Efficiency Rankings - Week 9

The efficiency rankings are beginning to settle down approaching the midpoint of the season. Big movers include San Diego climbing from 8 to 5 and Arizona dropping like a rock from 11 to 19.

The Cards played poorly against a bad team, and most teams are tightly bunched around the league average. So it doesn't take much movement in GWP to drop a number of spots from the top to the bottom of the "pack."

There continues to be a correlation between opponent-adjusted team strength and strength-schedule. This indicates that the death of parity is greatly exaggerated. It seems that so far this year there have been a disproportionate number of games between the very best and very worst teams. Ironically, this may indicate parity is stronger than ever. The scheduling system is supposed to give the top teams two extra games against other division leaders, and the bottom teams two extra games against the other dwellers. If a dweller or two turns into an elite team, or if an elite team suddenly becomes a doormat, the current disparity is what we'd get. Besides, parity isn't just about keeping teams from having very good or very bad records in any one year. It's also about making sure they aren't the same teams year after year.


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
IND
1
0.83
0.40
3
8
2
NO
2
0.82
0.47
1
9
3
DEN
3
0.77
0.53
13
1
4
PIT
4
0.75
0.45
6
6
5
SD
8
0.71
0.45
4
22
6
DAL
5
0.71
0.44
2
25
7
NE
9
0.71
0.48
5
14
8
GB
6
0.68
0.42
9
7
9
PHI
14
0.67
0.39
14
3
10
MIN
12
0.63
0.47
10
20
11
NYG
7
0.62
0.44
8
19
12
BAL
13
0.62
0.54
11
15
13
CIN
10
0.62
0.57
15
11
14
NYJ
19
0.56
0.48
21
2
15
HOU
15
0.53
0.43
7
31
16
ATL
18
0.53
0.57
12
24
17
CHI
17
0.49
0.48
19
12
18
SF
22
0.47
0.52
22
4
19
ARI
11
0.42
0.52
18
10
20
JAC
16
0.42
0.46
16
30
21
SEA
21
0.42
0.51
23
13
22
MIA
20
0.42
0.62
17
17
23
TEN
23
0.39
0.60
20
16
24
CAR
26
0.36
0.45
28
5
25
WAS
24
0.36
0.35
24
21
26
BUF
25
0.33
0.46
29
18
27
STL
27
0.25
0.50
25
28
28
OAK
29
0.20
0.59
31
23
29
DET
28
0.20
0.57
27
26
30
KC
30
0.17
0.56
30
29
31
TB
31
0.16
0.54
26
32
32
CLE
32
0.14
0.61
32
27

And here are the sortable raw team efficiency stats. Passing, running, and penalties are in yards per relevant play. Fumbles and interception stats are in turnovers per relevant play.



TEAM
OPASS
ORUN
OINT%
OFUM%
DPASS
DRUN
DINT%
PENRATE
ARI
6.0
3.3
3.8
1.7
6.3
3.8
2.9
0.44
ATL
6.6
4.0
3.9
1.2
6.6
4.5
2.3
0.38
BAL
6.7
4.6
2.0
0.8
6.6
3.5
3.1
0.59
BUF
5.0
4.0
4.0
0.3
5.3
5.1
5.4
0.44
CAR
5.3
4.7
6.7
1.7
5.3
4.5
4.1
0.38
CHI
6.2
3.9
4.6
1.8
5.7
4.0
3.0
0.40
CIN
6.4
4.3
3.1
1.1
6.5
3.9
3.1
0.35
CLE
3.9
3.8
5.6
1.5
7.0
4.9
1.6
0.34
DAL
7.6
5.4
1.7
0.9
6.1
4.2
1.6
0.48
DEN
6.4
4.2
0.4
1.1
5.2
3.4
2.8
0.35
DET
4.9
3.8
4.5
1.1
7.1
4.8
2.1
0.46
GB
7.0
4.3
0.9
0.5
5.9
3.5
5.3
0.55
HOU
7.6
3.3
2.5
1.6
6.4
4.7
2.3
0.38
IND
8.2
3.7
1.5
0.8
4.6
4.5
2.8
0.33
JAC
5.8
5.4
2.1
0.9
7.1
4.3
2.1
0.29
KC
4.4
3.5
2.4
1.4
7.3
4.4
1.3
0.39
MIA
4.8
4.6
2.5
0.5
7.1
3.6
2.8
0.37
MIN
6.7
4.1
1.1
0.7
6.2
4.1
2.2
0.31
NE
7.0
4.1
1.4
0.3
5.4
4.5
3.3
0.39
NO
8.0
4.5
2.6
1.2
5.5
4.4
6.0
0.39
NYG
7.1
4.4
3.0
0.9
5.7
4.6
2.8
0.40
NYJ
5.7
4.8
5.1
1.4
4.8
4.0
2.7
0.38
OAK
4.3
3.6
4.4
0.9
6.9
4.5
2.2
0.35
PHI
6.4
4.8
1.7
0.9
5.0
3.7
5.7
0.45
PIT
7.6
4.0
2.6
1.6
5.4
3.8
1.9
0.38
SD
7.6
3.1
1.7
0.0
5.7
4.2
4.0
0.38
SF
5.3
4.3
1.9
0.7
6.3
3.2
2.3
0.45
SEA
5.5
3.5
1.9
1.4
6.0
4.2
1.2
0.31
STL
5.0
4.6
2.7
1.1
7.1
4.2
2.3
0.48
TB
4.8
4.2
4.6
2.0
7.8
4.7
4.4
0.38
TEN
5.1
5.5
4.1
1.8
7.0
4.3
2.2
0.38
WAS
5.8
3.9
3.2
1.5
5.4
3.9
1.5
0.41
Avg
6.1
4.2
2.9
1.1
6.1
4.2
2.9
0.40

  • 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

11 Responses to “Efficiency Rankings - Week 9”

  1. Anonymous says:

    Lots of surprises still. San Diego is really shocking to me - they're 4-3 with wins over OAKx2(28 ranking) KC (30) and MIA (22). They do have good losses, but it doesn't seem like 4-3 with a mediocre point differential and 4 home games played should be the #5.

    MIN is another surprise, especially with their relation to GB. - MIN is 7-1 with 5 road games played, a sweep over GB, a better point differential and a harder schedule. I realize this is statistically based, but is it possible your system undervalues empiric results, as opposed to a poll-based or power rankings system which overvalue results?

    As an aside, it seems like there is a strong negative correlation between GWP and oppGWP. I see a few possible explanations, Strength of schedule is underrated as a metric (in my opinion unlikely), or good teams are deflating their opponents stats. Indy beats every team they play by 30, so their opponents look really weak when its actually the case that Indy is very strong. Perhaps you could calculate team X's oppGWP as a function of only opponents games played against not-team X?

  2. willkoky says:

    I don't agree with much of the preivous comment, but one good point was that perhaps yards gained or prevented on the road should count more then yards gained or prevented at home. Perhaps there is higher correlations between those stats and winning when computed seperately then when averaged together?
    To answer the guy above and out of curiosity, which is more predictive of future results, existing in season records or your model's predictions?
    At how many games does your model turn the corner into very good predictor? How many games of sample size are required?

  3. Drew says:

    Brian,
    I tried your contact link but it didn't work so I thought I'd try to just post a comment and hope you read it.

    I was wondering if you have ever researched the importance of scoring on the first drive of a game or second half? There was a blog post on JSOnline about Green Bay's lack of success in this regard with Rodgers as QB instead of Favre.

    http://www.jsonline.com/blogs/sports/69055962.html

    I also heard that New Orleans has scored 6 TDs and had 1 FG in their opening drives this year. Does this success correlate to an increased win probability? My guess is that it's the fact that they are scoring and not that it's the opening drive. I also think that most teams that score a lot also tend to score on their first drives and that when you score doesn't have much to do with your probability of winning....but I'm not smart enough to do the research. Thanks ahead of time if you are able to write about this. Love the site.

  4. Happy says:

    I wonder if, taking your sortable statistics, screening DOE could be run on the stats to see if there are any interactions? Then if potential interactions are found a full DOE could be run on the stats surrounding the potential interactions?

    For example lets say team A has a decent DPass and a decent Drun plays team B with a great OPass and a terrible Orun. The bad Orun doesn't come into play very much though; team B will pass on every down. I have oversimplified this, but where I'm going with this is to ask whether there are interactions that can explain expectations that shift on the basis of matchups.

    You may already do this - I just thought I'd ask.

    - Happy

  5. Brian Burke says:

    I did a lot of interaction testing in 2006 and found that it didn't add much predictive power. I thought that maybe there would be special significance to a game with a mismatch in one or more phases. For example, is run defense suddenly very important when the opponent has a strong run offense? Or vice versa? The interactions didn't add much at all, so I tended toward the simpler model without the interactions.

  6. James says:
    This comment has been removed by the author.
  7. James says:

    Brian, you gave the Eagles a .54 probability of winning the game against the Cowboys this week, despite the Cowboys being having a slightly higher GWP. Clearly this is due to home field advantage for the Eagles, which makes sense. My question is, do you factor in the diminished home field advantage you found for divisional games into your calculations? Thanks!

  8. Brian Burke says:

    James-No. The strength of HFA ranges from about 0.09 for perfectly matched opponents (making it 41/59) to about 0.04 for mismatches. Although I have confidence in the research regarding a reduced HFA for familiar opponents, the size of the effect in practical terms is small.

  9. Happy says:

    Brian - Thanks for your response. Its definately better to have a simple system well executed than a complex system that is difficult to manage. So if I read you right you found that, by and large, the interactions could be ignored.

  10. Brian Burke says:

    I'd say that with the data I had in 2006 the interactions had very little effect. Perhaps with more data the results would be different. It's something I'd like to look at again in the future.

  11. Mister says:

    I wish the schedule had this scheme:
    Play every team in your conference and then play
    the comparable (based on last year's play) team
    in the other conference.

Leave a Reply