody at FO wants DVOA to show zero correlation with W-L.
"Correlate" doesn't mean "correspond to".
I would say the popular wisdom is that Denver is heavily favored to make it to the Super Bowl as the AFC favorite even if Peyton can't carry them and the whole team can't prepare for and overcome New England.
Well that point is easily debunked. I'm sorry, but since a week ago Sunday, that's not the popular wisdom by a long shot. The Patriots just beat the Broncos, handily. The public doesn't care that it happened Foxborough. The public rates the Patriots ahead of the Broncos now. FO may not do so, and Vegas may not do so, and you are free to have the opinion that you have, as expressed above. But that's not the popular wisdom. Take a look at the power rankings poll at ESPN.com. The Patriots have more than twice as many 1st place votes as the Broncos. The Broncos are certainly not "heavily favored" by the public.
#174 by Tim F. // Nov 12, 2014 - 5:39pm
"I can assure you nobody at FO wants DVOA to show zero correlation with W-L."
Where did I say that FO wants DVOA to show zero correlation with W-L? I may have been unclear to you, but let's not twist the context to my statement to utter absurdity. Of course, FO wants picks and playoff predictions to have some level of accuracy, but I am saying that FO very likely doesn't want to compromise the current value of DVOA to try to be a more accurate game predictor than a completely different stat designed to be, and successful at being, a more accurate game predictor. The basic NFL turnover differential and probably a few other basic NFL stats are likely (but I honestly haven't looked) more correlative to W-L than DVOA. If DVOA wanted to be the best predictor to W-L record, they're certainly putting a lot of misplaced effort in all the wrong places.
"The public rates the Patriots ahead of the Broncos now. FO may not do so, and Vegas may not do so, and you are free to have the opinion that you have, as expressed above. But that's not the popular wisdom. Take a look at the power rankings poll at ESPN.com."
Lol. Online ESPN poll. You're right, no one considers Broncos the current #1 in the AFC. LOFinL. Irrefutable, ironclad evidence.
#73 by Vandal // Nov 11, 2014 - 11:09pm
You should be puzzled:
http://nflpickwatch.com/
538 is 5th of 130
FO, if it was listed, would 80th.
You can glib and dismissive all you want, but it order to be condescending, you need to speak from a position of superiority. Try to get there first.
If the advanced stats can't predict outcomes with any increased accuracy, they aren't worth a damn. Moneyball is a legend and movie, is because the advanced stats found a market inefficiency and exploited it.
If your advanced stats perform worse than a random pleb, then it's just a sabermetric circle-jerk.
#80 by Tim F. // Nov 11, 2014 - 11:19pm
Nonsense. You want advanced analytics to function as game predictors. I think if that is your intention then you aren't paying attention to what the analytics are actually measuring and telling us. I've been reading FO since it started, never valued it based on predictive results. I actually view it primarily from the opposite perspective of: what the game results and basic NFL stats do not tell me, what my own eyes and biases may not be telling me.
All advanced stats always perform worse than a random pleb as a game predictor, just a question of finding the right random pleb.
#85 by chemical burn // Nov 11, 2014 - 11:27pm
The problem is that fo doesn't sell it that way. Just because you recognize that their system has always been an uncomfortable mix of descriptor and predictor and have dealt with that fact doesn't mean they present it as only descriptor. In fact, they constantly emphasize components determining dvoa's predictive value.
#92 by Tim F. // Nov 11, 2014 - 11:40pm
I think you are overselling what you perceive are FO's claims for predictive value. But maybe you are right and I am wrong. I will leave it to the staff to provide their own answer. Of course, you want the data to be insightful, but I think the staff is far more cognizant of the odds changing week to week, understanding why a result may seem surprising but ultimately isn't and may change, and AGS occurring than some are claiming is a failing of past and future analysis by FO based on sports book results.
#95 by chemical burn // Nov 11, 2014 - 11:55pm
FO explicitly, in no uncertain terms states that certain factors influence DVOA for their predictive value. Their most notable insight in this regard has been fumble luck. It's not when slightly up for debate. You have chosen to ignore this aspect if DVOA'S intended making likely because it is so poor.
#99 by Tim F. // Nov 12, 2014 - 12:09am
I'm having difficulty parsing the grammar of your statement so I'm probably not properly following. I do understand FO's exploration of fumble luck however, but I'm not seeing how FO's analysis of fumble luck determines that FO thinks DVOA's primary and distinguishing value is as a game predictor, particularly more so than other advanced analytics which are clearly attempting to measure completely different things. I don't pay for advanced analytics though, so maybe there is a whole world of arrogance that I'm missing. ;-)
#102 by greybeard // Nov 12, 2014 - 12:21am
FO claims its predictive. They even release their predictions before the season begins. They also have not revisited how good their predictions are the last five years because they have been on an unlucky roll the last five years. As far as I can tell DVOA predictions correlate well only with DVOA. So they perfected a system where it can predict where it will be.
#103 by intel_chris // Nov 12, 2014 - 12:48am
I think you have to look at DVOA in perspective. As I recall it was originally invented by Aaron (a Patriots fan) to illustrate that one of the early teams was better than "conventional wisdom" suggested. In particular, that team won the "field position game" and slowly worked its way into scoring position and worked its opponent out of scoring position. Thus, DVOA tends to emphasize plays that move the chains more than it does ones that "score".
Winning the "field position game" can result in winning the football game, and against average and weak opponents often does. It does not always win the football game.
However, if you just look at DVOA as a metric, one that emphasizes that particular aspect, it can be a useful measurement, and all things considered one would generally rather win the field position game rather than lose it.
Note I think this actually explains to a certain extent why Denver's DVOA was high but they lost to the Patriots. Over the last few years, Denver has built a team that is particularly well suited to beating inferior (or just average) opponents. DVOA loves these kind of teams. Denver doesn't do as well against teams that are "elite" or "lucky". This years Patriots game, last year's SB, the previous year's playoff against Baltimore shows what can happen when an "elite" team plays Denver and gets a break. Denver's offense and defense are best suited to getting ahead and staying ahead until the other team collapses, but if the other team can get sufficiently ahead (or even just stay even long enough), Denver can be beat, even trounced. This shows that a great DVOA does not guarantee a win even over a lower DVOA team, especially not in "clutch" situations. It isn't designed to measure that. It's designed to measure how well a team grinds other teams into the ground by consistently moving the chains.
#104 by Tim F. // Nov 12, 2014 - 12:50am
You and I clearly have different weights for the self-ascribed predictive value that FO places on itself. Clearly, their preseason predictions are not based on the predictive value of this season's DVOA values (although since I don't look to them for predictive value, I don't know what the basis for those predictions are — but clearly their preseason prediction are not a statement about Week 8 DVOA predictive value). I have a reasonably strong understanding of FO's inputs into DVOA... why they don't measure some things which may be measurable because they aren't statistically predictive even if they could be determinative, what factors they don't measure even if they are predictive because they are subjective or difficult to record in a consistent matter or because they actually negate a greater predictive value.... All that said, if you are relying on and expecting DVOA to be predictive of game results week to week and throughout the playoffs and think that is something FO asserts it can deliver through DVOA because FO talks about predictive value in their statistical analysis, I think you are in for a world of hurt. (Again, I don't pay for Premium and I can't imagine using the small handful of DVOA outputs as a game predictor tool.)
But that's my viewpoint, certainly the staff can speak for themselves better. Maybe they should be addressing it for those who do pay for premium and think that they aren't getting their value's worth.
#109 by greybeard // Nov 12, 2014 - 2:50am
I have been following this site since 96. I am pretty familiar with what both they put into the DVOA and their stand on the predictive nature of it.
I don't put stock in DVOA. I think it is very flawed metric and has been proven to be not predictive. I also don't find it to be very descriptive either. It seems to me that you care and value it a lot, which suggests to me based on the history of the last 5 years you are in a world of hurt not I.
#118 by Tim F. // Nov 12, 2014 - 6:17am
Nope, unsure why you are seeing so much care on my part. I rate FO's stats about equal or slightly lesser than most advanced analytics. And I still trust my own judgment over all of the analytics that I try to ingest. But I still see value in it. Basically, I'm trying to do the calculus on how my own perceptions and various analytics differ to find some truth... or not. And I still come up short but since I'm only invested as a consumer and fan, not as someone who can have any direct effect on the outcomes or any direct benefit except entertainment and enjoyment (which does include the occasional wager but with the understanding that it is a gamble and entertainment), I don't much care.
I'd think it's the people that are angry that DVOA is not predicting game winners are the ones in a world of hurt.
#136 by intel_chris // Nov 12, 2014 - 11:30am
Hmmm, FO has only been around since 2003. I'll take your word for it. I would have guessed 2001, as I started a football drought around 2003 or 4 and have followed this site for some time. As to a functional web, depends upon your definition. I've been using some form of the web since about 1977, prior to that most computers were not connected, so it was site by site. It was just a lot harder to get to back then, being near a major university and/or working in the computer industry really helped.
#105 by Perfundle // Nov 12, 2014 - 1:02am
Here is Silver's own admission on his initial writeup for ELO:
http://fivethirtyeight.com/datalab/introducing-nfl-elo-ratings/
Can I use you to beat Vegas?
I wouldn’t try that. Vegas lines account for a much wider array of information than I do. When Nate backtested me, he found that I got 51 percent of games right against the point spread. That’s not nearly enough to cover the house’s cut, much less to make a living.
ELO is not supposed to be predictive, and the fact that his picks are so accurate this year is largely because teams are abnormally static this year: only 2 teams have records that differ by at least .35, compared to the average of 3.75 teams the previous four years, and only 5 teams went from a losing record to a winning one or vice versa, compared to the average of 8.5 teams the previous four years.
#116 by Perfundle // Nov 12, 2014 - 5:41am
Did you know that using last year's point differentials to make straight-up picks, without even accounting for home-field advantage (e.g. Detroit had a point differential of 19 last year, compared to -11 for Green Bay, so pick Detroit), would've gotten you a 104-42 record, which would've been third-best on that site, and incidentally bested Silver's record? That's how freaking static this season has been compared to last, and it's why his inertia-heavy ELO ratings are doing so well; it has nothing to do with diminishing returns. You need to stop putting so much credence into half a season's worth of picks in an unusual season. It's clear that his system would've been horrible picking straight up last year.
#47 by Vandal // Nov 11, 2014 - 10:09pm
I don't "want" anything, other than improvements to a system which seems to be getting worse, rather than better (FO's winning percentage in picks, is basically the poorest performing system this season, and has been getting worse for about 4 years now).
I picked New England and Denver, because I was on NFL.com today, and noticed they are both at +84 point def this season, so it's an easy example to illustrate my point.
I know ELO is flawed, absolutely, but I believe there is merit to diminishing returns as a part of the system. My main reason for this? Silver thinks it's important, and that ALONE makes it worth consideration. Also, saying ELO is good for chess, but not for football, is disingenuous. Chess ELO is linear, Football ELO is a modified system that uses point margin in it's ELO results (subject to DR).
Anyway, I don't care which teams are high, and which teams are low, I care that the quality of the information is trending in the wrong direction, and the rationalization for this, was "New England had one REALLY bad game, and that outweighs how many good games they had". And this flaw could be erased, by implementing DR.
Better yet: Test Diminishing returns on game scores, run the regression analysis, eliminate it if it hurts the system, implement if it helps. Scientific method... shouldn't be a tough sell.
#70 by intel_chris // Nov 11, 2014 - 11:02pm
Thanks, Will. In fact, the test has been done. DVOA has a diminishing returns factor. It's called "weighted DVOA". It just kicks in slowly enough that it is only now having an effect. The reason it kicks in slowly, is precisely because it *was* tested and the most consistent wDVOA results came from the weighting factors that are now used. Back a few years ago when Aaron was measuring the factors, he tried numerous experiments including a system that looped around re-weighting the games until the system "converged". It was more poorly predictive of the final wDVOA than the current "simple" diminishing returns system that only begins discounting games after they are about 8 games back.
You can call that "inertia" if you want. I think more it has to do with teams having a fundamental "character" but which needs more than one game to discover due to the random nature of football. In fact, I suspect it takes something on the order of 8 games to tease out that character (although you can approximate it much earlier), because otherwise teams that might have played a very easy or tough (for them) stretch of the schedule can look too good or bad if you don't factor in enough games. (SD is a good example of that. They looked like a playoff lock after the 1st few games, but are slowly falling.)
After 8 games, the teams finally begin to settle down into predictable patterns, at least at the ends, where a top 10 team will not likely fall to a bottom 10 team by the end of the season, barring some catastrophic event, like losing a franchise player.
And even with all that in account, the random nature of football is sufficient to make any one game unpredictable. For example, although many of us expect NE's DVOA to rise as the season progresses, I don't think it will rise enough to predict the pasting that NE gave Denver. It might rise enough to predict a NE victory, but not likely a blowout victory.
#56 by Perfundle // Nov 11, 2014 - 10:37pm
Also, saying ELO is good for chess, but not for football, is disingenuous. Chess ELO is linear, Football ELO is a modified system that uses point margin in it's ELO results (subject to DR).
That doesn't matter. Silver's ELO still maintains an enormous whack of preseason ranking that makes no sense for ranking teams with such fluid parts. The worst-place team starting the year winning every single game by 10 points wouldn't have them anywhere near #1.
Anyway, I don't care which teams are high, and which teams are low, I care that the quality of the information is trending in the wrong direction, and the rationalization for this, was "New England had one REALLY bad game, and that outweighs how many good games they had". And this flaw could be erased, by implementing DR.
Again, you've failed to explain why not implementing diminishing returns is the reason why New England is ranked so low. Any overemphasis on the blowout loss by Kansas City should've been more than cancelled out by a similar overemphasis on the blowout wins against Minnesota, Cincinnati, Chicago and Denver.
#57 by Vandal // Nov 11, 2014 - 10:39pm
Did you even read this article?
"The issue here is their performance early in the season, but more specifically, one game weighing them down: the Week 4 blowout by Kansas City. "
Why do you think I used the New England example. Jesus.
#65 by Perfundle // Nov 11, 2014 - 10:48pm
Settle down there. Given that you're comparing point differential, it sounds like what you're actually arguing is not that New England is being overly weighed down by Kansas City but that they're not being boosted enough by the blowout wins. After all, if they had 72.7% for each of their 20-point wins they wouldn't be ranked so low.
#75 by Perfundle // Nov 11, 2014 - 11:11pm
Spare me the condescension; I know exactly what diminishing returns mean, whereas you don't understand my point. I'm asking you why you're only directing your complaint about why the Chiefs game is not being diminished, when going by point differential the other four games should also be very positive in DVOA. And even if diminishing returns were implemented, it wouldn't nearly accomplish what you hope it would, considering New England's own win against Denver would be similarly diminished.
#89 by intel_chris // Nov 11, 2014 - 11:32pm
Ok, I read the Wiki article and I will re-answer your question in that light.
Diminishing returns is the decrease in the marginal (incremental) output of a production process as the amount of a single factor of production is incrementally increased....
DVOA actual explicitly incorporates that kind of diminishing returns, in that it caps values of long runs (and other plays) whose per game value is not representative of their predictive power going forward. We've actually had this discussion many times. DVOA explicitly incorporates diminishing returns on a play-by-play basis. It instead rewards teams that consistently run "successful" plays. Doing so it tends to reward "dink-and-dunk" type teams that win the "field position game". That is DVOA likes a team that makes enough yards each play that it tends to slowly grind its opponent into the ground. It was designed to measure exactly that effect.
Plays that may win the game but are "rare" enough to be non-predictive tend to be discounted. They aren't completely thrown out, but they are rewarded less than making consistent yardage.
Denver and Arizona this year are two good examples of the dichotomy. Denver doesn't generally have many back breaking plays. They just move the chains and score nearly each and every possession. That gives them the #1 offensive DVOA. Arizona in contrast plays the "boom or bust" game, throwing the strike down the field every-so-often. Arizona connects on those strikes just often enough to beat their opponents, but DVOA doesn't reward those "rare" events by as much as they win the games. Thus, Arizona has a much lower offensive DVOA (25th), even though they have a better record than Denver.
I can't tell you whether there are other levels of "diminishing returns" calculations beyond the per-play one. I know that the metric attempts to compare each play to "similar" plays by situation. That might be a "diminish return" type effect, e.g. once a team gets far ahead they often play differently (e.g. GB resting their starters v. Chicago) and that might be one of the factors that is counted in. I don't think Aaron has ever said exactly what makes plays similar, although I do recall a posting that once showed that field position was a factor, and if I recall correctly was a linear one, based upon the regressions he had run.
#69 by Perfundle // Nov 11, 2014 - 11:00pm
Since you're negatively comparing FO's ranking to other systems, I'll note that every other objective ranking system that I know of other than Silver's (FO, Sagarin, Advanced Football Analytics, Pro Football Reference, Massey-Peabody) has Denver firmly ahead of New England, so I guess every system needs to take into account diminishing returns? Or maybe Denver is simply better than New England this year.
#81 by Vandal // Nov 11, 2014 - 11:22pm
My point, my ONLY point, is that I think DVOA could have improved functionality, by implementing individual game diminishing returns.
(I know it's more complicated than this example, I'm just TRYING to get across diminishing return)
For Example:
In simple DVOA, a team plays 5 games. The game scores of those games are: -100%, 25%,25%, 25%,25%.
Simple DVOA scores this team as a 0% DVOA, with an expected win of 2.5
In a Diminishing Returns system, where the weighting is halved every 25% away from the baseline, that same team would have a game scores of: -47%, 25%, 25%, 25%, 25%.
For a DVOA of around 10%. Much more indicative of a 4-1 or 3-2 team.
Anyway, WIKI can explain Diminishing returns much better than I, hopefully you get the idea.
#94 by Perfundle // Nov 11, 2014 - 11:54pm
Again, you're not telling me anything I don't know. Are that wasn't your initial point, at least not with the evidence you provided.
First, chopping that -72.7% to, say, -30% would still put New England at around 18% in total DVOA, still nowhere near Denver.
Second, point differential with no other context isn't a good way to compare teams, because of differences in strength of schedule.
Third, you used DVOA itself to show that the two teams played roughly the same schedule. If DVOA is flawed as you claim, then an SOS calculated from it is just as flawed. In fact, DVOA's SOS suffers far more from not implementing diminishing returns than DVOA itself! Around this time last year, for instance, it would've said that playing San Francisco, New Orleans, Green Bay, New England, Cincinnati, Seattle, Oakland and Jacksonville would've been an average schedule, because Oakland and especially Jacksonville completely dragged down the numbers. Other SOS numbers show that Denver has played a considerably harder schedule than New England; Sagarin has Denver 5th and New England 21st, and Denver's simple SOS is .549 to New England's .457.
#98 by chemical burn // Nov 11, 2014 - 11:59pm
FO's sos number are horrible to the point of being worthless. It rates a schedule versus 4 teams with the dvia rankings of #1, #23, #24 & #25 as the same as playing against the #12, #13 ,#14 and #15 teams. That's clearly ludicrous.
#196 by Pat // Nov 14, 2014 - 12:44pm
How?
An average team which faces 8 super-mega-awesome teams and 8 utterly beyond pathetic teams would have a 8-8 record. An average team which faces 16 average teams would have an 8-8 winning record.
Are those two schedules identical? No - a *good* team facing the first schedule would have close to an 8-8 winning record, and a good team facing the second schedule would have close to a 16-0 winning record.
Repeat above statement. You can't boil down 16 random games to one number and have it mean something useful for all schedules and all teams. It's just not possible. All measures of strength-of-schedule are going to be limited.
#198 by Pat // Nov 14, 2014 - 1:54pm
Totally agree. Honestly, strength of schedule should get its own page - that would be one thing that would really, really make FO's stats more accessible to people.
You could imagine a chart showing any or all of the following:
1: Expected record for this team, against an average schedule
2: Difference between actual wins and expected wins against an average schedule (This is what I would order the chart by).
3: Expected record for an average team, against this schedule
4: Expected points scored for this team, against an average schedule
5: Expected points allowed for this team, against an average schedule
etc.
Wouldn't be too terribly hard to calculate each, although the last two (points scored/points allowed) you'd have to look at how to incorporate the baseline for each year a little closely.
It's funny that the previous poster suggested "expected record for an average team against this schedule" because I think that's the least interesting metric you could choose - you can't relate that to a boost that team got from its schedule, because that team isn't average.
The one that I think most people intuitively would relate to is "expected record for this team against an average schedule" - looking at the difference between that and W/L is the measure of how many wins or losses the schedule 'cost' that team.
(That being said, treating an average schedule as 16 games against a 0% team is a little doofy, since that schedule never exists. Another thing you could do would be to do a bunch of random draws against the *other* teams in the league, and average those wins. See how I said that there's no one perfect way to do this?)
#77 by Tim F. // Nov 11, 2014 - 11:11pm
FO picks? Do you mean Mike's previews at Bleacher Report? Does he even use DVOA to pick a winner? (Does it even seem responsible to use DVOA as a game predictor? I'm unsure.) I've never liked his picks and they've never seemed strongly based on what FO's data says. It just seems like he sprinkles some data gleamed from FO into his picks to justify his own feelings.
I don't see why an equal point differential points to equal quality team so it's not doing a very good job of illustrating your point, no.
Of course, variance is significant. I think you are discounting what FO does to account for it.
Just because this article states, NE is largely being dragged down by one game, it seems to me you are discounting much of the description of perceived differences in Power Rankings and DVOA rankings. The Dolphins and Raiders game are near average value now but the Vikings, Bills, Jets, and even the strong wins against the Broncos and Bears don't really show a team that is consistently successful against their opponent while Denver has been more consistently better than average against their opponents.
You claim you don't want anything, but it's clear that you think NE should be near the top, not just higher, in DVOA rating. That may be true later in the season and that may even be reflected in DVOA, but I see no reason to think DVOA should reflect NE as one of the strongest teams in the leagues at this point (based on performance), even if they happen to be one of the most dangerous (potential future performance). The FO staff does think they are performing better now and will climb up the DVOA in coming weeks. Understanding the data informs why they aren't where you might expect them to be.
#86 by Vandal // Nov 11, 2014 - 11:28pm
You read as thoroughly as Perfundle.
I chose New England, and use the one game example vs Kansas City, BECAUSE AARON USES IT IN THE ARTICLE.
Christ, you'd think if you have time to write 3 paragraphs in response, you'd at least have time to read the article I was referring to in the first place.
#100 by Tim F. // Nov 12, 2014 - 12:18am
You are making a distinction that is irrelevant to me. I know Aaron used it. What does that change for you, because it changes absolutely nothing for me? (I already mentioned other examples myself in reply to other comments — Baltimore and Miami.) If you aren't concerned with NE specifically and can demonstrate your general concern — go ahead and argue another team, surely the flaw effects others besides NE. It's irrelevant to me whether it's NE or someone else — I am arguing the value that DVOA assigns to one team (NE) versus what you seem to stridently claim is wrong according to your own views or 538's ELO (but you haven't presented or argued that for any team but NE).
But you sound awfully butt hurt in your last several posts so I guess I'll leave it at that.
#107 by Vandal // Nov 12, 2014 - 1:43am
Let me speak slowly, and I'll even refrain from using team names, and maybe this will sink in.
Here is my One, and ONLY point:
Premise:
DVOA's prognostication, might be improved, by using diminishing returns for individual game scores.
Reasoning:
Some other computer models, created by some of the best statisticians in the world, use this, and so far, they are significantly more accurate.
This is not butt hurt, son, it's frustration at having to refute arguments that were never made; because someone lacking reading comprehension imagined I was pimping a specific team.
#110 by Perfundle // Nov 12, 2014 - 3:02am
When you say some, you in fact mean just one, right? Because you've been endlessly pimping Nate Silver in every single one of your posts. Why do you assume that diminishing returns is the reason that DVOA is inaccurate as opposed to the heaps and heaps of other differences between the two systems? Sagarin's ratings also use diminishing returns and Denver is firmly ahead of New England there too.
#144 by Perfundle // Nov 12, 2014 - 12:25pm
I'm quite familiar with all the predictive models out there, and I haven't seen any other model that's both accurate at making straight-up predictions and explicitly takes into account diminishing returns. Nor has Vandal mentioned any other model other than Silver's, and as I have noted, no other model has New England first either. I don't see how the only conclusion isn't that he was enamored by how great Silver's model has been performing, read through its methodology, and saw that diminishing returns was one of the criteria used and latched onto it.
#145 by Will Allen // Nov 12, 2014 - 12:39pm
Has any predictive model, with its picks available to the public, held up as being superior for 4 or 5 years? I really haven't paid attention to this stuff, because I'm so dubious that anybody is going to do well, with a large enough sample size, to be meaningful, and if somebody actually thinks they are going to accomplish it, they aren't going to let me see it.
#149 by Perfundle // Nov 12, 2014 - 1:14pm
I would disagree with the second part of the claim, because I've read through quite a few sports betting forums, and there's no end of threads made by bettors eager to share their picks (with plenty claiming to have some predictive model aiding them). Posters with good track records across several years have the largest followings, and the threads are filled with appreciative comments made by bettors making the same picks, so it seems that wanting your ego stroked wins out over wanting to keep a winning system to yourself.
#119 by Tim F. // Nov 12, 2014 - 6:45am
I'm completely baffled by what you think I am confused by and not following (although, yes, I presume I could be confused to some extent and apologize in advance — but admittedly, partly, this is due to some real confusion on your part). And you are behaving like a petulant juvenile.
Here are my points:
1a. You are misusing "diminishing returns" entirely but I still understand you. "Diminishing returns" is an economics term that means that as you increase production, you see less and less return from the corresponding output, thus diminishing returns. The apparent corollary to football analytics would be victories (or defeats) of greater margin would count for less.
1b. Elo does have a function that corresponds to this to some measure. Its primary input is point differential per game, and there is a margin of victory adjustment (initially, bigger wins count for more points but only up to a point). Thus, the larger the point differential for games won by higher ranked teams over lesser teams, the less the point differential adds to the Elo score. Likewise, a larger point differential at the hands of a higher ranked opponent over a lesser ranked team counts for less of a score reduction for the lower ranked team. BUT, conversely, if a lesser team wins over a higher ranked team, the greater the point differential is, the more it adds to the cumulative Elo score. (Thus, despite some sense of correlation, using "diminishing returns" to mean something which can also mean the complete opposite is nonsensical. This is a margin-of-victory multiplier with an additional favored-unfavored multiplier directly related to their previous rank. This adjustment is more closely linked to opponent adjustment and success rate than the type of adjustment you want to make.)
1c. But the corresponding adjustment you appear to want in DVOA is not the same, could break DVOA, and/or already exists. DVOA isn't a game score; it is a success rate above or below an average opponent on a play-by-play basis. I'm unclear if you want to adjust past performances or highly variant performances or both. That is: would you adjust these series of game DVOAs the same? How differently?: 1) -100, 25, 25, 25, 25; 2) 25, 25, 25, 25, -100; 3) 25, 25, -100, 25, 25; 4) 50, 50, -50, -50, 0; 5) -50, 0, 0, 0, 50? Weighted-DVOA attempts to adjust for recency, where DVOA is trending. (I'd argue that variance you want to capture, not try to factor out, and I'll explain more below.)
1d. As far as I am aware, but I'd happily be corrected, Elo would not aim to nor does it adjust for highly variant performances or recency by comparing current and past results. It adjusts for high variance, trending, and autocorrelation almost solely by being a cumulative score based on previous Elo score and the margin of victory/opponent adjustment. Elo didn't have NE move up their ratings faster than they've moved up DVOA because of a "diminishing returns" adjustment that devalued early losses; they stayed on top because of a high preseason ranking based on last year and cumulative high margin scores recently (despite the margin of victory adjustment and early losses).
2. Both systems aim to be predictive at the beginning of the season, at the end of the season, want to quickly adjust to changes in data throughout the season, while avoiding autocorrelation (see being predictive at the beginning of the season). However, both are not measuring the same thing.
3. Elo attempts to rank each team and predict the margin of victory/loss against every other opposing team at each moment in the season. It's a handicapping and pairing system. Of course, as opposed to how I will describe DVOA below, it is going to be more predictive of game results than DVOA.
4. DVOA attempts to measure the most/least successful team on a play-by-play basis based on idealized (disregarding random or non-predictive data, weighting for success rate, etc) and controllable circumstances over what would be statistically average. The value of DVOA isn't directly tied to predicting game results but in evaluating the quality of a team regardless of the randomness of football or the actual game result or common wisdom. The very nature of a play-by-play metric is based on the tenant that every play has value and the teams that are most successful on any given play are the most successful, even if reality doesn't bare that out in game results.
4a. Thus, intra-game and inter-game variance is significant and should be measured. Intra-game variance will mostly be seen in average to low ratings; inter-game variance can be tracked in variance and weighted-DVOA. (I included 4a solely because I mentioned above that I would explain why you don't want to adjust variance out of the data.)
5. If you don't believe all or any of the above but do think the sole value of any and all analytics is solely to predict game results and know Elo to be the most accurate system, there is absolutely no need for DVOA to change. You simply need to disregard DVOA and use Elo.
6. If you believe all or any of the above and think that each advanced analysis has its unique value, it's not clear at all that DVOA needs to change or is currently too flawed for its purposes.
7. Considering the above, specifically in relation to comparing Denver and New England, I am entirely unswayed that DVOA is problematically flawed and in need of a fix, particularly, in relation to Elo. Or even in comparison to any other metric measuring something else, regardless of their success rate at game result prediction.
8. I believe I would be similarly unswayed if you made any other specific argument in relation to any other team or teams but am open to the possibility. Of course, DVOA aiming to measure the hidden, immeasurable value of a team on any given play means that I'm highly unlikely to be swayed by a conventional metric or another advanced metric based on an entirely different methodology.
9. DVOA is not without its flaws for its purposes, but the majority of them are statistically negated or are actually irrelevant when you remain aware of what it is and is not measuring.
10. It is possible that DVOA isn't adjusting as quickly as it could be, but I trust that simulations have been run that show that the adjustments that are available to be made to DVOA lead to less correlative results.
11. And... I'm not convinced that DVOA and Elo will not be more correlative with one another at the end of the season or further into the season when playoff situations are more relevant. In fact, I'm almost certain that will be the case.
#147 by Perfundle // Nov 12, 2014 - 12:57pm
3. Elo attempts to rank each team and predict the margin of victory/loss against every other opposing team at each moment in the season. It's a handicapping and pairing system. Of course, as opposed to how I will describe DVOA below, it is going to be more predictive of game results than DVOA.
I don't believe this for one second, and feel so because Silver himself doesn't either. You'd think if ELO was a good predictor that Silver wouldn't repeatedly tell readers not to use his numbers to bet. As I noted in post #116, a picking method using no input whatsoever from this year, but roughly approximating Silver's preseason ranking, has managed over a 70% success rate, mainly because the teams with at least a point differential of 50 from last year have cumulatively gone 64-33-2, and the teams with at most a point differential of -50 have cumulatively gone 27-65.
I tried to extrapolate his numbers to the start of last season to see for myself, but it turned out that the formula he provides doesn't remotely match up with the weekly changes in his numbers. For example, take the first game of the season. Seattle started off with a ELO rating of 1674 compared to 1495 for Green Bay, and the Seahawks won by 20 at home. His formula is Margin of Victory Multiplier = LN(ABS(PD)+1) * (2.2/((ELOW-ELOL)*.001+2.2)), and all that is multiplied by K which he sets at 20. But put those numbers into the formula (with Seattle's rating increased by 65 on account of home-field advantage) and you get 71.94, which is nowhere near the 11 Seattle goes up by the following week. Also, I don't see how Denver goes up more than Seattle when they only won by 7; inputting the numbers for Denver and Indianapolis gives you 20*(ln(7) + 1)*(2.2/((1619 + 65 - 1535)*0.001 + 2.2) = 55.18.
#171 by Tim F. // Nov 12, 2014 - 4:55pm
"I don't believe this for one second, and feel so because Silver himself doesn't either. You'd think if ELO was a good predictor that Silver wouldn't repeatedly tell readers not to use his numbers to bet."
Absolutely, I do. By the very nature of its inputs, Elo is going to be a better game predictor than DVOA. That's not to say that I would use it as one however, no. And, of course, Silver declaims it. It may be a reasonably good game predictor but it isn't designed to beat sportsbook spreads. Silver knows it has a success rate against the spread comparable to a newbie Blackjack player playing the conventions and that doesn't overcome the vig of sports betting. He'd lose absolutely all credibility if he claimed it was a good sports betting tool. But every responsible party trying to do these "predictions" is going to declaim liability — even the shady businesses that claim to sell winners and absolutely tout their success rate and ability to beat the system. Silver is literally the only rockstar statistician capable of creating his own media empire, leveraging his successes in baseball and the Baseball Prospectus into political success at the Daily Kos and NYT to a bestselling book to his own semi-independent media empire with fivethirtyeight at ESPN now covering every big data/statistical news story he can, with a stable of his own fellow statisticians/journalists... He's had to defend and declaim his own political work for several election cycles, explaining that it is simply statistical analysis. He would be absolutely insane to destroy all of that by claiming his NFL Elo is a good predictor of game results that people can successfully use to win weekly Pick'Ems, if not actual sports wagers. Even if Elo is getting good early results now, Silver knows it didn't predict a Giants SB win — why would he gamble his credibility when he's... a statistician who knows the math?! But just because he is a responsible, ethical statistician who understands the value of his own work doesn't make it a bad rating system. Just as, Elo being a better game predictor doesn't make DVOA a bad statistical analysis in its own right.
As to trying to work out his math, I haven't fully plunged down the rabbit hole, but you are making a few errors, I think. The complete formulae are not being fully elucidated as far as I can tell. It definitely has very few inputs but there is clearly more adjustment than the detailed margin of victory adjustment. (Silver, like FO, would be foolish to fully reveal their system, data, and simulations; this is how they make their money.)
It's easiest to understand a bit more by simplifying (using that simple k value). Let's assume k is always constant (most Elo systems have a curve to k) and there are absolutely no adjustments made (for expected result, past success, margin of victory, home field advantage, etc): a straight win would be +20, a straight loss would be -20, a tie would be 0.
Looking at weekly results we see, generally speaking, point increments in the max range of +/-50 and at minimum of +/-5 but generally more tightly grouped around that basic non-adjusted +/-20 k value. This suggests maximum adjustment of 2.5 and a minimum adjustment of .025. Seattle increased by +11 which equals a .55 adjustment, and Denver increased by +12 which equals a .6 adjustment. My math (admittedly, possible wrong) gives a straight MoV adjustment of 2.815 for Seattle and 2.003 for Denver, leaving an unknown adjustment factor of .195 for Seattle and .3 for Denver... still a ways apart but reasonable considering the potential for several unknown adjustments not being fully elucidated with precise formula and the differences between the two teams and their opponents. The out-of-range MoV adjustment for Seattle and the high MoV adjustment (but I admit, I moved fast, may have misunderstood, and may be in error) when we can see that there should be (and would expect) a very small adjustment tells me there are other adjustments being made beyond what Silver discusses but that these adjustments do not seem out-of-range or unexpected.
#175 by Perfundle // Nov 12, 2014 - 5:47pm
Silver, like FO, would be foolish to fully reveal their system, data, and simulations; this is how they make their money.
Then he should have mentioned that, as FO does, instead of saying of Elo, "I’m simple, transparent and easy to work with." No one reading that article would've walked away with the impression that there were other factors in the formula, and it's not like Elo ratings for other sports include extra components either. The fact that he went out of his way to get technical about the autocorrelation issue suggested that everything had already been revealed. And Silver has been very upfront with all the tweaks he makes to his political models, even unpopular ones, so having the numbers not come out right was an even bigger surprise.
#181 by Tim F. // Nov 12, 2014 - 7:05pm
I don't find "I’m simple, transparent and easy to work with." coming from an article that is using a light-spirited, anthropomorphic algorithm to try to explain complex statistics remotely contradictory with the belief that he hasn't fully detailed his calculations. Elo's in virtually any application (chess, online games, other team sports) are simple, transparent, and easy to work with. Scores are cumulative, provide ranking, provide anticipated winner/loser, and in some cases point differential in one number, and are going to increase for a win and decrease for a loss within a fairly narrow, simple range. I would describe that as simple and transparent and easy to work with without ever knowing the calculations or how to perform them.
"The fact that he went out of his way to get technical about the autocorrelation issue suggested that everything had already been revealed."
There is zero logic to that. Virtually every FO article ever explores some statistical detail of their process, does that imply to you that everything is known and fully revealed about DVOA and other FO stats?
I treat his silly conversation between himself and with his own algorithm as a basic FAQ covering the most obvious and basic questions to come up, not all — in fact, they specifically seemed design to answer the 3 or 4 questions to come from people who are familiar with FO. Take a look at the "First Time Here?" page on this site: doesn't reveal every detail of every stat presented here, does it?
Also, it's completely untrue that Silver has been 100% transparent with his electorate polling analyses... Yes, he has needed to and has explained different weightings and the reasoning behind weightings and various other elements of his system in the face of well-intended but usually misinformed criticism. But the actual algorithms, precise parameter values, actual data sets going in etc are 100% proprietary and secret. See http://www.washingtonpost.com/blogs/wonkblog/wp/2012/10/30/the-nate-silver-backlash/ He's not the best known and most successful statistician for nothing; 100% transparency means Nate Silver is largely irrelevant.
Some of your conclusions are far too literal-minded, misinformed and/or not based on sound logic. Sorry.
#183 by Perfundle // Nov 13, 2014 - 12:12am
There is zero logic to that. Virtually every FO article ever explores some statistical detail of their process, does that imply to you that everything is known and fully revealed about DVOA and other FO stats?
The key word is some, and because of that, it automatically does not imply to me that everything is known. It's very clear when reading FO people explain their system that a lot of their numbers are proprietary.
And, so much for your theory that Silver's formula needs to be proprietary to make money. I've figured out his formula from examining the original Elo formula. Here is the full formula, which matches his weekly changes 100% percent:
Adjustment to winning team = LN(ABS(PD) + 1)*(2.2/((ELOW - ELOL ± 65)*.001 + 2.2))*(1 - 1/(1 + 10^(-(ELOW - ELOL ± 65)/400)))
That final portion that he left out is the difference between the expected winning percentage and the actual winning percentage, the latter being 1. I should have realized he still retained that portion of the formula from the original, because he mentioned it when talking about how to calculate a team's probability of winning. I'm guessing he left it out because it would've looked too intimidating compared to the rest of his explanations. Here's how the above formula compares to his adjustments for each winning team in week 1:
Seattle
10.8 11
Atlanta
16.2 16
Buffalo
18.7 19
Houston
20.8 21
Tennessee
38.2 38
Miami
33.3 33
NYJ
9.5 10
Philadelphia
10.8 11
Pittsburgh
5.7 6
Minnesota
45.6 46
Cincinnati
23.2 23
San Francisco
18.3 18
Carolina
14.8 15
Denver
11.6 12
Arizona
5.7 6
Detroit
26.7 27
So there you have it. His formula is completely accessible to anybody, and other people can choose to copy his ideas and write their own article with the exact same spreads and updates, although there would clearly be no point to it.
#184 by Perfundle // Nov 13, 2014 - 11:20am
Armed with this, last season's starting Elo rankings can be calculated:
1631 New England Patriots
1627 San Francisco 49ers
1624 Baltimore Ravens
1603 Denver Broncos
1594 Seattle Seahawks
1585 Atlanta Falcons
1583 Green Bay Packers
1563 New York Giants
1559 Houston Texans
1552 Cincinnati Bengals
1541 Minnesota Vikings
1529 Chicago Bears
1529 New Orleans Saints
1523 Pittsburgh Steelers
1519 Washington Redskins
1513 Carolina Panthers
1503 Indianapolis Colts
1486 San Diego Chargers
1481 Dallas Cowboys
1478 Miami Dolphins
1463 Tampa Bay Buccaneers
1460 St. Louis Rams
1453 Tennessee Titans
1438 Buffalo Bills
1427 New York Jets
1426 Detroit Lions
1416 Cleveland Browns
1415 Arizona Cardinals
1397 Oakland Raiders
1396 Philadelphia Eagles
1353 Kansas City Chiefs
1339 Jacksonville Jaguars
As I expected, last year's straight up predictions through week 10 were horrible, getting 91 out of 147 for a meager 61.9% success rate. Elo is most useful near the end of the season, when enough data has been gathered, but it's simply too slow to react at the start.
#173 by EricL // Nov 12, 2014 - 5:38pm
One of the problems I have with using ELO for football is based in its source: chess. Your ELO rating in chess is considered to be "provisional" until you've got 30 (some organizations use 25) games rated. "Provisional," in this context, means "close, but may be inaccurate and subject to variance."
30 games in the NFL is nearly two seasons. If you compare the rosters of nearly any team at the end of season X and the beginning of season X+2, there's very little commonality.
By the time an ELO rating has enough sample size to be considered accurate, the team which its trying to measure has changed enough to effectively not be the same team.
ELO could work in baseball, hockey, or basketball. Not football.
#178 by Tim F. // Nov 12, 2014 - 6:34pm
Any chess organization is also going to usually have 100s-1000s of "players" (not 32) that are going to compete in 10s-100s of matches — that only have 3 basic outcomes, not scores — annually as well. Those several differences are not sufficient to say that Elo is completely useless for the NFL however. It's clear from the provided explanations that Elo was not only calculated for this season but also past seasons and that numerous simulations were then run to verify the relative accuracy of its assumptions and adjustments (see the mention of it not predicting a Giants victory in the SB).
I have no issue with Elo as a ranking system and game result predictor. I'm sure the statistics are sound. I just don't find it any more compelling than basic NFL stats, a Power Ranking, or Vegas line: it takes conventional wisdom and last year's ranks and only takes into account scores. Sagarin had been doing it for years (I think, I recall) with his own formula... just the cult of Nate Silver's celebrity has some thinking it is either the true Philosopher's Stone or a complete and utter sham of pseudoscience, like a cheap fortuneteller, when it's neither: it's a basic, well-established ranking algorithm, nothing more.
#199 by Pat // Nov 14, 2014 - 2:19pm
Point of note: Elo is not an acronym. It's a guy's name, be nice to him. It doesn't get capitalized. Doing so is wrong, and just because other people do it doesn't make it right.
The biggest problem with using Elo for football is that it's stupid. Elo developed it specifically because chess has no score - only "win, loss, or tie". Football has a score, therefore, using Elo is silly. You're specifically ignoring vast amounts of information which you could use. You're also right in that Elo was developed assuming you'll play a large number of games, and ratings should not move significantly for each game. So Elo ratings in football will do well in stretches where the best teams stay relatively static, and poorly when change happens, and in the long term, they'll fare relatively bad. (For a pure W/L based system, Laplace's method would be significantly better, since it would be a higher-order opponent correction as well).
If Elo outperforms a game score derived predictive system in any one season, it's pure chance, not because Elo is better. Game scores contain significant information, as has been shown a bajillion times over (the fact that Pythagorean wins correlate with wins tells you that straight away).
All that being said!
The thing is, football is like, 50% pure chance, from a statistical point of view. Judging a metric by how often it picks winners straight up is stupid. The repeated statement from people in this thread that "a metric needs to be best at picking winners!" is just wrong, flat out.
Why? Suppose Miami played San Francisco at home. By DVOA that would be a virtually even strength game - home field is about 15% or so. So you ask DVOA "who's going to win?" DVOA says "hell if I know. It's a toss-up." Then you force DVOA to make a decision, and it says "oh, hell, I dunno, a team like San Francisco wins this game like, 2% more than a team like Miami." Then Miami wins. And you say "stupid DVOA!"
The model predicted that it couldn't predict the game. In fact, a prediction system should be able to predict how well it should be doing. And that is the measure of how good a prediction system is - not how good it is at picking games, but how good it is at predicting the percentage of games it's correct at picking.
#200 by jbird1785 // Nov 14, 2014 - 3:21pm
The model predicted that it couldn't predict the game. In fact, a prediction system should be able to predict how well it should be doing. And that is the measure of how good a prediction system is - not how good it is at picking games, but how good it is at predicting the percentage of games it's correct at picking.
Ironically enough, Silver has made this point very recently with regard to election models.