There's a ranking I never thought I'd see. Peyton Manning's offense is ranked 21st in the league in efficiency. Last week, I got a few questions about how that could be , and the answer is opponent strength. The Colts have put up some slightly above average passing numbers, but they've been against some very weak defenses, including HOU twice, JAC, WAS, and DEN. Even NE's defense is ranked 23rd, giving up 7.0 net yards per pass attempt.
The team rankings below are in terms of generic win probability. The GWP is the probability a team would beat the league average team at a neutral site. Each team's opponent's average GWP is also listed, which can be considered to-date strength of schedule, and all ratings include adjustments for opponent strength.
Offensive rank (ORANK) is offensive generic win probability, which is based on each team's offensive efficiency stats only. In other words, it's the team's GWP assuming it had a league-average defense. DRANK is is a team's generic win probability rank assuming it had a league-average offense.
GWP is based on a logistic regression model applied to current team stats. The model includes offensive and defensive passing and running efficiency, offensive turnover rates, defensive interception rates, and team penalty rates. If you're scratching your head wondering why a team is ranked where it is, just scroll down to the second table to see the stats of all 32 teams.
Click on the table headers to sort.
| RANK | TEAM | LAST WK | GWP | Opp GWP | O RANK | D RANK |
| 1 |  SD | 1 | 0.82 | 0.41 | 1 | 2 |
| 2 |  PIT | 2 | 0.78 | 0.52 | 4 | 1 |
| 3 |  GB | 4 | 0.74 | 0.53 | 5 | 8 |
| 4 |  PHI | 5 | 0.71 | 0.51 | 3 | 5 |
| 5 |  NYG | 3 | 0.68 | 0.49 | 7 | 6 |
| 6 |  MIA | 6 | 0.67 | 0.59 | 17 | 9 |
| 7 |  TEN | 7 | 0.64 | 0.57 | 12 | 4 |
| 8 |  BAL | 8 | 0.61 | 0.51 | 15 | 13 |
| 9 |  NE | 12 | 0.59 | 0.60 | 6 | 23 |
| 10 |  NYJ | 10 | 0.58 | 0.53 | 20 | 7 |
| 11 |  KC | 9 | 0.58 | 0.44 | 14 | 12 |
| 12 |  IND | 11 | 0.57 | 0.53 | 21 | 10 |
| 13 |  MIN | 13 | 0.54 | 0.54 | 18 | 16 |
| 14 |  CHI | 19 | 0.54 | 0.49 | 29 | 3 |
| 15 |  HOU | 16 | 0.53 | 0.55 | 2 | 31 |
| 16 |  NO | 14 | 0.52 | 0.40 | 11 | 18 |
| 17 |  DAL | 17 | 0.52 | 0.55 | 8 | 24 |
| 18 |  CLE | 15 | 0.50 | 0.54 | 22 | 14 |
| 19 |  ATL | 20 | 0.47 | 0.48 | 13 | 22 |
| 20 |  CIN | 18 | 0.45 | 0.53 | 25 | 21 |
| 21 |  WAS | 21 | 0.45 | 0.55 | 16 | 27 |
| 22 |  TB | 23 | 0.44 | 0.39 | 10 | 28 |
| 23 |  BUF | 26 | 0.42 | 0.55 | 24 | 25 |
| 24 |  JAC | 27 | 0.38 | 0.57 | 19 | 30 |
| 25 |  DEN | 22 | 0.37 | 0.51 | 9 | 32 |
| 26 |  SF | 25 | 0.35 | 0.41 | 23 | 17 |
| 27 |  DET | 28 | 0.34 | 0.54 | 28 | 19 |
| 28 |  OAK | 24 | 0.31 | 0.47 | 27 | 15 |
| 29 |  SEA | 29 | 0.27 | 0.42 | 26 | 26 |
| 30 |  STL | 31 | 0.23 | 0.38 | 31 | 20 |
| 31 |  CAR | 30 | 0.22 | 0.48 | 32 | 11 |
| 32 |  ARI | 32 | 0.18 | 0.44 | 30 | 29 |
And here are each team's efficiency stats.
| TEAM | OPASS | ORUN | OINT% | OFUM% | DPASS | DRUN | DINT% | PENRATE |
| ARI | 5.0 | 4.3 | 3.7 | 1.0 | 7.0 | 4.4 | 2.8 | 0.45 |
| ATL | 6.2 | 4.1 | 1.3 | 0.2 | 6.7 | 4.3 | 4.3 | 0.32 |
| BAL | 6.6 | 3.7 | 2.1 | 0.9 | 5.9 | 4.2 | 3.0 | 0.36 |
| BUF | 5.7 | 4.5 | 3.3 | 1.1 | 6.4 | 4.7 | 1.3 | 0.32 |
| CAR | 4.4 | 3.8 | 5.0 | 2.7 | 6.0 | 4.0 | 3.4 | 0.42 |
| CHI | 5.7 | 3.8 | 4.9 | 0.2 | 5.4 | 3.5 | 4.0 | 0.41 |
| CIN | 6.0 | 3.8 | 3.3 | 1.4 | 6.4 | 4.4 | 3.5 | 0.37 |
| CLE | 6.1 | 4.1 | 3.1 | 1.7 | 6.5 | 4.0 | 4.0 | 0.39 |
| DAL | 7.0 | 3.7 | 3.8 | 0.2 | 6.9 | 4.3 | 2.5 | 0.49 |
| DEN | 6.8 | 3.3 | 1.5 | 1.4 | 7.4 | 4.4 | 2.0 | 0.48 |
| DET | 5.6 | 3.5 | 2.5 | 1.0 | 6.2 | 4.6 | 2.8 | 0.56 |
| GB | 6.9 | 4.1 | 2.7 | 0.4 | 5.5 | 4.5 | 4.2 | 0.34 |
| HOU | 6.9 | 4.9 | 2.1 | 0.6 | 7.7 | 4.0 | 1.6 | 0.32 |
| IND | 6.7 | 3.6 | 1.6 | 0.7 | 6.0 | 5.0 | 2.7 | 0.36 |
| JAC | 6.3 | 4.4 | 5.1 | 1.1 | 7.9 | 4.2 | 2.8 | 0.35 |
| KC | 6.5 | 4.8 | 1.4 | 0.6 | 6.0 | 4.1 | 1.8 | 0.37 |
| MIA | 6.3 | 3.8 | 3.7 | 1.4 | 5.8 | 3.9 | 2.3 | 0.28 |
| MIN | 6.2 | 4.5 | 5.1 | 0.8 | 6.4 | 3.7 | 2.5 | 0.41 |
| NE | 6.6 | 4.2 | 1.5 | 0.2 | 7.0 | 4.1 | 3.3 | 0.39 |
| NO | 6.7 | 3.9 | 3.4 | 0.7 | 5.6 | 4.2 | 1.9 | 0.39 |
| NYG | 6.9 | 4.5 | 4.5 | 1.8 | 5.7 | 3.7 | 3.2 | 0.45 |
| NYJ | 6.2 | 4.4 | 2.1 | 1.5 | 5.9 | 3.5 | 1.4 | 0.52 |
| OAK | 5.5 | 4.8 | 3.5 | 1.2 | 5.9 | 4.6 | 1.7 | 0.67 |
| PHI | 6.7 | 5.4 | 1.2 | 0.6 | 5.7 | 4.0 | 5.4 | 0.63 |
| PIT | 7.1 | 4.1 | 3.0 | 1.0 | 5.7 | 2.9 | 3.4 | 0.45 |
| SD | 8.1 | 4.0 | 2.5 | 1.7 | 5.3 | 3.6 | 2.8 | 0.40 |
| SF | 6.1 | 3.9 | 3.3 | 0.9 | 6.3 | 3.7 | 2.4 | 0.54 |
| SEA | 5.9 | 3.5 | 2.9 | 0.4 | 6.8 | 3.9 | 2.3 | 0.47 |
| STL | 5.1 | 3.8 | 2.4 | 0.2 | 5.9 | 4.2 | 2.2 | 0.50 |
| TB | 6.4 | 4.3 | 1.6 | 0.8 | 6.1 | 4.8 | 4.8 | 0.41 |
| TEN | 6.4 | 4.5 | 2.6 | 1.1 | 5.9 | 3.9 | 3.7 | 0.54 |
| WAS | 6.3 | 4.1 | 3.3 | 1.1 | 6.7 | 5.1 | 2.5 | 0.35 |
| Avg | 6.3 | 4.1 | 2.9 | 1.0 | 6.3 | 4.1 | 2.9 | 0.43 |
Brian, you mention that your regression model includes offensive turnover rates and defensive interception rates. You've argued recently that interceptions are somewhat random, and I've seen the FO guys argue that fumble recovery is almost completely random. Given the amount of randomness inherent in these turnovers, is it fair to judge a team's ability based on turnovers?
(I'm assuming, based on their inclusion in the table, that you're not simply using these rates as a control for the regression outcome).
Fumble recoveries are not in the model at all. Offensive and defensive int rates, as well as offensive fumble rates are slightly predictive. They are included in the model, but are very heavily regressed.
Have you accounted for # of home/road games in your strength of schedule ratings?
No. At this point in the season, the biggest difference a 1-game imbalance could make is .007 in terms of GWP.
I still don't see how Washington's offense is better than Indianapolis's. I get the (dis)pleasure of watching them as you probably do also considering you live in the area. They have one of the worst 3rd down conversion% in NFL history. They are not a good offense.
All I can tell you is that IND is less efficient than WAS, accounting for opponent efficiency. Strange but true.
The article says "offensive turnover rates." I assumed you simply meant times when the ball is turned over, which would be a consequence of both fumbling and not recovering. I was actually going to suggest that fumbles might be a better component than turnovers, but it sounds like you're saying that it's actually fumbles, and not fumbles lost, that's in the model. Is that right?
Yes. The model assumes league-average recovery rates.
Thanks for explaining. Sorry to give you a hard time about something you're doing perfectly correctly. Just got confused by the wording.
No need to apologize. Those are excellent questions. I guess it's been a while since I explained all this stuff.
If one wants to compare the impact of success rate to that of efficiency (yds per play) in the passing game, at least on D, there'll never be a better example than Rex Ryan's D, last year to this.
Last year it was #1 in lowest pass completion pct allowed and #2 in fewest yards per completion allowed. Compound those and it was a true dominating pass D. Considering what it was the year before, with basically the same players, it was an amazing coaching job by Rex.
This year it is still #1 in lowest pass completion pct allowed (even lower than last year) but as of last week #31 in yds per completion. After giving up 20-per completion in the 4th Q collapse against Houston, 160 yards in one Q, I doubt if its numbers got better. And this is after reinforcing the DBs with Cormartie and a 1st-rd pick.
Rex's philosophy is to bet the house on D success rate more than any other coach I've ever seen. Everything is on the line to both stuff the run and stop the pass completion. Last year, amazingly, he did it while also keeping completed passes short somehow. But this year it looks like opposing OCs have figured out how to make him pay on the completions.
Point is, measure the effect on total D effectiveness of yds/completion falling from 2 to 31 holding all else constant, and one might see something. A pretty extreme natural experiment.
In regards to the Colts efficiency. They are really selling out on "success rate" as compared to efficiency. The Colts offense is #1 in turning a first down into a new series or TD. They are also #1 in total yards per drive.
What they have done is played a very conservative passing attack given how awful both their defense and special teams are due to the extreme number of injuries on that side of the ball. They can't afford to give up a pick or punt because most likely the other team will put points on the board in response.
Additionally the offensive line is absolutely terrible so there is very little time for a play down field to develop.
In my opinion the offense has still been pretty good - but not as great as normal.
So the Colts offense is inefficient, but still successful because of its consistency? Like the team that gets 3 yards per run every time (except not quite that good).
Brian, have you ever considered looking at the total yards gained in relation to the maximum possible yards? The Chargers are averaging the most yards on offense and the least yards on defense in the league, but I wonder how much of that is due to their terrible special teams play, particularly early in the season.
If a team gets one or two extra (or fewer) possessions in a game due to a KR, PR or INT for a TD it's extremely easy (or difficult) to rack up a lot of yards. Similarly, it's difficult if your defense gets a lot of turnovers close to your opponent's endzone.
Something like total yards/potential yards or yards/possession would seem more informed than simply total yards.
ChrisZ-Right on.
James-Perhaps EPA per play would be the appropriate measure for that. If your offense takes the field with 40 yards to go for a TD, their starting EP is going to be high, limiting their potential total EPA. Dividing over the # of plays would account for number of possessions.
Brian - can you explain why the Packers are ahead of the Eagles, even though the Eagles are ranked higher in Off and Def rank? I know the opponents are tougher for the Packers but I thought that was accounted for in the orank/drank? Even if it wasn't, does that small difference in opponent strength make that large difference?
Jim-It's their penalty rates. GB is 5th best and PHI is 2nd worst.
The Vikings have roughly the same efficiency rating as last year (.54 in 2009, .57 in 2010).
If this rating is accurate then the Vikings or 10 and 10 are the most glowing example of huge a factor luck plays in any one season.
If you take the two GWPs you come up with the Vikings having a 14.52 wins over the 2009 and 2010 regular seasons. They've actually won 15.
When I watch the two teams they don't look similar in quality. But the stats are hitting it bang on. The Vikings were just lucky last year and unlucky this year.
Mr. Burke,
Have you ever added up all the seasons to see which teams are the luckiest/unluckiest over a number of seasons.
I presume they all regress towards their predicted win levels. Have any teams failed to do so in a statistically significant way?
Yes, a long while ago, 2007 I think. All the teams regress back and forth around their GWP from year to year, except one--NE. They over-perform their efficiency nearly every year.
Favre-led teams have usually been very lucky too. But this year, the magic has run out.
Thanks Brian,
That Belichick is simply smarter than everyone else.
Brian,
Have you done much work trying to spot situations where your GWP model is consistently biased (e.g. your mention of the Pats and Favre-led teams above)? There's certainly a difference between "luck" and the sum of (a) random chance and (b) a statistical model's inability to capture sitation-specific or team-specific characteristics that may significantly alter WP. Referring to all of the unexplained variance as "luck" implies the model accounts for all team characteristics (no matter how complex and difficult to model) that are significant predictors of winning.
Since I believe your model is linear-regression-based, I'm wondering if it performs poorly when one of the predictor variables is an outlier relative to the data used to obtain parameter estimates. That is, if an independent variable has a non-linear effect on the dependent variable, it seems the difference between the model and reality would be more noticeable when the value of the independent variable greatly deviates from the "baseline" value of the independent variable.
A reader noted above that the Vikings have similar efficiency ratings as last year and surmised that the change in their record could be explained by a difference in luck.
As an avid Vikings fan, I find this very hard to believe. I have a hunch that Favre's INT rate is so high relative to the baseline data that the GWP model isn't capturing how detrimental this has been. It seems weird that their efficiency and GWP stats would be so similar when Favre's EPA and WPA numbers have taken such a brutal downturn (due to what appears to be a brutal downturn in accuracy and decision making skills).
The model is a logistic model, at least this incarnation. This model is tuned to be predictive rather than explanatory. Certainly Favre's interceptions this year have been a very large cause of the Vikings downturn since last year. But here, for this purpose, the question is, what is Favre's most likely interception rate going forward?
To answer your question, yes. Each year I have another 256 games of data and re-tune the coefficients. There are outliers and there always will be, so I don't over-react to them. But when I do see some systematic biases, I dig deeper. One of the big things I learned this past year was about run SR, so I plan to incorporate that in future versions.
Thanks for the response Brian. I think an article on the outliers, or "teams that make their own luck" / "defy the odds", would be very interesting.
Interesting to note that the Chargers are ripping out the highest GWP vs one of the softest schedules according to their vsGWP.
And they are only 6-5. Lucky plays be damned, you are getting 10 games vs the AFC and NFC West.
You are mediocre if you can't crush those tomato cans.