That brings us to the first team efficiency rankings of the 2013 season. For those who need a refresher, these rankings consider passing, running, turnover and penalty efficiency to create a logistic regression model. Using these values of team efficiency, we can determine the Generic Win Probability (GWP), a theoretical measure of a team's long-term true winning percentage. For more details, check out Brian Burke's explanation from the start of last season.
Again, three games is hardly a significant sample size. And as you might expect, that has led to some wacky results thus far.
Surprises Near the Top
No one should be surprised by the top three, as the Seahawks, Broncos and Saints are all undefeated and universally considered contenders. But just below them are three teams who few consider true contenders: the Lions, Jets and Panthers.
The Lions and Jets have excelled in one dimension of the passing game, an indispensable quality in today's NFL. Detroit possesses the third-ranked passing attack, while the Jets are one of three passing defenses (the Chiefs and Seahawks being the others) that have separated themselves thus far.
The Panthers are a little trickier, as despite middling rankings in nearly every stat, they've generated the 12th-ranked offense and 8th-ranked defense in terms of Expected Points Added. Those stats are different from the efficiency-based ones listed here, and may be skewed by Carolina's demolition of the Giants last week. The Panthers are off in Week 4, so we'll have to wait a little longer to see if Cam Newton and co. are capable of delivering a winning season.
You Are What Your Record Says You Are...Unless You're Not
The undefeated Bears and winless Giants are ranked 11th and 27th this week. That is, Chicago is 27th, and New York is 11th.
As befuddling as this sounds, the peripheral stats do paint the Giants as an unlucky team. Eli Manning's stratospheric 7.6 interception percentage will surely come down; as Grantland's Bill Barnwell details, the quarterback has run into a glut of fluky turnovers. Moreover, the G-Men possess an above-average passing attack and an excellent penalty rate, and are effectively average at everything else. As horrific as the Panthers game was, the Giants are certainly not dead, especially in the mediocre NFC East.
Meanwhile, the Bears have defeated the winless Vikings and Steelers, and Chicago was an underdog for most of the fourth quarter against Minnesota. They are essentially league average or worse in every category besides pass defense and interception rate, two hallmarks of recent seasons. The Bears had eight defensive touchdowns in 2012, and while you might expect that to come down, they've already amassed three this season. If those extra points ever dry up, the Bears might find themselves on the fringes of a playoff spot once again.
Super Bowl Chumps?
Seven months ago, the Ravens and 49ers played for a championship. But both are off to inauspicious starts, a perception reflected in these rankings.
The Ravens have a passable 2-1 record, but everything screams regression. Baltimore has the eighth- and second-worst passing and rushing attacks respectively, the latter a reflection on Ray Rice's putrid start. Their defense has been a neutral presence, and seems unlikely to prove capable of carrying a significant load. The 30-9 win over the Texans looks impressive, but subtract the interception and punt return touchdowns and you get a game much like the ugly 14-6 slugfest they played against Cleveland.
On the other hand, after Colin Kaepernick looked like a cross between Peyton Manning, Randall Cunningham and Thor against the Packers, he and the 49ers offense have been considerably subdued the past two weeks. Big deficits and Frank Gore's ineffectual performances have put too much pressure on Kaepernick, who still looks a bit raw trying to process coverages. San Francisco has too much talent to collapse, even with the Aldon Smith saga, but even with the small sample size caveat, it's fair to wonder if the 49ers are still among the NFC elite.
Again, just to reiterate one last time: these rankings are by no means the complete picture yet. Many of our perceptions are still based on 2012, so as 2013 wears on, these rankings should feel a little less surprising. But for now, here are the inaugural rankings of the season:
RANK | TEAM | LAST WK | GWP | Opp GWP | O RANK | D RANK |
1 | SEA | -- | 0.74 | 0.45 | 3 | 4 |
2 | DEN | -- | 0.70 | 0.45 | 1 | 20 |
3 | NO | -- | 0.64 | 0.51 | 9 | 8 |
4 | IND | -- | 0.60 | 0.48 | 11 | 24 |
5 | DET | -- | 0.59 | 0.38 | 5 | 16 |
6 | ATL | -- | 0.58 | 0.52 | 4 | 26 |
7 | NYJ | -- | 0.57 | 0.52 | 7 | 10 |
8 | CAR | -- | 0.56 | 0.60 | 13 | 14 |
9 | KC | -- | 0.55 | 0.47 | 26 | 3 |
10 | MIA | -- | 0.55 | 0.55 | 17 | 12 |
11 | NYG | -- | 0.54 | 0.60 | 21 | 9 |
12 | NE | -- | 0.53 | 0.53 | 18 | 11 |
13 | DAL | -- | 0.53 | 0.49 | 16 | 18 |
14 | PHI | -- | 0.52 | 0.44 | 8 | 25 |
15 | TB | -- | 0.51 | 0.58 | 14 | 5 |
16 | BUF | -- | 0.51 | 0.55 | 10 | 23 |
17 | HOU | -- | 0.49 | 0.42 | 30 | 1 |
18 | GB | -- | 0.47 | 0.42 | 6 | 30 |
19 | CIN | -- | 0.47 | 0.44 | 25 | 6 |
20 | CLE | -- | 0.46 | 0.44 | 32 | 15 |
21 | TEN | -- | 0.46 | 0.45 | 23 | 7 |
22 | SF | -- | 0.45 | 0.61 | 19 | 2 |
23 | OAK | -- | 0.45 | 0.55 | 15 | 28 |
24 | PIT | -- | 0.44 | 0.45 | 28 | 13 |
25 | SD | -- | 0.44 | 0.49 | 2 | 32 |
26 | ARI | -- | 0.43 | 0.53 | 22 | 19 |
27 | CHI | -- | 0.41 | 0.43 | 12 | 31 |
28 | MIN | -- | 0.39 | 0.49 | 27 | 21 |
29 | STL | -- | 0.37 | 0.52 | 29 | 29 |
30 | BAL | -- | 0.37 | 0.55 | 20 | 22 |
31 | JAC | -- | 0.35 | 0.58 | 31 | 17 |
32 | WAS | -- | 0.32 | 0.53 | 24 | 27 |
TEAM | OPASS | ORUNSR% | OINT% | OFUM% | DPASS | DRUNSR% | DINT% | PENRATE |
ARI | 5.9 | 41 | 3.5 | 1.8 | 7.1 | 63 | 1.7 | 0.33 |
ATL | 7.0 | 44 | 1.7 | 1.5 | 6.7 | 59 | 2.4 | 0.35 |
BAL | 5.5 | 27 | 1.7 | 1.8 | 6.5 | 55 | 0.9 | 0.45 |
BUF | 5.5 | 41 | 0.9 | 2.7 | 6.1 | 59 | 3.4 | 0.49 |
CAR | 5.3 | 46 | 2.3 | 1.4 | 6.2 | 56 | 3.0 | 0.28 |
CHI | 6.5 | 38 | 3.0 | 1.7 | 8.1 | 60 | 4.8 | 0.32 |
CIN | 7.0 | 39 | 2.8 | 8.9 | 5.8 | 59 | 3.5 | 0.55 |
CLE | 4.7 | 41 | 4.1 | 1.3 | 5.2 | 66 | 1.8 | 0.32 |
DAL | 6.0 | 36 | 0.9 | 3.4 | 5.9 | 70 | 2.4 | 0.36 |
DEN | 8.9 | 43 | 0.0 | 4.1 | 6.5 | 70 | 4.2 | 0.55 |
DET | 8.2 | 39 | 1.7 | 0.0 | 6.3 | 67 | 4.2 | 0.66 |
GB | 7.4 | 48 | 2.5 | 2.6 | 8.2 | 61 | 1.9 | 0.35 |
HOU | 5.8 | 40 | 3.1 | 0.0 | 5.2 | 71 | 1.2 | 0.58 |
IND | 6.1 | 51 | 1.1 | 0.0 | 6.4 | 52 | 3.3 | 0.14 |
JAC | 4.1 | 34 | 3.4 | 3.0 | 6.5 | 59 | 1.1 | 0.37 |
KC | 5.4 | 37 | 0.0 | 0.0 | 4.3 | 50 | 3.5 | 0.33 |
MIA | 6.2 | 34 | 1.9 | 3.6 | 5.5 | 47 | 3.7 | 0.23 |
MIN | 5.9 | 37 | 5.0 | 3.8 | 6.8 | 51 | 4.4 | 0.27 |
NE | 4.9 | 37 | 1.6 | 0.0 | 5.1 | 55 | 3.9 | 0.27 |
NO | 7.0 | 36 | 3.1 | 0.0 | 5.4 | 52 | 4.2 | 0.32 |
NYG | 6.8 | 27 | 7.6 | 2.0 | 6.2 | 62 | 1.7 | 0.28 |
NYJ | 6.6 | 38 | 5.8 | 1.4 | 4.6 | 70 | 0.9 | 0.68 |
OAK | 6.7 | 33 | 2.4 | 2.0 | 6.8 | 52 | 0.0 | 0.44 |
PHI | 7.4 | 54 | 2.2 | 6.3 | 6.9 | 55 | 1.5 | 0.44 |
PIT | 6.4 | 25 | 3.6 | 6.2 | 5.6 | 64 | 0.0 | 0.36 |
SD | 7.3 | 45 | 1.0 | 1.9 | 8.2 | 48 | 0.8 | 0.34 |
SF | 6.4 | 35 | 4.3 | 1.9 | 6.6 | 60 | 2.4 | 0.68 |
SEA | 8.3 | 40 | 2.5 | 1.5 | 4.5 | 59 | 5.6 | 0.57 |
STL | 5.7 | 31 | 1.4 | 1.5 | 7.6 | 62 | 0.9 | 0.39 |
TB | 5.2 | 38 | 3.2 | 0.0 | 5.4 | 64 | 3.3 | 0.68 |
TEN | 5.6 | 41 | 0.0 | 2.9 | 5.4 | 58 | 2.9 | 0.64 |
WAS | 6.4 | 44 | 2.9 | 4.1 | 8.5 | 55 | 0.9 | 0.48 |
Avg | 6.3 | 39 | 2.5 | 2.3 | 6.3 | 59 | 2.5 | 0.42 |
Since you're generating these from a logistic regression model, it would be nice if you would also report the confidence intervals around each team's GWP.
Tables don't sort?
One thing I have never understood is how important the strength of schedule adjustment is. It seems like you have the core stats (passing eff. run sr, etc) and then you have the SOS adjustment. What if the core stats you have collected have no dependence on opponent strength? What if they are completely dependent on team strength? It seems like the regression needs to be told how important the opponent strength is. Is this accounted for in some way?
tables not sorting
Why does NE have a 0% OFUM%? They had a fumble week 1.
Would it be a good idea to give last year's performance in these stats some weight that slowly decreases as the season wears on?
Its pretty obvious that, while run success rate doesn't usually regress much to the mean, Baltimore won't keep a 27% SR, and will probably wind up having one a little bit closer to last season's.
I don't think you can incorporate last year's stats because so many players change teams. You have five teams with new starting QBs so far (six if you count the Browns with Hoyer; seven once Glennon starts for Tampa next week). Numerous teams have changed their coaching staff and offensive and defensive schemes.
I believe ANS does not weight early season stats any differently from late season stats, however. Personnel changes happen in-season and teams can change, so I feel like early season stats should be weighted less, even if only by a little bit. I've never seen an explanation as to why this isn't done.
@Anon
It's not done because it doesn't affect the performance of the model. I saw a post by Brian that explained this last year. Given no objective reason to do so, he won't do so. "so I feel like" won't cut it. I like this, it weighs the objective information the best way possible and allows us, the thinking readers, to assess how much personnel changes and injuries should affect the teams going forward
@Anon
It's not done because it doesn't affect the performance of the model. I saw a post by Brian that explained this last year. Given no objective reason to do so, he won't do so. "so I feel like" won't cut it. I like this, it weighs the objective information the best way possible and allows us, the thinking readers, to assess how much personnel changes and injuries should affect the teams going forward
What would be the best way to use these rankings for office pool purposes?
how is strength of schedule measured? i'm assuming you just take the average of opponent's GWPs. Wondering if it would be more accurate to use offense vs. defense (and vice versa). For example, for Seattle, weigh their offense with respect to the strength 3 of the defenses they've played (and the reverse for their defense). instead of comparing team to team. Would that make it more accurate?
Does GWP already account for Opponent's GWP or do would we need to adjust your GWP variable for Strength of schedule (your Opponent's GWP) to come up with a True GWP?
Should we be able to replicate your calculations for GWP using the data you present? Either way,
1. Would it be possible to obtain the full model including coefficients and constant?
2. Is there other data you collect and consider for the model, but reject as insignificant?
@Steve
The GWP already accounts the Opponent's GWP.
I'll be honest, some of you stat guys really missed some forest for the trees with Kaepernick. His 79% completion rate in the 21-to-30 yard splits were a huge fluke. So while he did have the streak, over the long-run nobody does that. Nobody has been anywhere close to that over a career.
So while the streak it made him 'look good' it also disguised a lot of his problems. Now he's (last I read) a respectable 40% and the offense, which was totally predicated on his big-plays when he was the starter last year, isn't making enough big plays to over-come his problems.
In fairness, Gore actually averaged two yards more per CARRY than Kaep did PASSING against the Colts. Not Frank's fault Roman decided to abandon the run in the second half.
MosesZD: Kap made plenty of big plays against Green Bay. He still has lots to learn when it comes to reading defenses, and has regressed a bit when it comes to putting touch on his passes, problems I think are related. But I suspect the Niners' struggles on offense have much more to do with the fact they are missing their two best WRs (not to mention their stud TE and all-everything LBs).