Writers of Pro Football Prospectus 2008

Most Recent FO Features

RiceSid07.jpg

» Sidney Rice: What Could Have Been?

Sidney Rice has retired. Is he the most random single-season DYAR leader ever? One-year wonder? Injury prone? We offer a career retrospective for the second-best wide receiver named Rice in NFL history.

22 Dec 2010

Week 15 DVOA Ratings

by Aaron Schatz

Thanks for bearing with us this week, readers. I was super sick for about 36 hours, Monday night into Tuesday night, so I wasn't even able to sit down and knock out DVOA until Wednesday morning. Let's see if we can hit the high points on this week quickly before we get into the numbers.

The 2010 New England Patriots have now passed their 2007 brethren for the top offensive DVOA of all-time, and their win over Green Bay launched them into consideration as the greatest team of the DVOA Era. Only the 1995 49ers and 2007 Patriots had higher ratings through 15 weeks. (For those of you who may not remember past commentaries on early seasons of DVOA, 1993-1995 unexpectedly see the 49ers higher than the Cowboys in DVOA in the years the Cowboys won the Super Bowl, and vice versa.) You may ask yourself how the Patriots scored so high, given that they barely beat the Packers. Well, a couple reasons there. First, the opponent adjustments don't know that Matt Flynn is not Aaron Rodgers. That is, understandably, an asterisk next to the Patriots' current total rating as well as their improved defensive DVOA, but it doesn't mean anything for the Patriots' offensive DVOA. In addition, the Patriots' rating is not hurt by the surprise onside kick that started the game. Onside kicks are not included in our special teams rating, and because DVOA is a per-play metric, a strategy that gives the Packers more offensive plays means the Packers have a better chance to win the game but doesn't actually affect the Patriots' DVOA rating. It makes sense, if you think about it: the surprise onside kick is generally a strategy used to try to gain one more possession so you can beat your opponent without actually playing better overall. As long as you beat your opponent on that one specific play, the onside kick, you get a huge gain.

This was the sixth straight week the Patriots had offensive DVOA above 40%. They have not put up a game with offensive DVOA below 0% all season.

You also might notice that out of the five teams with over 40% total DVOA through Week 15, not one actually won the Super Bowl, although the Patriots still might and three of the five teams ranked six through ten did.

The Patriots' domination over the rest of the league looks even larger when we look at weighted DVOA. New England is nearly 15 percentage points ahead of the rest of the NFL. The San Diego Chargers are now second in weighted DVOA, as their early-season special teams failures move further back in the rearview mirror. The Chargers also make a big move this week, going ahead of the 1997 Seahawks and 2000 Bills so that they no longer rank as the worst special teams DVOA has ever measured.

And, in a similar move, we're removing the "worst defense ever" from the "Best and Worst Ever" watch at the bottom of the page. The 2010 Broncos, Jaguars, and Texans have improved to the point where only one of these teams would still be in the bottom ten (Denver, which would be eighth).

Thank you everyone for bearing with my lateness this week. The specific position and team stats pages are not yet updated, but I'll try to get all that taken care of in the next hour or two, and will post here when the updates are finished. We'll get premium and picks against the spread done by the end of today as well.

* * * * *

These are the Football Outsiders team efficiency ratings through 15 weeks of 2010, measured by our proprietary Defense-adjusted Value Over Average (DVOA) system that breaks down every single play and compares a team's performance to the league average based on situation in order to determine value over average. (Explained further here.)

OFFENSE and DEFENSE DVOA are adjusted to consider all fumbles, kept or lost, as equal value. SPECIAL TEAMS DVOA is adjusted for type of stadium (warm, cold, dome, Denver) and week of season. 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 NE 40.9% 1 46.4% 1 12-2 48.3% 1 9.5% 24 2.1% 9
2 PIT 32.7% 2 31.7% 3 10-4 14.8% 6 -16.7% 1 1.2% 14
3 PHI 27.0% 3 26.5% 6 10-4 26.4% 2 1.0% 11 1.6% 11
4 SD 24.0% 6 32.1% 2 8-6 21.5% 4 -11.8% 3 -9.3% 32
5 BAL 23.8% 5 30.7% 4 10-4 10.4% 13 -7.2% 6 6.1% 4
6 NYG 22.0% 4 27.0% 5 9-5 13.7% 8 -13.4% 2 -5.1% 31
7 GB 17.8% 7 16.0% 8 8-6 11.5% 10 -9.3% 5 -3.0% 29
8 ATL 15.7% 8 13.7% 9 12-2 13.0% 9 2.3% 15 5.1% 6
9 NYJ 13.8% 11 7.5% 11 10-4 2.6% 17 -5.0% 8 6.2% 3
10 NO 13.3% 9 16.4% 7 10-4 14.1% 7 -1.2% 10 -2.0% 24
11 TEN 12.4% 10 8.2% 10 6-8 0.9% 20 -6.6% 7 4.9% 7
12 MIA 8.2% 12 7.3% 12 7-7 5.8% 15 -3.3% 9 -0.9% 19
13 KC 3.9% 14 -5.1% 22 9-5 11.0% 11 5.1% 18 -2.0% 23
14 IND 3.6% 13 1.5% 17 8-6 15.2% 5 6.6% 20 -5.0% 30
15 HOU 1.8% 15 4.9% 13 5-9 25.0% 3 20.5% 30 -2.7% 28
16 CHI 1.7% 18 4.8% 14 10-4 -14.5% 30 -9.4% 4 6.9% 1
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 0.1% 16 0.5% 18 5-9 0.0% 22 1.9% 14 2.0% 10
18 TB -3.8% 17 -2.1% 19 8-6 8.3% 14 11.1% 26 -1.0% 20
19 JAC -4.4% 19 2.6% 16 8-6 11.0% 12 21.4% 31 6.0% 5
20 DET -5.7% 20 -4.2% 20 4-10 0.7% 21 7.8% 21 1.4% 13
21 OAK -8.2% 23 4.3% 15 7-7 -4.7% 24 3.8% 17 0.2% 18
22 CIN -9.1% 24 -13.1% 24 3-11 1.9% 18 8.8% 23 -2.3% 26
23 DAL -10.9% 26 -15.5% 25 5-9 1.2% 19 13.6% 28 1.5% 12
24 SF -11.1% 21 -4.4% 21 5-9 -6.4% 26 2.8% 16 -1.9% 22
25 BUF -12.3% 25 -9.5% 23 4-10 -1.9% 23 11.0% 25 0.6% 17
26 MIN -16.6% 22 -19.8% 28 5-9 -13.7% 27 1.1% 12 -1.8% 21
27 DEN -19.4% 29 -17.7% 27 3-11 4.8% 16 21.8% 32 -2.3% 27
28 STL -19.5% 28 -16.7% 26 6-8 -14.4% 29 6.2% 19 1.1% 15
29 WAS -20.3% 27 -21.9% 29 5-9 -5.2% 25 12.9% 27 -2.2% 25
30 SEA -22.5% 30 -33.8% 31 6-8 -14.3% 28 15.0% 29 6.7% 2
31 CAR -32.4% 32 -29.6% 30 2-12 -32.2% 32 1.2% 13 1.0% 16
32 ARI -36.5% 31 -37.9% 32 4-10 -30.5% 31 8.3% 22 2.3% 8
  • 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 NE 40.9% 12-2 34.6% 11.9 1 6.8% 7 -2.0% 17 18.2% 20
2 PIT 32.7% 10-4 29.6% 10.3 3 7.8% 4 -16.2% 30 12.9% 13
3 PHI 27.0% 10-4 28.3% 10.4 2 1.7% 17 -13.7% 28 10.0% 8
4 SD 24.0% 8-6 26.9% 9.4 6 -4.5% 26 -14.3% 29 23.4% 27
5 BAL 23.8% 10-4 14.8% 10.1 4 5.9% 8 -4.5% 19 5.1% 2
6 NYG 22.0% 9-5 24.2% 9.9 5 -3.6% 24 -1.3% 16 22.7% 25
7 GB 17.8% 8-6 18.2% 8.8 8 0.6% 18 11.9% 8 14.8% 16
8 ATL 15.7% 12-2 18.8% 9.2 7 -1.7% 20 -9.6% 24 5.0% 1
9 NYJ 13.8% 10-4 14.3% 8.4 11 8.0% 3 -5.3% 22 17.0% 18
10 NO 13.3% 10-4 19.0% 8.7 9 -8.8% 31 6.0% 11 8.6% 6
11 TEN 12.4% 6-8 7.8% 8.1 12 3.8% 14 3.8% 12 24.9% 29
12 MIA 8.2% 7-7 7.2% 8.5 10 7.0% 6 17.6% 5 10.7% 9
13 KC 3.9% 9-5 11.1% 7.5 14 -7.1% 29 2.1% 15 23.8% 28
14 IND 3.6% 8-6 2.9% 7.8 13 4.7% 10 2.1% 14 8.3% 5
15 HOU 1.8% 5-9 -1.6% 6.9 18 7.3% 5 -11.9% 27 19.8% 22
16 CHI 1.7% 10-4 0.7% 7.0 15 -1.9% 23 15.8% 6 28.2% 31
TEAM TOTAL
DVOA
W-L NON-ADJ
TOT VOA
ESTIM.
WINS
RANK PAST
SCHED
RANK FUTURE
SCHED
RANK VAR. RANK
17 CLE 0.1% 5-9 1.3% 6.9 17 5.8% 9 28.2% 1 16.8% 17
18 TB -3.8% 8-6 0.4% 6.9 16 -4.7% 27 -4.6% 20 12.0% 11
19 JAC -4.4% 8-6 -7.2% 6.3 20 4.3% 11 -9.3% 23 19.0% 21
20 DET -5.7% 4-10 -5.4% 6.3 19 4.2% 12 -4.2% 18 8.0% 3
21 OAK -8.2% 7-7 -2.8% 5.7 23 -1.8% 22 3.8% 13 36.2% 32
22 CIN -9.1% 3-11 -15.1% 5.6 25 9.7% 1 23.9% 4 8.7% 7
23 DAL -10.9% 5-9 -11.3% 5.8 22 3.9% 13 -4.8% 21 22.4% 24
24 SF -11.1% 5-9 -8.7% 6.3 21 -4.5% 25 -28.0% 32 23.1% 26
25 BUF -12.3% 4-10 -16.7% 5.7 24 8.2% 2 27.4% 2 13.3% 14
26 MIN -16.6% 5-9 -19.2% 5.6 26 3.7% 15 10.7% 9 14.3% 15
27 DEN -19.4% 3-11 -18.4% 4.9 30 -1.8% 21 12.9% 7 21.3% 23
28 STL -19.5% 6-8 -11.8% 5.0 28 -10.0% 32 -16.8% 31 11.8% 10
29 WAS -20.3% 5-9 -19.5% 5.0 29 3.3% 16 8.8% 10 8.1% 4
30 SEA -22.5% 6-8 -21.7% 5.5 27 -6.7% 28 -11.7% 26 25.5% 30
31 CAR -32.4% 2-12 -32.5% 3.5 31 -1.2% 19 24.2% 3 12.2% 12
32 ARI -36.5% 4-10 -27.0% 3.4 32 -8.5% 30 -11.0% 25 18.2% 19

Best and Worst DVOA Ever Watch

BEST TOTAL DVOA
AFTER WEEK 15
  BEST OFFENSIVE DVOA
AFTER WEEK 15
  WORST OFFENSIVE DVOA
AFTER WEEK 15
  WORST SPECIAL TEAMS
AFTER WEEK 15
2007 NE 56.8% x 2010 NE 48.3% x 2005 SF -47.6% x 1997 SEA -11.5%
1995 SF 44.2% x 2007 NE 45.8% x 1997 NO -43.2% x 2000 BUF -9.8%
2010 NE 40.9% x 2002 KC 42.6% x 2002 HOU -41.5% x 2010 SD -9.3%
2004 PIT 40.5% x 2004 IND 38.7% x 2004 CHI -37.9% x 1997 STL -8.2%
2005 IND 40.2% x 2003 KC 35.6% x 2006 OAK -37.0% x 1995 PHI -8.1%
1999 STL 39.1% x 2004 KC 35.3% x 1997 SD -32.7% x 2009 GB -7.9%
1996 GB 37.8% x 1998 DEN 32.5% x 2010 CAR -32.2% x 1996 NYJ -7.9%
2004 PHI 37.0% x 1995 DAL 32.5% x 1999 ARI -32.1% x 1998 OAK -7.7%
1994 DAL 36.8% x 2009 NO 30.2% x 1993 TB -31.0% x 2002 CIN -7.7%
2002 TB 36.0% x 2004 NYJ 28.9% x 1996 STL -30.9% x 2008 MIN -7.4%

Posted by: Aaron Schatz on 22 Dec 2010

149 comments, Last at 24 Dec 2010, 10:00pm by Andrew Potter

Comments

1
by JZ (not verified) :: Wed, 12/22/2010 - 2:14pm

First?

Seriously though, it would be nice to see not just Weighted Total DVOA, but also Weighted Offensive, Defensive, and Special Teams DVOA. It would give a clearer picture of just how much the Chargers ST has progressed over the season, for example.

5
by dmb :: Wed, 12/22/2010 - 2:20pm

Those are included on the specific pages for each unit, whose links are under the "Statistics" tab. The Chargers' ST unit is dead last in Weighted DVOA, but is "only" at -7.7%, as opposed to their unweighted -10.2%. But they're still really awful by the weighted measure; the next-worst ST Weighted DVOA is Buffalo, at -4.5%.

2
by Podge (not verified) :: Wed, 12/22/2010 - 2:15pm

Any chance of some sort of "Worst Division Ever" comparison? Surely the NFC West is as bad as any division has been?

6
by Bobman :: Wed, 12/22/2010 - 2:24pm

I think that award gets renamed in "honor" of the NFCW.

Also, Aaron, feel better.

7
by Karl Cuba :: Wed, 12/22/2010 - 2:24pm

While I have no time for those that complaint about wildcard teams not getting homefield advantage or suggest that if the NFC West winner has alosing record then they shouldn't get a playoff berth, I do think that whoever wins it should get a banner that says, 'Participant' instead of champion.

21
by Joshua Northey (not verified) :: Wed, 12/22/2010 - 2:40pm

This is exactly my attitude. :)

58
by Counterfactual (not verified) :: Wed, 12/22/2010 - 5:24pm

So after all the time I have made for your ridiculous complaints, now you say you have no time for mine!? Well, make time dammit!

40
by jimbohead :: Wed, 12/22/2010 - 3:57pm

I second this request. This is a historically bad division, and needs to be recognized and compared to other historically bad divisions!

91
by Mr Shush :: Wed, 12/22/2010 - 9:42pm

Totally not as bad as the 2004 version.

Also, the Rams may or may not win it this year, but for my money this is the last time for a long time that they won't. 2011-onwards Sam Bradford will be a lot like 1999-onwards Peyton Manning. With weaker divisional and conference opponents. Happy Christmas, 49er and Seahawk fans. Cardinal fans . . . I'm sorry. I'm so, so, sorry. Hey, at least you got to go to a Superbowl, right?

121
by dryheat :: Thu, 12/23/2010 - 9:54am

I'm not sure about that. I think that the Cardinals or especially the 49ers could be dominant in that division if they could get a B/B- quarterback. The 49ers may need a new coaching staff as well, but if you put a guy like Sanchez or Freeman on the team, they'd be much improved. A guy like Eli, and they would dominate the division.

I'm really not sure why the 49ers gave up on Shaun Hill. As good as Bradford has played at times, I think he would be the best QB in the division this year.

51
by TomC :: Wed, 12/22/2010 - 5:08pm

Request thirded.

However, having just finished an hour-long shouting match with collaborators over how to properly combine correlated posterior probability distributions, I wish Aaron the best of luck with this endeavor.

3
by Q (not verified) :: Wed, 12/22/2010 - 2:16pm

I know Dvoa is in love with short passing teams, but seeing GB's defense decline in DVOA after that game seems odd. Granted, DVOA does not know about dropped INT luck, but still its offense really only scored 17 points and struggled at times in the game. Even the 1st drive needed multiple 3rd and long conversions which cannot be that sustainable.

14
by Pat (filler) (not verified) :: Wed, 12/22/2010 - 2:33pm

but still its offense really only scored 17 points

On 8 drives! That's 2.125 points/drive, against a defense that had been giving up 1.3 points/drive.

24
by MJK :: Wed, 12/22/2010 - 2:55pm

I'm confused about what you're asking... Are you questioning why GB's defensive DVOA went down? Do you mean New England only scored 17 points on offense? (They technically scored 24 offensive points, but I get your apparent point that the TD where they started on the 4 yard line after Connolly's return should only partly count...)

As Pat suggested, let's look at drives. Leave out the 4 yard TD drive for now. NE had eight drives:
from NE 27, 7 plays, 73 yds, TD
from NE 9, 9 plays, 31 yds, P
from NE 30, 3 plays, 2 yds, P
from NE 34, 5 plays, 14 yds, P
from NE 20, 3 plays, -9 yds, P
from NE 27, 6 plays, 53 yds, FG
from NE 37, 6 plays, 63 yds, TD
from NE 11, 3 plays, 4 yds, P

I would call that 4 successful drives by the GB defense causing brief-and-outs, one semi-successful drive giving up a field position change but no points, one semi-failed drive giving up a long drive and a FG, and two unsuccessful drives giving up TD's.

Plus the 4 yard drive. Granted, a team will score a TD with a fairly high probability from the 4 yard line, but the Patriots scored a TD 100% of the time in this game when they started on the 4 yard line.

All this was against one of the best offenses of recent times. So I agree with you...Green Bay completely shut down one of the best offenses 50% of the time, and had mixed success the other 50%, which seems pretty good.

That said, GB was one of the better defenses going into the game, so the standards for them are higher in a sense...they need to do really well to keep their DVOA from dropping. And apparently they didn't do enough (no turnovers, for example, although they did cause a Brady fumble that works to their credit).

4
by Karl Cuba :: Wed, 12/22/2010 - 2:19pm

"First, the opponent adjustments don't know that Matt Flynn is not Aaron Rodgers. That is, understandably, an asterisk next to the Patriots' current total rating as well as their improved defensive DVOA, but it doesn't mean anything for the Patriots' offensive DVOA."

Has FO determined that good offensive performance does not help that team's defense? It seems obvious to me that a team that can move the ball and score points will end up with a more rested defense and that putting points on the board will end up forcing the offense to become one dimensional. There are many factors in play but I do think that a team's performance on one side of the ball will affect their performance on the other side.

P.S. Hope Aaron is feeling better.

18
by MJK :: Wed, 12/22/2010 - 2:36pm

Agree in general. I think you're right that a team with a good offense is going to get a better performance out of its defense because its defensive players are better rested. Also, a good offense is more likely to put up points quickly, forcing the opposing offense to pass more and allowing it's defense to be more aggressive.

However, in this specific instance, I would argue that the Patriots defense played so poorly against Green Bay that, even if Aaron Rodgers had played, the Packers defense couldn't have been much better rested than they were. The Packers had roughly a 2:1 advantage over the Patriots in that game both in time of possession and in offensive snaps. And the Patriots trailed for most of the game, so there was just as much pressure on them to pass and catch up as there would have been if Rodgers had been in. So in this particular instance, changing Flynn for Rodgers probably would not have affected the Patriots offensive performance much.

141
by R (not verified) :: Thu, 12/23/2010 - 8:58pm

The Pats defense played poorly? They gave up basically a fieldgoal and a TD on 6 drives in the 2nd half, while forcing 2 turnovers, and 2 punts in the 2nd half, while the offense had a grand total of -5 yards on all their drives in the 2nd half, barring the two scoring drives.

The reason the time of possesion and snaps were so differnt is that the Packers had SIGNIFICANTLY MORE POSSESIONS than the Patriots. The Patriots had a possesion less ( in each half) because of the coin/onside kick, and then gave away two more possessions with the kick return, and the pick-6. When the other team gets 4 more possesions, you expect a huge difference in TOP and plays, especially when its a pretty slowly paced game.

144
by dryheat :: Fri, 12/24/2010 - 10:46am

Considering that the Packers were converting 3rd downs over half the time, and Brandon Jackson and John Kuhn ran up and down the Pats, yes. The Pats defense played poorly. I'd bet any member of the defense or coaching staff would say so.

20
by tally :: Wed, 12/22/2010 - 2:39pm

The defense resting on the laurels of the offense can cut both ways.

A dink and dunk offense that chews up a lot of clock can certainly keep its defense off the field.

On the other hand, an explosive offense that scores quickly would "lose" the time of possession battle.

Both such hypothetical offenses could be very good according to DVOA (or any conventional measure) but could have vastly different effects on its defense assuming your theory holds true. Thus, I don't know if such a study comparing defenses based on its offense's DVOA would produce anything that couldn't be interpreted in different ways. Perhaps defensive DVOA by time of possession instead?

27
by Jimmy :: Wed, 12/22/2010 - 2:58pm

I once sent Aaron a question asking if there was any noticable difference in defensive DVOA as the number of snaps played increased. I remember hearing a colour commentator saying that defensive coordinators fully expect their defense to be gassed after 50 snaps and for its play to drop off badly. Aaron said he was a bit too busy and he wasn't sure how to get the number of plays stat as it wasn't in the play by play.

30
by MJK :: Wed, 12/22/2010 - 3:13pm

This is a really good idea. I doubt you could do this meaningfully for an individual team, because, as Aaron says, what players are in the game isn't in the play-by-play. But assuming all teams substitute players at roughly the same rate, give or take, you could probably look league wide and maybe see an interesting trend, comparable to the 370 carry rule.

I would look league wide and calculate a split at some parameterized N snaps. Calculate league-wide DVOA for all plays which occur before a defense has played N snaps in a game, and for all plays for which the play occurred after N snaps in the game. Try this for different values of N and see if there is a particular N that maximizes the difference between the two... If the difference is significant, then that would be a powerful tool for evaluating how play time gasses a defense collectively, even taking substitutions into account.

It might even be a good way to correct DVOA and make it more accurate, by discounting defensive snaps played after a certain number to get an idea of the caliber of a team's "fresh" defense...

36
by Jimmy :: Wed, 12/22/2010 - 3:35pm

I thought of it more as a league baseline like you suggest. If the talking head was correct then it might suggest that a viable strategy for trying to win would be to try to gas the defense as fast as possible to mean that you can score points almost at will towards the end of the game. Teams sure seem to get pretty tired trying to contain Vick for sixty minutes (see Giants game, possibly Bears game) and I know he is a very unique player but he may have the effect of exagerrating the effects we are discussing.

It might also be interesting to see if playing the run or pass has more of an effect. Traditional memes would say the run tires the defense out faster but they have been proven to be false before.

38
by Kulko :: Wed, 12/22/2010 - 3:48pm

I think the idea is nice, biut your methodology is wrong. You are just looking for a point with a high split, which will be mostly random effect in the samples, much like the number 370 in the RB Overuse article.

I think its netter to just plot Defensive DVOA by Snapnumber and run an extrapolation on it. If you have established, that that line drops slowly (which is far from certain I would say) you could look at the shape of the curve to see if there is a greater dropoff late in the game.

The second problem with the methodology is the sample bias. later defensive Snapps are more likely to be prevent defense snaps. But prevent Snaps give up a higher number of successes to the offense in tradeoff for time taken off the clock, so I would expect that 4th Q defensive DVOA in general is lower then average Def DVOA.

And third, worse defenses are more likely to play more snaps, as they tend to give up longer drives(, unless they drop to atrocious when they start giving up very short drives). So the high snap counts should contain an disproportional number of bad defenses.

Maybe it would be better to use a drive based metric like FEI. This would also fit with the thing you are trying to measure, as you say, better offense means less drives, while the number of snaps is controlled mostly by the Defensive ability.

72
by tally :: Wed, 12/22/2010 - 6:53pm

I compared the slope of a linear regression of defensive DVOA by quarters with time of possession and found zero correlation in 2010 across all teams. Somewhat quick, very dirty, and obviously a lot of noise and assumptions, but there you have it.

Team Slope TOP (sec/game)
ARI 1.27% 1568
ATL -5.85% 1948
BAL 0.94% 1878
BUF -6.45% 1684
CAR -5.96% 1668
CHI -3.75% 1780
CIN 8.79% 1863
CLE 4.68% 1684
DAL -4.90% 1844
DEN 10.60% 1684
DET 6.97% 1791
GB 1.91% 1898
HOU 0.71% 1743
IND 3.05% 1789
JAC 4.86% 1908
KC 1.95% 1851
MIA -3.76% 1836
MIN -4.30% 1813
NE 7.82% 1721
NO 7.67% 1891
NYG -5.82% 1993
NYJ 4.25% 1890
OAK 9.35% 1861
PHI 3.45% 1883
PIT -1.80% 1888
SD -3.86% 1990
SEA -2.09% 1727
SF 0.35% 1617
STL 4.71% 1868
TB -10.00% 1853
TEN 1.46% 1554
WAS -7.56% 1673

90
by Jimmy :: Wed, 12/22/2010 - 9:08pm

I have no idea what slope means in this context (I am fairly statistically ignorant - or at least I feel like it at times in this site). Could somebody explain?

92
by Pat (filler) (not verified) :: Wed, 12/22/2010 - 9:45pm

Yeah, time of possession isn't what you want to look at: that has everything to do with play selection. You really want to look at snaps and actual time on the field, which you don't have information for.

108
by tally :: Thu, 12/23/2010 - 1:08am

Snaps would be a little more difficult to dig out. I suppose I could compile pass and rush attempts against and penalties to get an approximation.

145
by jtual15008 (not verified) :: Fri, 12/24/2010 - 11:56am

The 370 Carry "rule" is not a rule. It is a statistical trick: selection bias. It's all about regression to the mean. Having 370 carries means the back is having a career peak year. So it would be wrong to call it a curse, because you would expect a back having a career year to regress towards the mean the next year. There just happens to be a group of backs who were in the 360-369 group who did better the next year, and a group in the 370-373 range who did worse the next year. So wall-ah! look at the data and draw the line where it fits with your theory, and you have a statistically significant "rule." But if you move the line to like 374 carries, then it means nothing. Here is the data to back me up

So people, PLEASE stop citing this a credible statistic.

39
by jimbohead :: Wed, 12/22/2010 - 3:50pm

One interesting approach to this question is to look at year-to-year changes in off/def dvoa and see if they're correlated. It wouldn't establish causality or anything like that, but it would tell you something.

9
by Bobman :: Wed, 12/22/2010 - 2:27pm

Hey, why is the time keyed to what looks like Hawaii time? Galapagos time? Easter Island? It's 10:25 out here in Seattle, lunchtime at FO's bustling corporate campus in Massachusetts (name a new building after me!), yet my time stamp says 7:28 am.... More proof of the mid-Pacific bias the media talks so much about.

8
by jay stokes (not verified) :: Wed, 12/22/2010 - 2:27pm

Playoff odds are still Week 14. As a Chargers fan I am grimly awaiting the news. Stupid ST/bad luck/Norv Turner/STL. The little PO Odds box suggests that it is through 15, but when I click on it, it goes back to Week 14.

11
by Bobman :: Wed, 12/22/2010 - 2:29pm

Looks right to me--Indy is finally above Jax, which was NOT the case a week ago.

17
by jay stokes (not verified) :: Wed, 12/22/2010 - 2:35pm

Weird, it is fine now. Maybe a cache thing.

10
by MJK :: Wed, 12/22/2010 - 2:29pm

Sorry to hear about your illness Aaron... Hope you feel better!

I was especially looking forward to DVOA this week in light of all the division structure hurting the playoffs talk. If we use DVOA as a metric of who has played the best this season, this is what the (probable) playoff picture looks like:

AFC:
AFCE: Patriots (#1, +40.9%)
AFCN: Steelers (#2, +32.7%) or Ravens (#5, +23.8%)
AFCW: Chiefs (#13, +3.9%) or Chargers (#4, +24.0%)
AFCS: Colts (#14, +3.6%) or Jaguars (#19, -4.4%)
WC: Steelers/Ravens, Jets (#9, +13.8%)

AFC teams in the top 12 possibly/likely missing the playoffs:
Chargers, Titans (#11, +12.4%), Miami (#12, +8.2%)

NFC:
NFCE: Eagles (#3, +27.0%)
NFCN: Bears (#16, +1.7%)
NFCW: Rams (#28, -19.5%), Seahawks (#30, -22.5%) or Niners (#24, -11.1%)
NFCS: Falcons (#8, +15.7%)
WC: Saints (#10, +13.3%), Giants (#6, +22%) or Packers (#7, +17.8%)

NFC teams in the top 12 possibly/likely missing the playoffs:
Giants/Packers

Seems like the fans with the biggest beef are San Diego fans (who could see the 4th best team in the league by DVOA miss the playoffs while the 13th, 14th, or even 19th best team gets in), and fans of whichever of the Packers or the Giants misses the playoffs, who see the 6th or 7th best team in the league miss while the 16th team gets in AND the 24th, 28th, or 30th best team gets in.

If you think the playoffs should be about each geographic region/set of rivalries putting forth their best to battle for the SB crown, then the system is working fine (and the NFC teams who aren't winning their division are struggling extra hard to get the #5 seed and face the NFCW, instead of the #6 seed and face the Colts).

But if you think the playoffs should be about getting the best teams in the league to play in a single-elimination tournament to find the SB winner, the system is obviously broken and this year has ample proof of that.

16
by Bobman :: Wed, 12/22/2010 - 2:35pm

Adding to that is the home-team factor. A few years ago Colts fans were ticked that their 12-win wildcard team had to go on the road to lose to a 9 or 10-win SD division champ. There was some discussion (first time I ever noticed it) about letting all div champs in of course, but giving HFA to the teams with better records (which would help keep week 17 games meaningful). Colts fans were certainly in favor of that.

I wonder if that sentiment has changed now that the Colts might win the div with a lesser record than one or boith WC teams from the AFC. I was kind of neutral then. I thought it sucked for my team, but seemed right for a div winner to host. This year of course, I am even more in favor of it ;-)

MJK, You are overlapping AFC and NFC in your penultimate graf--confusing me.

23
by dryheat :: Wed, 12/22/2010 - 2:50pm

Well, yes, but the playoffs are about getting the teams with the best record to play in a tournament, not the best by DVOA. So if San Diego misses the playoffs, it's because Kansas City has more wins/owns a tiebreaker. I don't think the system's broken because the #4 team in DVOA isn't in the playoffs any more than I do because the #4 team in turnover differential isn't in the playoffs (disclaimer: I have no idea who the #4 team in turnover differential is).

The only valid point your argument has is the NFC West thing: a team not good enough to win their division or earn a wild card has to sit while an inferior team wins their division to punch their ticket. But that would be the case if the Seahawks, for example, were 9-7 instead of 7-9, and nobody would be that upset about it except for fans of the 10-6 or 11-5 team that got left out.

37
by oi! (not verified) :: Wed, 12/22/2010 - 3:39pm

Before losing to the Ravens this week, the Saints looked likely to end the season tied for the best record in the NFC, yet playing on the road against a team with a .500 record. The idea that a 13-3 team should play at an 8-8 team just feels wrong, especially since the NFC West has seemed to get worse as the season has progressed.

I'm not sure how to fix this "problem". I don't think a wild card team should ever get a first round bye, but that's what the Saints would have if you just reseeded based on record. Maybe the two best records for division champions get a bye, then seed 3-6 based on record, with the tiebreaker going to division winners?

44
by JIPanick :: Wed, 12/22/2010 - 4:18pm

No dog in this fight, but saying that scenario was likely implies not 1. A better than 50-50 chance to win for NOR on the road against a superior team that already beat them in the Superdome AND to match ATL's record against @SEA and CAR with games @BAL and TB.

Assuming that the Falcons-Saints rematch is a toss-up (slightly generous to the Saints) the Ravens beat the Saints 70% of the time and the Bucs beat 'em 30% of the time (very rough estimates, obviously) combined with a 20% chance of an upset in each Falcons game, NOR would have only about 22.5% chance of finishing with Atlanta's record or better. That's not likely by any stretch of the word.

That said, I want to point out that if you're travelling to an 8-8 team with a 13-3 squad and you're as good as you say you are, you're gonna blow them away anyway. The system is fine.

46
by Anonymouse (not verified) :: Wed, 12/22/2010 - 4:55pm

I think there should be a rule that states that a team must have better than a .500 record to qualify for the playoffs. If a division champion doesn't qualify for the playoffs, it opens up an additional wildcard spot in that conference. With all other seedings being the same, the wildcard qualifier with the best record would be the 4-seed, hosting the 4-5 game.

52
by DeltaWhiskey :: Wed, 12/22/2010 - 5:08pm

I certainly understand the discomfort with a 7-9 team making the playoffs, while team with a winning record stays home; however, quality wise, I don't know if there's that much difference between a team that goes 7-9 and 9-7...both are pretty much average. I think, and I'm not sure I advocate this yet, if we're going to argue banning sub .500 teams, we have to abandon the division format for making the playoffs.

101
by RickD :: Thu, 12/23/2010 - 12:32am

I hope you all understand that comparing records of teams from different divisions isn't terribly meaningful. There's a reason to pick a division winner: it has done the best of a small group of teams that have almost identical schedule difficulties.

In contrast, a team from a different division in the same conference might have a considerably easier schedule. The division with the bad records might have been forced to play all the teams in the toughest division in each conference, while the division with the good records might have gotten the easiest team in the other conference and perhaps an easier division also in the same conference.

We face this every year in the AFC, where the division that has to play the NFC East always has a tougher time than the team that has to play the NFC West.

Bottom line is that wild cards have no right to play in the playoffs. They know the path: it's to win the division.

And it hardly seems sporting to tell the all the fans of a given division that their divisional contests have been meaningless. So sure, right now it looks like the Saints will get an easy road game against the NFC West champ, while either the Giants or Packers will miss the playoffs in spite of having a much better team. But that's life: the league is not predicated on the idea of picking the 12 best teams for the playoffs. And it's hard to imagine a system that guaranteed such a result.

12
by spenczar :: Wed, 12/22/2010 - 2:32pm

Aaron, I'm curious whether you personally think that this 2010 NE offense is better than the 2007 edition. It doesn't seem like it to me, but that may be because the more traditional stats aren't as flashy. I've only seen two Patriots games, and both games the Pats looked real good, but never historically good.

28
by MJK :: Wed, 12/22/2010 - 3:07pm

I think there are a couple of factors in play as to why DVOA thinks these Patriots are better than the 2007 edition, at least on offense.

First, strength of schedule. The 2007 Pats faced a pretty easy slate of pass defenses, while the 2010 Pats have faced a much more difficult slate. I don't remember for sure, but I think Aaron made a comment to the effect that they only face a few elite defenses (Pittsburgh, Chicago), but don't really get to face any cupcakes.

Second, the fantasy effect. I don't know if you play fantasy, but I've noticed that people that do tend to give extra weight to teams that put up a lot of fantasy points and underrate teams that don't. The 2007 Patriots were a great fantasy team. Brady was throwing gaudy bombs to Moss, giving both of them tons of fantasy points, and Welker was a reception machine making him a top scoring WR in all ppr leagues. Plus Sammy Morris got hurt, so the Patriots were forced to platoon their running backs less, making Maroney a passable fantasy RB. Finally, the offense put up so many points so early against weaker defenses, that their defense keyed in on the passing game and got a healthy amount of turnovers, making the 2007 Patriots fantasy defense a fairly valuable defense.

In 2010, the Patriots are a horrible fantasy team. Brady faced some tough defenses early in the season and put up decent, but not great, fantasy numbers. Scores of fantasy owners got burned by Moss being worked out of the offense, traded, and then dumped. Welker has not reproduced his earlier fantasy numbers. And the Patriots have been spreading the ball out so well among multiple recievers, TE's and RB's, and platooning their RB's so effectively, that very few Patriots players have made an impact in a fantasy league. Their defense has been bad most of the season, and has only recently started producing turnovers, so it's not a good fantasy defense. So anyone that is not a NE fan and follows teams mainly by fantasy will greatly overrate the 2007 Patriots and underrate the 2010 Patriots.

Finally, there's the defensive effect, on the offense. The 2007 NE defense probably looked better than it really was, but was pretty clearly better than this year's defense. That means the offense has had to work even harder to win in shootouts.

29
by chemical burn :: Wed, 12/22/2010 - 3:10pm

Also, I think it can't be overstated how amazing and remarkable it is for a team to avoid turnovers as well this 2010 team. DVOA kills teams for turnovers and, obviously, this team has gone a huge stretch without giving the ball away in any capacity - it's really unbelievable...

31
by dryheat :: Wed, 12/22/2010 - 3:15pm

As a fantasy owner of Brady, Welker, Bennie, and Hernandez, I disagree with the "horrible fantasy team" part.

69
by tuluse :: Wed, 12/22/2010 - 6:37pm

There is another factor why a person might think the 2010 offense is better. The 2007 offense started off just blowing everyone they came across out of the water, but got worse as the year went on. This year the Patriots started ok good but not great on offense, and have gotten better as the years has gone on.

2007 NE ODVOA after week 8: 49.0%
2010 NE ODVOA after week 8: 32.7%

81
by jaster7474 (not verified) :: Wed, 12/22/2010 - 7:41pm

Or maybe DVOA is just wrong. Considering the 2007 Pats scored 589 pts, and this year's version is on pace for 510, that does not seem a much more obvious conclusion. Occum's razor and all that.

82
by jaster7474 (not verified) :: Wed, 12/22/2010 - 7:43pm

Or maybe DVOA is just wrong. Considering the 2007 Pats scored 589 pts, and this year's version is on pace for 510, that does seem a much more obvious conclusion. Occum's razor and all that.

85
by chisox24 :: Wed, 12/22/2010 - 8:57pm

Not to put too fine of a point on it, but maybe you should know what DVOA means before saying it is wrong. Defense-adjusted. Defense-adjusted. Defense-adjusted.

Additionally, based on pure speculation which I am not backing up at all by looking at numbers, 2007 probably had more quick strike scores, while 2010 has more long sustained drives for touchdowns. FO has said that offenses that consistently move the chains are regarded well by the DVOA formula.

89
by tuluse :: Wed, 12/22/2010 - 9:06pm

It's also a per play metric. If the 2007 Pats ran more plays, they would have more points with the same DVOA.

102
by RickD :: Thu, 12/23/2010 - 12:35am

Yeah, um, DVOA is supposed to be a bit more sophisticated than "Points Scored" as a statistic.

126
by Marcumzilla :: Thu, 12/23/2010 - 10:43am

"making Maroney a passable fantasy RB"

Speaking as someone who spent a second round pick on him that year, I strongly disagree.

15
by nat :: Wed, 12/22/2010 - 2:34pm

That is, understandably, an asterisk next to the Patriots' current total rating as well as their improved defensive DVOA, but it doesn't mean anything for the Patriots' offensive DVOA.

Haven't most teams faced a backup QB one or two times this year? While I understand an asterisk on the single-game numbers, it's normal to face a backup QB by this time each year, right?

At this time of the season, most teams should get an asterisk on their total DVOA by this standard. In fact, if you don't have an asterisk on your total DVOA by this time of the year, you're getting misleading numbers and should get an asterisk.

Everybody gets an asterisk!

100
by Jerry :: Thu, 12/23/2010 - 12:04am

To be fair, if Aaron hadn't mentioned the backup QB, someone would have brought it up in the comments.

124
by nat :: Thu, 12/23/2010 - 10:29am

Actually, it's surprising how seldom the effect of backup QBs on opponent's DVOA has come up in DVOA threads this year.

Week 9 had a brief mention of the Giant's unusual number of games against backups. Week 11 had a general discussion on adjusting DVOA for injuries, which included comments on the difficulties of defining "backup" QB, giving several examples. There have been other discussions of injuries that could probably be considered the same general topic.

I suspect you're right about this week, though. Not because the backup QB issue is legitimate or the asterisk deserved. Because top teams attract bogus negative analysis.

127
by Eddo :: Thu, 12/23/2010 - 11:22am

My guess is that it hasn't come up because there have been so many games started by backup QBs that the perception (and I don't know if it's correct or not) is that most teams have been affected equally by facing backups.

13
by PatsFan :: Wed, 12/22/2010 - 2:33pm

NE's defense up to 24? Sacrilege! It must remain #27.

19
by Bobman :: Wed, 12/22/2010 - 2:37pm

But 24 (genuflect twice)is a multiple of 12 (genuflect), and we all know what that means.

Damn, for a couple weeks back in, what was it, 2004, it was pretty funny seeing how creative we were in working that into almost every post.

34
by Anonymous1 (not verified) :: Wed, 12/22/2010 - 3:24pm

"NE's defense up to 24? Sacrilege! It must remain #27."

NE's improvement here is fascinating to me considering how poorly the D played to the naked eye. They jumped as much as they had after trouncing both NY and Chicago. Yes, I know those two teams aren't offensive juggernauts, but 10 total points between them is still pretty impressive.

All told, I think this is closer to where NE should be ranked, IMO. Yards and 3rd down conversions come easy, but they consistently make plays when they need it and teams don't put nearly as many points on the board as the other stats would appear to indicate.

103
by RickD :: Thu, 12/23/2010 - 12:36am

Yeah, I have to say I'm baffled.

22
by Biebs (not verified) :: Wed, 12/22/2010 - 2:49pm

I'm assuming I'm missing something, but how is it that the Jets are 70% likely to get the #5 seed for the wild card, and only 25% likely to get the 6 seed. By my count, the only way the Jets get the 5 seed
is to be tied with Pitt (Jets lose vs. Chicago or Buffalo AND Pitt loses to Carolina or Cleveland)
Or for them to be ahead of Baltimore (Jets sweep Chicago and Buffalo AND Baltimore loses to either Cincinnati or Cleveland)

Then any other scenario where the Jets end up as a 6 seed (Short of NE losing final two and Jets winning final two, which does account for the final 2%)

25
by spenczar :: Wed, 12/22/2010 - 2:56pm

The odds come from doing 10,000 simulations of the league. In 7,000 simulations, the Jets got the #5 seed. In 2,500 simulations, the Jets got the #6 seed. In 200 simulations, the Jets won the division.

The odds that are then written up aren't cumulative.

104
by RickD :: Thu, 12/23/2010 - 12:51am

I agree that's very odd.

The Jets certainly have the roughest schedule of the three teams, with a road game in Chicago before a home game against Buffalo.

Pittsburgh would lose the tiebreaker to the Jets, but they play Carolina and Cleveland.

Baltimore would win the tiebreaker with the Jets, and they play Cincy and Cleveland. But it's not even that the playoff odds think that Baltimore will be winning the division. The odds page has the Steelers as the favorite over Baltimore 77%-22%. So nearly half the time, the random generator thinks that Baltimore and the Jets will be the wildcards, but that the Ravens will have a worse record.

Going by the numbers alone:

Baltimore and Pittsburgh have much higher DVOA's than the Jets.
Baltimore and Pittsburgh have schedules against weaker DVOA opponents than the Jets.

And yet, more often than not, the simulator gives the Jets the #5 seed.

Has the simulator forgotten that the Ravens beat the Jets earlier this season? That's the only explanation I can think of. The Ravens should be winning the tiebreaker, but the only way these results can be understood is if the simulator were giving the #5 seed to the Jets when the teams were tied.

26
by Aaron Schatz :: Wed, 12/22/2010 - 2:57pm

Team DVOA stats are now updated along with playoff odds. Individual, OL, DL to come soon. I need to rest first.

41
by Kulko :: Wed, 12/22/2010 - 3:57pm

Get well soon Aaron! And after that could you please give us the odds of NFCW Champion being 7-9?

I know its like gawking at an accident site, but I cant stop it.

32
by stephenmcf :: Wed, 12/22/2010 - 3:15pm

I know they don't deserve it, but Arizona is actually 4-10, not 3-10. Is there a record for offensive fumble recovery touchdowns in a season? I think Arizona has 3..

33
by GoDillos (not verified) :: Wed, 12/22/2010 - 3:18pm

Wow - The Colts are dead last in the AFC South in weighted DVOA. I wonder is that has EVER happened before.

105
by RickD :: Thu, 12/23/2010 - 12:53am

That's weird. Given the collapses of the Titans and the Texans, it's especially strange to see this happen with Weighted DVOA.

35
by Raiderjoe :: Wed, 12/22/2010 - 3:33pm

Raiders going to makr Clots deader thid Sunday on Cbs. Takr the Raiders +3 if sane person.

93
by Mr Shush :: Wed, 12/22/2010 - 10:15pm

Riadres a mraige cratered by sterngth of sechduel. Clots ont sa dumb as tehm. Pytone trwho for tol of tards afianst ghem. Godot becuase he'a on my fnasty tema. Bada bcusea heat Clots - Texans fan.

96
by Raiderjoe :: Wed, 12/22/2010 - 10:27pm

Not relally sure what you wrote there but ammde out thing about Raiders scheudle stength. Not ture. Raiders and Cotls have similar scheudle strenght numebrs. KNow this ebcuause base picks off it.

99
by tuluse :: Wed, 12/22/2010 - 10:34pm

Are you saying you're waiting for Godot?

42
by Karl Cuba :: Wed, 12/22/2010 - 4:05pm

I find it hard to believe that noone here has anything to say about women with pretty feet.

60
by MilkmanDanimal :: Wed, 12/22/2010 - 5:26pm

Aaron still needs to calculate the PVOA (toenail Polish-adjusted Value Over Average). Ms. Ryan was wearing Lollipop Cherry that day, so it bodes well for her performance.

And, in all likelihood, Rex's performance. Heyoooooooooooooooo!!!!!!

74
by TomC :: Wed, 12/22/2010 - 7:10pm

I actually had to google this, which clearly means I'm not spending enough time listening to sports radio. My reaction is twofold: 1) Unlike the Stubbly-Jenn situation, no one has been harmed, and if it were up to me we'd all shrug and move on; 2) I've seen some comments to the effect of "this is a private matter between Ryan and his wife"---no, your sexual quirks are a private matter until you start posting them on the internet; after that, not so much.

80
by Raiderjoe :: Wed, 12/22/2010 - 7:39pm

raiders 4-0 vs hors etea,ms so fra this seaosn. Goign to make it 5-0 on sudnay

106
by RickD :: Thu, 12/23/2010 - 12:54am

So the horseshoes on their helmets are going to be the downfall of the Clots? Interesting...

94
by Mr Shush :: Wed, 12/22/2010 - 10:18pm

No such thing. There are the comparatively rare inoffensive feet, and the far more common feet feet. Feet are bad. You might as well ask for comments on the good Alex Smith.

43
by DeltaWhiskey :: Wed, 12/22/2010 - 4:11pm

After last week, playing with the clustering and using new cuts. They are as follows (values are in relationship to the mean, ie >1.5SD means more than one SD above the mean.

ELITE: > +1.5SD
GOOD: +0.5SD TO +1.5SD
AVG: -0.5SD TO +0.5SD
BAD: -0.5SD TO -1.5SD
HORRID: < -1.5SD

TOT DVOA
MEAN: 0.94%
SD: 19.01%

ELITE: NE PIT
GOOD: PHI SD BAL NYG GB ATL NYJ NO TEN
AVG: MIA KC IND HOU CHI CLE TB JAC DET OAK
BAD: CIN DAL SF BUF MIN DEN STL WAS SEA
HORRID: CAR ARI

COMMENTS:
1. Range of AVG teams is from 8.2% to -8.2%. The symmetry is beautiful, but moreover, as noted last week, 7.5% DVOA appoximates 1 win, so the AVG range runs from about 7 to 9 wins.
2. Could argure that OAK CIN DAL SF belong in the AVG group or could argue that TB
JAC DET OAK don't belong in AVG.

OFF DVOA
MEAN: 3.88%
SD: 16.20%

ELITE: NE
GOOD: PHI HOU SD IND PIT NO NYG ATL
AVG: GB KC JAC BAL TB MIA DEN NYJ CIN DAL TEN DET CLE BUF OAK
BAD: WAS SF MIN SEA STL CHI
HORRID: ARI CAR

COMMENT:
ARI CAR offense is almost as far behind CHI's offense as NE offense is ahead of PHI (okay, not almost, but the level of relative 2010 suckitude is almost as stunning as NE's awesomeness).

DEF DVOA
MEAN: 3.43%
SD: 10.07%

ELITE: PIT NYG
GOOD: SD CHI GB BAL TEN NYJ MIA
AVG: NO PHI MIN CAR CLE ATL SF OAK KC STL IND DET ARI
BAD: CIN NE BUF TB WAS DAL SEA
HORRID: HOU JAC DEN

S.T. DVOA
MEAN: 0.48%
SD: 3.84%

ELITE: CHI SEA
GOOD: NYJ BAL JAC ATL TEN
AVG: ARI NE CLE PHI DAL DET PIT STL CAR BUF OAK MIA TB
BAD: MIN SF KC NO WAS CIN DEN HOU GB IND NYG
HORRID: SD

GENERAL COMMENTS:
1. SD, NYG and PIT are the only teams that are GOOD on both OFF and DEF.
2. OAK, CLE, and DET are across the board AVG
3. WAS is across the board BAD and SEA is the only other team that is BAD on OFF and DEF.

45
by Andrew Potter :: Wed, 12/22/2010 - 4:39pm

My favourite part of the comments thread every week. I'm glad you're using the different splits, and that very much passes the eyeball test.

Chicago's special teams looks incredible. How far is that from being top 10 historically?

47
by DeltaWhiskey :: Wed, 12/22/2010 - 5:00pm

Thanks for the props.

I agree that this is passing the eyeball test much better than my initial efforts.

My initial goal was to try and get away from looking at rankings and recognize/demonstrae that clusters are a better way to look at teams, that is the argument that the #8 ranked team was somehow much better than the #16 ranked team didn't hold water as they both were easily within a SD of the mean and therefore not likely to be statistically different; however, when I set up my original model (+/- 1 SD from the mean), it was very easy to demonstrate this and it was based on a sound statistical principals. Unfortunately, as many noted, the AVG group (while statistically justified) was just to damn big to be of any meaning. The current setup, is statitically weaker, but I think has more face validity and utility and is therefore satisfactory at present. With a little time, I'll look at previous years to see how they shook out.

Finally, I have to thank all who kept making the point about my AVG group being overly huge, and especially nat who provided a split I could live with for the impetus to make the change.

55
by Athelas :: Wed, 12/22/2010 - 5:18pm

My 6th grade son has to find some sports statistics to graph--I like the bell curve look of yours.

And I agree, passes the eyeball test.

57
by DeltaWhiskey :: Wed, 12/22/2010 - 5:21pm

What kind of cool class is your son in that he gets to graph sports stats? Might be careful, he could learn something if he's not careful.

62
by Athelas :: Wed, 12/22/2010 - 5:28pm

Plain old Math--but a really good teacher!

135
by EasyLikeSundayMorning (not verified) :: Thu, 12/23/2010 - 3:31pm

Me, too. Good work.

71
by tuluse :: Wed, 12/22/2010 - 6:48pm

1. Range of AVG teams is from 8.2% to -8.2%. The symmetry is beautiful, but moreover, as noted last week, 7.5% DVOA appoximates 1 win, so the AVG range runs from about 7 to 9 wins.

That's pretty awesome.

I also like the fact we have at least one team in each ELITE category now.

Good work on all this.

While I'm here, I wonder if you could look at weighted DVOA too especially as the playoffs get close.

95
by chemical burn :: Wed, 12/22/2010 - 10:20pm

I know this is maybe a strange thing to ask, but I would love to see these presented as an actual regular FO column next year - I'll be really bummed if Deltawhiskey gets tired of posting them or is otherwise unable to continue. It's really just such a great way of looking at things...

115
by DeltaWhiskey :: Thu, 12/23/2010 - 8:58am

No problem.

WGTDVOA

Mean: 1.59%
SD: 20.43%

ELITE: NE
GOOD: SD PIT BAL NYG PHI NO GB ATL
AVG: TEN NYJ MIA HOU CHI OAK JAC IND CLE TB DET SF KC
BAD: BUF CIN DAL STL DEN MIN WAS
HORRID: CAR SEA ARI

48
by yakul :: Wed, 12/22/2010 - 5:04pm

Question for those of you who spend a lot more time thinking about DVOA than I do: how plausible is it that even DVOA undervalues the importance of Special Teams? The extreme examples that jump to mind are of course San Diego and Chicago. I realize the understandable bias that FO has against unsupported judgments of conventional wisdom, but in two cases, the conventional wisdom argument doesn't sound any worse than what we seem to pick up from DVOA:

1) SD
"The San Diego Chargers have an amazing offense and defense. They'd be one of the best teams 2-4 teams in the NFL if their Special Teams weren't so terrible."

versus

"The San Diego Chargers have the fourth-highest DVOA in the league, having feasted one of the easiest schedules in the NFL. They have accomplished all this despite historically bad special teams. Their lack of wins is a little confusing."

2) CHI
"The Chicago Bears are having an amazing season thanks in no small part to incredible Special Teams play and very good defense, which may or may not coincide with the good fortune to be facing the opposing team's backup quarterback."

versus

"The Chicago Bears have the 30th rated Offensive DVOA, 4th rated Defense and top rated Special Teams. For some reason they have significantly overperformed against their estimated wins."

I don't have any axe to grind here and am a huge fan of FO. However, the valuation that FO places on special teams seems like it may fire wide of the mark in terms of calculating Estimated Wins. Of course I've cherry-picked the data but I am genuinely curious.

59
by TomC :: Wed, 12/22/2010 - 5:25pm

I can't intelligently comment on SD, but as a Bears fan who's watched every game, I feel like 7 estimated wins is pretty bang on, or at least no more than 1 win off. They've won four games handily, three against lousy teams (2x MIN, CAR) and one against the Thigpen Dolphins. They've been beaten soundly twice, both by very good teams (NE, NYG). All of the other games could have gone either way, and they're 6-2 in those games. So with average luck against an average NFL schedule, 7 wins out of 14 games seems like a totally reasonable outcome. The Bear fan part of me likes to think that they are clearly trending upward, and that they are more likely to win a playoff game or two now than they would have been early in the year, but that's a WDVOA question not an Estimated Wins question.

I also feel that special teams are weighted about where intuition says they should be: important enough to swing a close game either way, but not so important that they can overcome serious deficiencies in the other two phases.

73
by tuluse :: Wed, 12/22/2010 - 7:01pm

Another thing to consider is that the Bears have had pretty amazing injury luck this year. Williams, Garza, Tinoisamoa, Cutler, and Briggs are the only starters to miss time. Cutler and Briggs each missed a single game, and linebacker is probably the Bear's deepest position.

70
by tuluse :: Wed, 12/22/2010 - 6:47pm

I wonder if special teams have an odd effect on games when they are way better or worse than normal. I mean if they're within ~5% of average they have the normal effect you would see on games, but when you get outside that range the effect magnifies. Pat made a good comment earlier this year about coaches not adapting to terrible special teams because things like punting are usually "safe" plays. They don't expect to get 2 punts in a game blocked for TDs.

78
by jaster7474 (not verified) :: Wed, 12/22/2010 - 7:33pm

To this point ... In the thread after this one, I argue for doubling defensive and special teams DVOA. But maybe the right answer is actually not a linear scaling by a factor of 2, but a quadratic scaling who's total effect ends up being a factor of 2. Maybe in the middle, there is no scaling, but on the extremes the factor is 4 or more.

114
by fool_rider (not verified) :: Thu, 12/23/2010 - 5:48am

I think this is probably a much more reasonable way to value them also.

49
by jaster7474 (not verified) :: Wed, 12/22/2010 - 5:07pm

I think there is a bit of a bias or weakness in DVOA.

Difference between top O DVOA and worst O DVOA = 80.5%
Difference between top D DVOA and worst D DVOA = 38.5%
Difference between top ST DVOA and worst ST DVOA = 15.9%

According to the current DVOA model, offensive supremecy is worth twice as much as defensive supremecy, and 5x as much as special teams supremecy. That doesnt compute to me. I would double the Defense DVOA. Id also double the special teams DVOA. This would result in O:42.5% D:40.7% ST:16.8% split. Almost the 40-40-20 guideline often mentioned when discussing relative importance of the three phases. Based on this [apologies for the formatting]:

OLD TEAM TOTAL New New
RANK DVOA Total Rank
2 PIT 32.70% 50.60% 1
5 BAL 23.80% 37.00% 2
1 NE 40.90% 33.50% 3
6 NYG 22.00% 30.30% 4
3 PHI 27.00% 27.60% 5
4 SD 24.00% 26.50% 6
9 NYJ 13.80% 25.00% 7
7 GB 17.80% 24.10% 8
11 TEN 12.40% 23.90% 9
8 ATL 15.70% 18.60% 10
16 CHI 1.70% 18.10% 11
10 NO 13.30% 12.50% 12
12 MIA 8.20% 10.60% 13
17 CLE 0.10% 0.20% 14
13 KC 3.90% -3.20% 15
14 IND 3.60% -8.00% 16
21 OAK -8.20% -11.90% 17
20 DET -5.70% -12.10% 18
24 SF -11.10% -15.80% 19
18 TB -3.80% -15.90% 20
26 MIN -16.60% -19.50% 21
19 JAC -4.40% -19.80% 22
22 CIN -9.10% -20.30% 23
15 HOU 1.80% -21.40% 24
25 BUF -12.30% -22.70% 25
23 DAL -10.90% -23.00% 26
28 STL -19.50% -24.60% 27
30 SEA -22.50% -30.90% 28
31 CAR -32.40% -32.60% 29
29 WAS -20.30% -35.40% 30
32 ARI -36.50% -42.50% 31
27 DEN -19.40% -43.40% 32

Results of this new balancing?

Old TEAM New Diff
16 CHI 11 5
24 SF 19 5
26 MIN 21 5
21 OAK 17 4
5 BAL 2 3
17 CLE 14 3
6 NYG 4 2
9 NYJ 7 2
11 TEN 9 2
20 DET 18 2
30 SEA 28 2
31 CAR 29 2
2 PIT 1 1
28 STL 27 1
32 ARI 31 1
25 BUF 25 0
7 GB 8 -1
12 MIA 13 -1
22 CIN 23 -1
29 WAS 30 -1
1 NE 3 -2
3 PHI 5 -2
4 SD 6 -2
8 ATL 10 -2
10 NO 12 -2
13 KC 15 -2
14 IND 16 -2
18 TB 20 -2
19 JAC 22 -3
23 DAL 26 -3
27 DEN 32 -5
15 HOU 24 -9

Houston (-9) is now a bad team instead of an average team.
Denver (-5) is the worst team instead of a bad team.
Chicago (+5) is a playoff team instead of an average team.
SF/Min (+5) are below average teams instead of bad teams.
Oakland (+4) is an average team instead of a bad team.

Personally, these ranking seem better.

53
by DeltaWhiskey :: Wed, 12/22/2010 - 5:14pm

While the range may be a problem, the other concern is the unequal variance:

TOT DVOA SD: 29.45%
OFF DVOA SD: 28.18%
DEF DVOA SD: 18.53%
S.T.DVOA SD: 6.24%

I'll have to play with what you've done to see if makes a difference.

Alternate thought: Perhaps this truly reflects the state of things in the NFL...high powered offense more important than defense.

75
by jaster7474 (not verified) :: Wed, 12/22/2010 - 7:19pm

But the notable changes this new valuation makes (listed at the bottom) better reflext reality, at least in my view. I am not saying the O spread and D spread need to be equal. I am saying making them equal seem to improve the results.

Of course, this is subjective. But so is DVOA, so whenever I see or hear that a team is ranked way off from their results, I'm lokking for ways to improve DVOA. Not to declare DVOA perfect, and reality as being "luck".

107
by RickD :: Thu, 12/23/2010 - 1:07am

Well, the Pats are 12-2 after all. And they are an outlier when considered to offenses of the past few decades.

I don't think we should say "hey, the Pats' offense is an outlier so the statistic must be messed up." Maybe they're just really, really good?

There's no reason to suppose a balance between offense and defense. If we kept the defenses fixed, while improved all the offenses dramatically, many more points would be scored. Must we pretend that it is sensible to lock in some mathematical relationship between offense and defense that needn't reflect reality (and, indeed, is positively counter-intuitive, once you view the matter from a certain angle).

Bottom line is this: DVOA must allow for the possibility that offenses are better than defenses. And it must allow for the opposite possibility. If it doesn't do so, it is deliberately introducing a bias.

Bigger issue here: if data shows that one statistic is greater than another, it is shallow criticism to use the sledgehammer term bias. People throw around the word "bias" all the time, and I suspect that well over 99% of the time, it's being used in a grossly non-rigorous way.

So no, if there is more variation between offenses, that fact by itself doesn't mean the statistic is "biased". Football is not a zero sum game. There is no reason to expect any kind of parity relationship between offense and defense. It's quite conceivable that one year offenses will dominate, and scores will increase, while another year, defenses will dominate, scores will go down (and the NFL will step in and change the rules to ensure that offenses dominate again).

142
by Pat (filler) (not verified) :: Fri, 12/24/2010 - 1:18am

Alternate thought: Perhaps this truly reflects the state of things in the NFL...high powered offense more important than defense.

Or just more controllable. Might seem weird that a phase that is less controllable has less variance (you normally think of something that's uncontrollable being, well, all over the place) but if you're simply less able to build a dominant defense, you'll end up with a tighter distribution.

You can also imagine it as being a resource scarcity issue: if good players on offense are rare, and good players on defense are common, it'll be easy to generate an average/good defense (in an absolute sense), leading to a small variance, and it'll be hard to generate a good offense, leading to a high variance. Or also, it can be thought of like a salary cap providing parity: make resources equivalent across the league, and teams pull together.

Given that there are teams that obviously focus on defense first (the cluster of good defense/bad offense teams in the past 10 years is pretty striking), it's pretty indicative that it's likely to be just impossible to create as dominant a defense as a dominating offense.

143
by Andrew Potter :: Fri, 12/24/2010 - 1:37am

You can also imagine it as being a resource scarcity issue

One word: Quarterbacks.

64
by Eddo :: Wed, 12/22/2010 - 5:36pm

" I would double the Defense DVOA. Id also double the special teams DVOA."

While I share your amazement at the exaggerated importance of offense this year (and DeltaWhiskey's point below about variance also _appears_ to be a problem), you can't just do something like double defensive and special teams DVOA because you feel like it.

Yes, DVOA is subjective in that Aaron chooses what goes into it. But special teams is valued at 1/3 of offense and defense because that results in the best year-to-year correlations for DVOA.

I think it would be really cool if Aaron or someone dug deeper into why offense is rated so much more importantly by DVOA this year, but I don't really have any suggestions as to why.

76
by jaster7474 (not verified) :: Wed, 12/22/2010 - 7:28pm

This presupposes the system is perfect. Maybe DVOA values offense more highly, because DVOA is wrong. I know many people here arent open to considering that possibility. But when I hear things like, DVOA always overvalues the Eagles, or when people pooh pooh obviously incorrect rankings by waving their hands and calling it "luck" or "variance" ... well thats not the scientific method. Thats being dogmatic. They should be continually improving their methods. This is just one suggestion on how they may do so.

83
by Eddo :: Wed, 12/22/2010 - 8:04pm

I'm not saying it's perfect. No one is. Aaron constantly talks about improvements and things he's looking into.

However, there's a method to it, at least. He doesn't just say, "I'm doubling factor X, which makes the numbers look better." DVOA should correlate with itself as best as possible, as it's designed to be predictive. He tests any changes he makes in this manner, to my knowledge.

I would like to see him do what you're suggesting, to see if it makes DVOA better. But simply doubling defense, as you've done, in fact makes previous years skewed too highly in favor of defense.

I'm not saying DVOA is perfect, or even the best tool there is. For one, I wonder if there is something to extreme special teams affecting games disproportionately. The Bears and Chargers would be the extreme examples.

117
by DeltaWhiskey :: Thu, 12/23/2010 - 9:08am

"However, there's a method to it, at least. He doesn't just say, "I'm doubling factor X, which makes the numbers look better." DVOA should correlate with itself as best as possible, as it's designed to be predictive. He tests any changes he makes in this manner, to my knowledge."

How he makes changes is much clearer to you than me. I've never gotten a good feel as to what he does or if what he does actually makes it better. I just push the "I believe" button and go on.

128
by Eddo :: Thu, 12/23/2010 - 11:25am

To be fair, I'm only going by how Aaron has described the process in the past. My impression is Aaron will look at the ratings, say, "Oh, I wonder if this looks weird because I'm not weighting XXX enough," then tweak how XXX is rated, and see if the correlations improve.

Not being a statistician, I'm not sure that it's the right way to test changes to the model, but at least it's a better method than "the final results just 'look' better."

130
by DeltaWhiskey :: Thu, 12/23/2010 - 12:01pm

Your impression is similar to my own. This methodology is not necessarily a bad one, but it appears somewhat haphazard to me. In the end though, what is important is the degree of improvement in the correlations, and whether the improvement is statistically significant. This is something that Aaron has not done in the past or if he has, has not reported - hence my comment regarding pushing the "I believe" button. At one point I beat this dead horse to death to the point that I was reprimanded.

87
by dryheat :: Wed, 12/22/2010 - 8:59pm

Well, DVOA isn't wrong. Maybe it isn't the best predictive measurement out there, but it can't be wrong any more than passer rating can be wrong.

109
by RickD :: Thu, 12/23/2010 - 1:11am

Indeed, what does it mean to say "DVOA is wrong".

The offensive and defensive numbers are not pegged in a way so that they must have equal impact on overall DVOA. And why should they be? There's nothing wrong with having average offensive DVOA further from zero than average defensive DVOA.

People are simply imagining constraints that do not exist.

146
by tua15008 (not verified) :: Fri, 12/24/2010 - 6:34pm

I have to preface this with a question: I do NOT mean to troll or whatever you internet geeks (just kidding) call it. I just am wondering if there is a thread somewhere to discuss the sort of large-stroke pros and cons of using DVOA. Of course in the interest not of bashing it, but of improving it through good faith debate. JS Mills said that the truth is only found through the constant questioning of the status quo. This is more true no where than on the internet. I feel like I see this happen every week, every season on FO. A team is ranked below most of the teams it has beaten. Like Dallas, with its 5 wins! I know this is just a subjective anecdotal evidence or whatever. But REALLY? 4 of Dallas' 5 wins came over teams ranked HIGHER than it in DVOA. Obviously there are very sound reasons for this I won't even bother going over, but look at it anecdotally. Houston, which Dallas beat by TWO TOUCHDOWNS, has the same record (i know record doesn't "matter") but is Houston 50-guys-in-a-bar really 25% DVOA better than Dallas? What about Detroit at 4-10? It is 6 % points better than Dallas, a team it lost to by more than TWO TOUCHDOWNS? Obvoiusly, Indy and NYG, which the Cowboys beat, are better than Dallas (I say "obviously" and no one is gonna disagree from Diehard DVOA supporters and detractors alike, but this "obviousness" comes from subjective observation. Like, there is a point at which a team is simply just that much better that it becomes "obvious" and DVOA is supposed to deal with the not-so-obvious, as it claims to. But at what point are we not seeing the forest for the TREES here?)

The same goes for Chicago. I know, I know Chicago is what, lucky? Yes, football is largely based on luck. Or at least the arguement goes that the big reason real records don't match DVOA or w/e is luck. But OBVIOUSLY, and i say OBVIOUSLY in the same context as above, Chicago at 10-4 is better than Houston, Tennessee, probably Miami (7-7). I mean chicago SHUT OUT Miami. So, obviously, Chicago should be rated below Miami, if nothing else. at SOME point the REAL record of a team HAS to count for SOMETHING. you can talk about luck all you want but a system that is intuitively just plain wrong over and over again, it cant always be luck or whatever. This FO system is good for telling small differences between close teams i think but when its about OVERALL comparisons of a teams ranks, i think the more general it gets the less descriptive of reality that it gets. Saying that Chicago is not as good as Miami, or that Dallas is not as good as Detroit . . . this is INCORRECT. How do i prove it? Miami LOST to Chicago CONVINCINGLY 16-0. Detroit LOST to Dallas by TWO TOUCHDOWNS. Nothing is more "real" than these results. The results of games should be the TEST of the THEORY that is DVOA. And when the THEORY is shown to have faults, YOU MUST BLAME THE THEORY, not REALITY for not complying. JESUS.

147
by tuluse :: Fri, 12/24/2010 - 9:13pm

I only skimmed your post, but you need to understand DVOA is just a stat, no one is claiming it's perfect or that it gets every team right. It's might be the best stat, but it still gets things wrong.

Other than that, head-to-head games are not always the best measure of if one team is better than another. In 2003, the Bills opened the season by beating the Patriots 31-0, were the Bills better than the Patriots that year?

148
by Jerry :: Fri, 12/24/2010 - 9:32pm

If you're looking for rankings by wins and losses, every newspaper runs standings. If you're looking for rankings based entirely on who beat who, take a look at beatpaths.com.

Regardless of how rankings are arrived at, lower-ranked teams are going to beat higher-ranked ones from time to time - the #1 team is rarely undefeated and #32 is rarely winless.

DVOA tries to measure how successful teams are at running offensive and defensive plays. It's not perfect, but I find it useful. When DVOA disagrees with the standings, or even with subjective observations, it's a good chance to look a little deeper at what's actually going on. If you think DVOA is consistently and egregiously wrong, maybe you should ignore it.

149
by Andrew Potter :: Fri, 12/24/2010 - 10:00pm

Yes, and CLEVELAND beat NEW ENGLAND by TWENTY POINTS. CLEARLY we should disregard EVERY other game the two teams played in favour of this ONE result.

Look, nobody's claiming DVOA is perfect (well, most people aren't - any who are, are almost certainly wrong), but neither is picking one-game samples to argue against it. For example, Chicago played Miami with Miami's third string quarterback starting on a short week of practice. How many teams would you expect to perform up to their usual standard in that situation? Over the course of the season, Miami's offense has played better than it did in that one game. Is Chicago better than Miami? Maybe. Does that one game prove it? Not so much. Therein lies the problem with using one-game samples. Cleveland may have thrashed New England - that does not automatically mean the Browns are the better team.

112
by Intropy (not verified) :: Thu, 12/23/2010 - 1:29am

I agree with the intuition. Offense and defense are on the field at the same time so they must be equally as important as one another. But the range of offense being greater than the range on defense doesn't tell you that DVOA values offense more than defense. It could just as easily be showing you that this year the 32 teams in the NFL have much more widely varying degrees of talent on offense as on defense.

Consider, what if every defense was precisely identical? Ignore special teams. Every team would score the exact same number of points every game. Those scores would tell you precisely how good each offense was and they could vary wildly in DVOA while the defenses have identical DVOAs. DVOA would still be telling you the truth. And that doesn't mean it's saying defense doesn't matter. A single team with twice as good a defense could allow half the points to be scored. It's just saying that the defenses are all the same.

A better thing to measure, in my opinion, would be variance. You'd have better reason to expect them to come out similar, but that still doesn't get around what I mentioned above. Also interesting would be raw sum. I would expect the sum of defense DVOAs to be zero and likewise with offensive DVOAs.

113
by RickD :: Thu, 12/23/2010 - 2:46am


Also interesting would be raw sum. I would expect the sum of defense DVOAs to be zero and likewise with offensive DVOAs.

Why?

Look, you're either measuring the actual performance, or you're imposing a zero sum to the statistics. You cannot do both.

If you're measuring the actual performances, you have to allow for the possibility that all the defenses are playing worse (on average, at least).

Football is not a zero sum game. There's no a priori reason to think it should have zero sum statistics.

119
by DeltaWhiskey :: Thu, 12/23/2010 - 9:43am

Sorry for the format, just some thoughts on the above.

jaster7474
“But the notable changes this new valuation makes (listed at the bottom) better reflext reality, at least in my view. I am not saying the O spread and D spread need to be equal. I am saying making them equal seem to improve the results. “

Re: Improve the results – what is the standard you’re measuring against. This is a central question that has to be answered prior to any tweaking. Are we talking correlation with WINS, Pts, Pts allowed, yds/game, etc.?

RickD
“There's no reason to suppose a balance between offense and defense….It's quite conceivable that one year offenses will dominate, and scores will increase, while another year, defenses will dominate, scores will go down (and the NFL will step in and change the rules to ensure that offenses dominate again)."

A well thought out and reasoned point.

jaster7474
“This presupposes the system is perfect. Maybe DVOA values offense more highly, because DVOA is wrong. “

I suspect DVOA values offense more b/c the nature of the game now is an offensive game. I was listening to Pat Kirwan yesterday on Sirius NFL and he (or maybe his cohost) observed that to win you no longer need a running game, you need a high powered passing offense and you hope that your defense doesn’t lose the game for you.

by Intropy (not verified)
“I agree with the intuition…A better thing to measure, in my opinion, would be variance. You'd have better reason to expect them to come out similar, but that still doesn't get around what I mentioned above. Also interesting would be raw sum. I would expect the sum of defense DVOAs to be zero and likewise with offensive DVOAs.”
by RickD ::
“There's no a priori reason to think it should have zero sum statistics.”

If every team played the same schedule, zero sum might be expected, no?

Awesome discussion. About 3 or 4 times per year these type of discussions come up and awesome.

129
by Eddo :: Thu, 12/23/2010 - 11:36am

DeltaWhiskey, your and RickD's posts in this chain have been excellent. A really good primer on the purpose and general usefulness of metrics like DVOA.

And even if you were to think that offense is over-represented this year, remember that DVOA is a year-to-year metric. So, let's look at DVOA spreads for the past several years:


Year __ Off __ Def __ ST __ notes
2010 _ 80.5 _ 38.5 _ 16.2 _ offense has much greater spread
2009 _ 56.0 _ 45.6 _ 14.3 _ slight offense
2008 _ 48.1 _ 56.1 _ 12.6 _ slight defense

Also, don't discount the possibility that DVOA is showing us that there's really not that big a difference between the best and worst defenses, while there is when it comes to offenses. This could be a very important point, implying that teams, when given the choice, should try to improve a mediocre offense instead of a mediocre defense.

132
by DeltaWhiskey :: Thu, 12/23/2010 - 12:35pm

I think a question that has to be asked is "Is it even valid to try and compare OFF DVOA to DEF DVOA?" Or "what is the best way to compare OFF DVOA to DEF DVOA?" For the time being forget about S.T. In other words, what does 1% OFF DVOA mean compared to 1% DEF DVOA? Here's one way to look at it.

The correlation between OFF DVOA and Points is r = 0.822. The regression equation for this correlation is y = 380.83x + 341.05 (y=points, x=DVOA).

Therefore a team with a OFF DVOA of 0% will be predicted to avg 21.3
pts per game (y/16, 341.05/16=21.3). This is very close to the average number of points scored per game over the DVOA era.

For defense, the correlation between DEF DVOA and Points is r = 0.776. The regression equation for this correlation is y = 406.5x + 343.09 (y=points, x=DVOA).

Therefore a team with a DEF DVOA of 0% will be predicted to allow an avg of 21.4
pts per game (y/16, 343.09/16=21.4). Again, this is very close to the average number of points scored per game over the DVOA era and fortunately close to the offense numbers as well.

Now here's where things get interesting
10% OFF DVOA = 23.7 pts/game or a delta of 0.4 pts/game
10% DEF DVOA = 24.0 pts/game or a delta of 0.6 pts/game allowed

Now looking at this year

Offense:
NE OFF DVOA 48.3% predicted 32.8 pts/game
CAR OFF DVOA -32.2% predicted 13.7 pts/game
Delta NE - CAR = 19.1 pts/game

However, given that NE is acknowledged as blowing the lid off of DVOA this year, looking at PHI instead:
PHI OFF DVOA 26.4 predicted 27.6 pts/game
Delta PHI - CAR = 13.9 pts/game

Defense:
PIT DEF DVOA -16.7% predicted 17.2 pts/game allowed
DEN DEF DVOA 21.8% predicted 27.0 pts/game allowed
Delta PIT - DEN = 9.8 pts/game

Replacing Eddo's numbers with pts/game
Year __ Off __ Def
2009 _ 13.3 _ 11.6
2008 _ 11.5 _ 14.3

Now I've got to think about what these numbers mean, but tenatively I'd venture, that if you double the value of DEF DVOA in calculating total DVOA, you're going to f8ck up TOT DVOA.

OFF DVOA of 0% can b

133
by Eddo :: Thu, 12/23/2010 - 12:48pm

Nice work, DeltaWhiskey. Though, were you to simply double defensive DVOAs, would the effect be similar? That is, the Pittsburgh-Carolina gap would double, but it would still come out to the same amount of points? (The slope of the regression line would double, while the intercept stayed roughly the same, right?)

I'm not disagreeing with your overall point, which is a very good one. You're absolutely right that one point of offensive DVOA isn't inherently supposed to be equivalent to one point of defensive DVOA.

138
by DeltaWhiskey :: Thu, 12/23/2010 - 6:14pm

Hmm...will have to think a little more...maybe after Xmas hangover wears off.

136
by Intropy (not verified) :: Thu, 12/23/2010 - 5:49pm

It _is_ both. DVOA doesn't measure anything discrete. It's a value meant to be used only relative to other DVOA values, which is still a measurement of actual performance. And zero is defined to be average, so you are imposing a zero sum as follows:
mean = DVOA sum / count
0 = DVOA sum / count
0 * count = DVOA sum
0 = DVOA sum
So it should be expected that the sum of all DVOAs is zero.

137
by DeltaWhiskey :: Thu, 12/23/2010 - 6:12pm

"So it should be expected that the sum of all DVOAs is zero."
Each week (and ea. season) we only get a sample of DVOA's, so it will not sum to zero. To sum to zero, every team would have to play each other an equal number of times.

In the ideal world, yes a zero sum would be expected. However, reality doesn't always cooperate.

139
by Intropy (not verified) :: Thu, 12/23/2010 - 6:36pm

No, it's not necessary for each team pair to play an equal number of times. Consider a degenerate set where one team plays zero times, one team plays ten times, and ten teams play one time.

The zero team is obviously average. 0 DVOA on both sides. Every time the ten team plays a one team you would expect the changes in DVOA to cancel. Ten wins big against 1_1. Now 1_1 is negative and 10 is positive. 10 loses big to 1_2. Now 1_2 is very positive for beating a good team, 10 is more average, and 1_1 creeps even lower since its one opponent just got worse. Repeat.

What is required is that every game that goes into determining how DVOA is calculated be included in the set of games being summed, which of course means more than just one season.

140
by Andrew Potter :: Thu, 12/23/2010 - 7:03pm

Unless I'm missing something, DVOA doesn't work like that in part because it only includes predictive elements.

Take a missed 20 yard field goal for example. If your team misses a 20 yard field goal, it says something about the quality of your field goal unit. Your DVOA is rightfully penalised for that. If your OPPONENT misses a 20 yard field goal, the majority of the time it doesn't actually say anything about the quality of your own special teams. Thus it isn't predictive, and your special teams DVOA does not go up as a result. One is predictive (hey, our kicker might not be that good!), the other isn't (your opponents don't all have the same potentially-not-very-good kicker and the result of field goals is mostly out of your own control).

So we have one example of a play that reduces one team's DVOA, but doesn't increase the other's. I know next to nothing about statistics, but doesn't that automatically mean it's not a zero sum game?

118
by Andrew Potter :: Thu, 12/23/2010 - 9:25am

The reason for that seems simple enough: this year, there are no truly elite defenses. New England's offense is good enough to shred any defense - for evidence of that, look at what they did to the number 1 and 4 defenses by DVOA this season (Pittsburgh and Chicago, they also put up an above-average day on the road against #3 San Diego). That level of offensive dominance looks fairly clearly, by results thus far, to be more valuable than the level of defensive strength Pittsburgh has.

Meanwhile, Carolina and Arizona have offenses so wretched that any defense in the NFL can usually stand up to them, hence the massive range in offensive value but lesser range in defensive. It doesn't really need to be any more complicated than that.

120
by dryheat :: Thu, 12/23/2010 - 9:50am

I like where you're going with this. To grossly simplify, one could say that given the current rules of the NFL, a great offense will usually beat a great defense.

122
by DeltaWhiskey :: Thu, 12/23/2010 - 10:04am

Way to suck the winds out of a perfectly pompous and pointless discussion with valid points.

123
by Kevin from Philly :: Thu, 12/23/2010 - 10:28am

Don't put it that way. A little pre-Christmas wonking is a beautiful thing. Makes me think of being back in college, sweating out a Stat final.

50
by The Other Ben Johnson (not verified) :: Wed, 12/22/2010 - 5:08pm

is clearly ranked because . is way better than this.

54
by Aaron Schatz :: Wed, 12/22/2010 - 5:17pm

We forgot to do the odds on a 7-9 team winning the NFC West: Now 66.7 percent.

56
by Athelas :: Wed, 12/22/2010 - 5:20pm

Oy.

61
by Chip Paint (not verified) :: Wed, 12/22/2010 - 5:26pm

A 5-9 team has a 1 in 3 chance of making the playoffs. G-d bless the NFCW.

63
by Jetspete :: Wed, 12/22/2010 - 5:36pm

i just find it humorous that of games coming up in next two weeks, one de facto playoff game involves the 6-7 dvoa teams, and the other involves the 28-30 dvoa teams.

66
by Anonymus (not verified) :: Wed, 12/22/2010 - 6:03pm

The final 3 weeks should schedule only divisional matchups!

65
by Aaron Schatz :: Wed, 12/22/2010 - 5:40pm

All individual stats pages now updated.

67
by a fake Rodney Harrison (not verified) :: Wed, 12/22/2010 - 6:24pm

According to the playoff odds, there is slight but mathematically possible 0,3% chance that NE comes in 6th in the AFC.

Do not get it: if NE does not take the East, the Jets do, then BAL and PIT are the only other possible wild card 12 wins teams, should that be the case one of them will be the division champ of the North, the other one (actually both) lost against NE which is first tiebreaker.

Any hint how it is possible?

77
by MJK :: Wed, 12/22/2010 - 7:29pm

Let's see...

The only way that NE would not win the AFCE is if they lose their next 2, finishing 12-4. BAL and PIT can each finish with at most a 12-4 record, and both have lost to NE, so for NE to end up as the #6 seed, one of those two would have to miss the playoffs.

It is certainly possible that either PIT or BAL could miss the playoffs...but then the other wildcard would have to be either Indy, Jacksonville, KC, or San Diego. However, none of these teams can finish better than 11-5 (most can't finish better than 10-6), meaning that NE would still get seeded higher than them.

You're right...I don't see how it is possible for NE to end up with the #6 seed... If they somehow lose their division, they are locked into the #5 seed...

Maybe there's a bug in the playoff odds calculator...

125
by Kevin from Philly :: Thu, 12/23/2010 - 10:32am

Maybe the FO supercomputer knows something about Spygate that we don't?

79
by Joseph :: Wed, 12/22/2010 - 7:39pm

Playoff odds don't take into account tiebreakers.

97
by Mr Shush :: Wed, 12/22/2010 - 10:30pm

Indeed. A pretty substantial failing, frankly.

110
by RickD :: Thu, 12/23/2010 - 1:17am

They don't? Are you sure of that?

Explain how the Steelers are winning the AFC North 7/9 of the time.

I suspect you're wrong.

Now it's certainly possible that the simulator doesn't correctly account for tiebreakers (and I've argued above that it is doing so w.r.t. the Jets' odds for getting the #5 seed).

Coding tiebreakers should not be beyond the task at hand here. And really, if they are going to simulate the playoffs themselves, it seems silly to get the tiebreakers wrong. It's not like calculating the tiebreakers would be such a computational hurdle to jump.

68
by Raiderjoe :: Wed, 12/22/2010 - 6:28pm

Nfc West good chance all 4 temas finish witj more pints allowed tban scored. Would. be 4tb time happen.

Others- 2006 nfc west
2004 nfc wedt
1978 nfc cdntral

86
by dryheat :: Wed, 12/22/2010 - 8:57pm

Wow....we are fortunate to be living in such times. We are all witnesses to the near decade-long putridity of the NFC West/Wedt.

88
by Raiderjoe :: Wed, 12/22/2010 - 9:01pm

yeah one thingabouit it is nfc west now wuth 4 teams but 78 nfc cemtral was 5-teamer. all of yhem outscored by oppoknents for season. kind of amazing when thijl about it

98
by Mr Shush :: Wed, 12/22/2010 - 10:33pm

Amazing. But not as amazing as how awful the 2004 NFC West was. Honestly, that was just something else. I really think DVOA is missing something about the Rams, who for my money are an ok team.

Not good, I hasten to add. Just ok. At least as good as the 2004 Seahawks, for whatever little that's worth.

84
by bigtencrazy (not verified) :: Wed, 12/22/2010 - 8:47pm

I really struggle accepting Minnesota as an above average defense.

I know there is sporadic ability. But that secondary! Yowsers

116
by DeltaWhiskey :: Thu, 12/23/2010 - 9:07am

I say they're in the AVG range.

111
by RickD :: Thu, 12/23/2010 - 1:24am

I'm wondering who won the AFC East the .4% of the time it was neither the Pats nor the Jets.

These are simulations, not measurements, right? There shouldn't be problems with round-off error. Not of this magnitude.

131
by Travis :: Thu, 12/23/2010 - 12:04pm

Are you looking at the #1 seed, or the AFC East? The Ravens win the #1 seed the other 0.4% of the time neither the Patriots nor Jets do, while the AFC East odds do total 100.0%.

134
by RickD :: Thu, 12/23/2010 - 2:39pm

My bad - I was looking at the #1 seed. I knew that the Pats could not get the #2 seed (since they've beaten both the Ravens and the Steelers) but forgot that the Jets could.