Week 4 Team Efficiency Rankings

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, and team penalty rates. A full explanation of the methodology can be found here.



RANKTEAMLAST WKGWPOpp GWPO RANKD RANK
1 CHI-0.720.5818
2 SD-0.710.55211
3 DAL-0.670.64101
4 KC-0.640.56153
5 GB-0.630.5754
6 SEA-0.620.57192
7 HOU-0.620.62619
8 IND-0.610.57712
9 PIT-
0.580.43165
10 WAS-0.580.511114
11 DEN-0.560.541326
12 PHI-0.550.491710
13 NYG-0.540.49418
14 NYJ-0.530.5189
15 BAL-0.520.50246
16 NE-0.520.46332
17 CLE-0.520.53925
18 MIA-0.500.461821
19 TEN-0.490.47127
20 DET-0.450.572222
21 SF-0.450.552524
22 CIN-0.440.472715
23 MIN-0.430.452813
24 TB-0.430.492030
25 BUF-0.420.553116
26 NO-0.400.422131
27 JAC-0.390.612928
28 ATL-0.370.421429
29 CAR-0.360.473217
30 OAK-0.290.342320
31 ARI-0.270.302623
32 STL-0.250.383027

Raw efficiency stats for each team are listed below.


TEAMOPASSORUNOINT%OFUM%DPASSDRUNDINT%PENRATE
ARI4.95.84.01.75.54.33.30.72
ATL6.14.00.90.56.95.16.20.35
BAL5.73.14.60.74.14.70.00.39
BUF4.84.25.00.07.14.20.00.32
CAR4.63.66.23.86.63.15.40.40
CHI8.43.22.20.06.32.13.10.42
CIN5.53.22.51.85.44.54.70.34
CLE6.24.53.22.96.53.83.40.51
DAL7.23.51.60.06.53.92.20.50
DEN8.12.51.61.27.13.73.00.43
DET5.63.04.10.77.05.02.90.40
GB6.83.82.90.74.75.03.80.63
HOU7.05.04.00.68.53.00.00.33
IND7.73.30.01.86.75.03.10.39
JAC4.93.75.41.38.74.63.20.40
KC6.14.63.90.06.03.21.80.28
MIA6.23.81.12.05.64.53.10.22
MIN5.45.06.20.75.83.62.10.57
NE7.44.62.00.06.94.42.80.38
NO7.02.61.80.67.14.43.40.30
NYG7.04.35.91.36.13.83.90.54
NYJ6.14.90.00.86.62.43.40.72
OAK5.24.53.71.25.14.62.90.77
PHI6.16.00.00.74.74.14.70.61
PIT6.24.75.02.15.22.64.30.22
SD8.14.43.31.95.33.95.20.36
SF5.53.94.20.07.23.92.30.43
SEA6.34.05.60.06.82.63.00.35
STL4.93.64.30.06.35.12.70.58
TB5.93.42.20.76.64.87.10.36
TEN5.94.04.41.34.84.42.90.63
WAS7.43.91.00.86.83.51.50.42
Avg6.34.03.21.06.34.03.20.45

  • Spread The Love
  • Digg This Post
  • Tweet This Post
  • Stumble This Post
  • Submit This Post To Delicious
  • Submit This Post To Reddit
  • Submit This Post To Mixx

17 Responses to “Week 4 Team Efficiency Rankings”

  1. Brian Burke says:

    The ratings don't make sense, yeah yeah. I feel the same way, as I do every year. But somehow, they end up being more right than wrong...

  2. Anonymous says:

    Being a Bronco fan, this makes much more sense than the power rankings I've been seeing!

    I've learned much on this site about how the underdog needs to play a higher variance game to win. That's exactly what McDaniels did, in my opinion....and the Broncos came closer to defeating the Colts than is recognized.

    The win probability graph of the game bears this out. The Colts were in control most of the game, to be sure....but in the past, the graph would shoot up to .95 wp for the Colts and stay there. Denver at one point in the fourth quarter actually had a .37 wp.

    JudoJohn

  3. Chuck Winkler says:

    Do you have this for the season to date?

  4. nottom says:

    Is there a reason you don't have any sort of Special Teams rating? Have you looked at it and decided it is non-predictive? Certainly SD's WP would look a lot different if their awful STs were accounted for.

  5. Adam Davis says:

    You're right, the ratings don't make sense. I'm a Jaguars season ticket holder and I can tell you based on empirical observation that they're not nearly as good as your efficiency ratings would suggest.

  6. Chris says:

    I compared the week 4 GWP to the end of season GWP for the 2009 and 2010 seasons. The standard deviation of the difference of the week 4 and end-of-season GWP is about 0.12, which would be roughly equivalent to two games per season (0.12*16 = 1.92). Teams that are ranked high or low in GWP at week 4 tend not to move as much as teams that are ranked in the middle of GWP. The distribution of the differences in week 4 and end-of-season GWP isn't quite normal; instead, it tends to be skewed toward possible values. The largest improvement of GWP over the course of a single season was 0.29 (more than 4.5 games) and the greatest decrease in GWP was -0.23 (more than 3.5 games).

    The message is to take these early season rankings with a good pinch of salt. The samples size is small and teams are going to move around a good bit. On the other hand, surprise teams like KC are probably no worse than an average team (which would be a good improvement for one year). Also, it is likely that NO will be dealing with their post-super-bowl slump all season long. I conclude this by assuming they have a historic improvement in GWP and move from 0.4 to 0.7. That still would not put them at the top of the league in terms of GWP (which is typically 0.75 or better).

  7. Chris says:

    "skewed toward *positive* values".

  8. Chris says:

    And of course I meant the 2009 and 2008 seasons. I wish I had the end of season GWP for 2010.

  9. Brian Burke says:

    Thanks, Chris. Awesome work. Pinch of salt to say the least. On the other hand, the purpose of GWP/the prediction model is to look to next week as much as to the end of the season.

  10. Brian Burke says:

    nottom-I've looked at ST ratings. It's very hard to actually assess ST performance to begin with. And ST is particularly random. There are relatively few ST plays each game, and each type is unique in its own way (FG, KO, punt). Plus, there's no telling if ST is going to be a difference maker in a game. Bottom line is that although ST can help explain past outcomes, ST is not very predictive of future outcomes.

  11. JG says:

    Oddities:

    Tenn is #12 on O and #7 on D, both above average, but has only a 49% GWP.

    The Jets are #8 on O and #9 on D, but have only a 53% GWP.

    Denver is #13 on O and only #26 on D, much worse than the Jets, but has a better 56% GWP.

  12. Brian Burke says:

    JG-The rankings can be very sensitive to packs of teams concentrated just above or below any particular team.

  13. Anonymous says:

    Looks like the OFUM% for CHI is wrong. They fumbled three times against Detroit in week 1, but you have them at 0.0%.

  14. Andy says:

    Brian, have you ever looked at the delta in rankings between week 3 and week 17 (or week 16 if that is more accurate)? Do teams actually tend to stay close to where they were ranked at week 3? I realize that the system does great at actually predicting games, I just wonder if the actual rankings can change a lot along the way.

    Im thinking you could use the same methods that you used for evaluating preseason picks, where the error (maybe error in GWP?) was squared? I dont remember the exact method.

  15. Brian Burke says:

    Andy-See Chris's comment above.

  16. Andy says:

    Oh, thanks

  17. Michael L says:

    Brian, I have to think that while seeking a composite ST efficiency score may be unhelpful, that a few ST metrics WOULD be useful.

    Average kickoff distance, KO touchback rate, average starting field position (on KO or all possessions) are all fairly consistent by team. Possibly Return Yardage Over Average/Play earned and allowed would help account for good returners/schemes as well as poor coverages. Obviously TDs, fumbles, and blocks are rare events, though respectively you could add a RYOA bonus, track STFM% and STFF%, and probably ignore blocks (though CHI's got to be sigmas above average DBLK%).

    IIRC you calculate WP by correlative analysis, not a simulation model right? So the factors I mentioned might be too fine-grained, but I think they could prove predictive...

Leave a Reply