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. 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 | 0.85 | 0.49 | 1 | 7 |
2 | NO | 2 | 0.75 | 0.50 | 2 | 12 |
3 | DEN | 3 | 0.70 | 0.38 | 7 | 5 |
4 | NYG | 5 | 0.68 | 0.40 | 6 | 4 |
5 | PHI | 4 | 0.67 | 0.46 | 12 | 3 |
6 | PIT | 10 | 0.65 | 0.52 | 4 | 14 |
7 | JAC | 12 | 0.63 | 0.62 | 10 | 13 |
8 | NYJ | 7 | 0.59 | 0.58 | 24 | 1 |
9 | TEN | 11 | 0.58 | 0.60 | 16 | 9 |
10 | BAL | 6 | 0.56 | 0.39 | 11 | 21 |
11 | DAL | 8 | 0.55 | 0.50 | 3 | 30 |
12 | CHI | 14 | 0.54 | 0.47 | 22 | 8 |
13 | ARI | 15 | 0.53 | 0.66 | 15 | 6 |
14 | HOU | 19 | 0.53 | 0.51 | 8 | 31 |
15 | SD | 9 | 0.52 | 0.47 | 5 | 28 |
16 | GB | 13 | 0.51 | 0.42 | 9 | 26 |
17 | SF | 18 | 0.50 | 0.41 | 25 | 11 |
18 | NE | 21 | 0.48 | 0.51 | 14 | 18 |
19 | ATL | 16 | 0.48 | 0.40 | 13 | 32 |
20 | MIN | 22 | 0.48 | 0.40 | 18 | 17 |
21 | WAS | 17 | 0.45 | 0.38 | 19 | 23 |
22 | CIN | 23 | 0.45 | 0.53 | 21 | 16 |
23 | BUF | 20 | 0.42 | 0.48 | 23 | 20 |
24 | SEA | 24 | 0.40 | 0.53 | 17 | 22 |
25 | MIA | 25 | 0.40 | 0.57 | 28 | 2 |
26 | CAR | 27 | 0.33 | 0.57 | 31 | 10 |
27 | DET | 30 | 0.31 | 0.56 | 20 | 27 |
28 | KC | 26 | 0.30 | 0.54 | 26 | 24 |
29 | TB | 29 | 0.29 | 0.53 | 27 | 25 |
30 | CLE | 32 | 0.27 | 0.55 | 30 | 19 |
31 | OAK | 28 | 0.26 | 0.51 | 32 | 15 |
32 | STL | 31 | 0.24 | 0.47 | 29 | 29 |
TEAM | OPASS | ORUN | OINT% | OFUM% | DPASS | DRUN | DINT% | PENRATE |
ARI | 6.0 | 3.2 | 3.1 | 2.7 | 7.2 | 3.1 | 1.8 | 0.45 |
ATL | 6.8 | 3.4 | 1.1 | 1.4 | 6.2 | 4.7 | 1.8 | 0.34 |
BAL | 6.7 | 5.0 | 1.9 | 0.9 | 6.7 | 2.6 | 4.7 | 0.53 |
BUF | 5.4 | 4.8 | 4.2 | 0.6 | 5.5 | 4.8 | 1.9 | 0.51 |
CAR | 4.9 | 4.3 | 7.4 | 2.1 | 5.8 | 5.4 | 2.2 | 0.33 |
CHI | 6.2 | 3.8 | 3.9 | 1.1 | 5.4 | 3.8 | 1.9 | 0.40 |
CIN | 5.4 | 4.5 | 3.6 | 1.5 | 6.1 | 4.2 | 1.4 | 0.40 |
CLE | 4.5 | 3.7 | 5.0 | 1.0 | 6.2 | 5.4 | 0.7 | 0.38 |
DAL | 6.8 | 6.0 | 3.1 | 0.0 | 6.9 | 4.7 | 1.4 | 0.49 |
DEN | 7.1 | 4.7 | 0.0 | 1.5 | 4.6 | 3.2 | 4.7 | 0.41 |
DET | 5.4 | 3.5 | 4.0 | 0.9 | 7.0 | 5.2 | 1.5 | 0.46 |
GB | 6.6 | 4.2 | 0.8 | 0.0 | 6.8 | 3.5 | 5.6 | 0.46 |
HOU | 7.3 | 3.2 | 2.3 | 2.1 | 6.2 | 5.5 | 1.6 | 0.29 |
IND | 9.5 | 3.5 | 2.2 | 1.5 | 4.7 | 4.0 | 1.9 | 0.30 |
JAC | 6.3 | 4.9 | 0.7 | 1.0 | 7.2 | 4.0 | 2.6 | 0.28 |
KC | 4.6 | 3.7 | 1.8 | 2.6 | 7.0 | 4.1 | 1.4 | 0.45 |
MIA | 4.3 | 5.0 | 2.6 | 0.9 | 7.4 | 2.9 | 2.5 | 0.27 |
MIN | 5.8 | 4.2 | 0.8 | 1.4 | 5.9 | 3.7 | 3.9 | 0.29 |
NE | 6.2 | 3.7 | 1.1 | 0.0 | 5.9 | 4.5 | 0.8 | 0.39 |
NO | 7.5 | 5.0 | 1.6 | 1.3 | 5.2 | 3.7 | 6.6 | 0.42 |
NYG | 7.7 | 4.2 | 1.5 | 0.4 | 3.8 | 5.5 | 4.4 | 0.38 |
NYJ | 5.6 | 4.0 | 4.5 | 1.5 | 4.6 | 4.2 | 2.7 | 0.36 |
OAK | 4.1 | 3.5 | 3.6 | 1.9 | 6.5 | 4.0 | 3.3 | 0.40 |
PHI | 6.6 | 4.4 | 3.4 | 1.3 | 4.8 | 3.6 | 6.9 | 0.37 |
PIT | 7.3 | 3.9 | 2.8 | 0.4 | 5.7 | 3.6 | 0.7 | 0.46 |
SD | 7.4 | 2.7 | 2.0 | 0.0 | 6.6 | 4.6 | 3.2 | 0.37 |
SF | 5.2 | 4.0 | 0.9 | 1.1 | 5.1 | 3.1 | 3.2 | 0.36 |
SEA | 5.5 | 3.7 | 2.4 | 0.9 | 6.3 | 5.1 | 1.5 | 0.31 |
STL | 4.2 | 4.5 | 1.6 | 1.7 | 7.1 | 4.1 | 1.7 | 0.49 |
TB | 5.0 | 4.3 | 2.9 | 1.8 | 7.8 | 4.8 | 3.7 | 0.40 |
TEN | 5.6 | 5.5 | 3.9 | 0.5 | 7.1 | 2.8 | 2.0 | 0.33 |
WAS | 6.7 | 3.9 | 4.0 | 1.0 | 5.7 | 4.3 | 1.7 | 0.43 |
Avg | 6.1 | 4.2 | 2.7 | 1.2 | 6.1 | 4.1 | 2.7 | 0.39 |
Carolina went from #1 at the end of last year to to just plain bad...
Brian - I know that the relative ranking of teams based on week-to-week statistics can lead to some apparent anomalies so I normally don't say anything. But are you sure that the Titans numbers are right? According to your numbers, they actually ROSE two places after receiving a nasty beat-down from the Jags. I can see that some of this is caused by teams above them falling, but it is shocking that the Titans can still hold water under your regression model when they're 0-4 and coming off their worst loss of the season.
In doing similar research, I am finding it difficult to acheive significance for each of the specific offensive and defensive variables you have listed above. What significance level are you setting for the model to accept each of these variables at? Are you removing those that don't meet the significance level you set? Or are you including all of the variables, regardless of individual significance, at the expense of variance accounted for, to capture the true nature of team skills? Thanks.
TEN isn't that bad. They're playing just below average in terms of team efficiency, but they have had the 3rd toughest schedule so far. And that's partly why they went up instead of down. The Jags themselves went up, plus their other previous opponents had strong outings, improving TEN's rating.
Additionally, there were some teams ahead of them that fell. It's so early in the season, and the teams can be so tightly grouped, that a change up or down a couple of places doesn't mean much.
John-Except for run def, they were all sig at p=0.05. But I'm keeping that in the model because 1)I know it matters, at least a little; 2)It's marginally significant; 3) It's sign is as predicted; and 4) My linear model of season win totals shows significance for it.
"They're playing just below average in terms of team efficiency"
Hmmm, that gets me to thinking. Although I have a great deal of respect for your statistical approach, I would hypothesize that the model is somewhat skewed by a lot of "garbage time" plays.
Example: When you look at the stats for Jags-Titans it really doesn't look like the Titans did THAT badly. But I was there and I can tell you that the Titans were godawful. The only reason their stats (and, presumably, their efficiency) looks decent is because the whole 2nd half was literally garbage time. Similarly, the only reason that the Jags looked "decent" in the Jags-Cardinals game was because they enjoyed an entire 2nd half of garbage time.
It would be really cool if you could filter the stats according to the GWP. In other words, only calculate efficiency based upon plays that were made when neither team was above X% in GWP. (I would blindly guess that most plays made when either team is above 80% GWP are not strong indicators of EITHER team's true efficiency.)
What day are you shooting for each week this year to publish your head to head probabilities?
Wow, Indy's OPASS is off the charts. Have you considered doing another QB rating post?
Now that you have the Win Probability tool, you could actually rate every play on how much in increased the probability of a victory. This could potentially lead to a FO-type approach where every play contributes to a team's rating.
A more accurate rating would probably be found by rating the play by the WP-differential it would cause if it were the first play of a scoreless game. This may overrate garbage time and would penalize kneel-downs or desperation plays, though, so some more clever weighting formula based on whether the game was in doubt could help.
brian;
Your origional formula included def.int
then,you took them outduetothe fact that they werenot repeatable..
then i recall in a recent post you decided toput them back in. But, in this post you say that it only includes "offensive turnover rates"...where arewe at with def. int. ?? ( I knowyou don'tput in def ff right?...thanks Dan
sorry about last post..I need a new keyboard argh!
does winning have any value in ur rankings titans 0-4 #9, vikings 4-0 #20 patriots 3-1 #18... What?
i agree getting rid of the garbage time stats too the vikings given up at least 10-20 meaningless points when they were ahead big late
jackson....???
Brian's model is predictive (not just explanatory)
His model went 13 wins 1 tie last week..and has been consistently predicitve at a very high %
Best on the web!
50% of every result is luck ..therefore we can't trust our eyes or something as simplistic as win lost records after 4 weeks..
your short-sigted attitude is why Vegas is rich...
Dan
What do you think of Denver's O and D being rated similarly? Or just Denver's O being in the top ten? I've had the impression that Denver's been all D, all the time, and their O has only been sort of along for the ride.
This site is amazing! I'm learning so much about statistics?
Brian, you seem to suggest there is an adjustment for opponents when you calculate GWP, I thought that wasn't the case?
Trash time is a shortcoming. I freely admit that. But what I like about this model is that it is relatively simple. It just takes a team's core efficiency stats and weights them according to how well those stats predict winning.
Yes, opponent adjustments are baked in.
I talk a lot about luck that occurs outside of the core stats, such as bad snaps or timely penalties. But I'm guessing Denver's good fortune is showing up in their passing efficiency. The Stokely tip-catch-TD and the Marshall game-winning TD probably account for most of their off pass efficiency. I doubt they can do that week in and week out. But who knows?
Garbage time happens because one team played exceptionally well or because other team played exceptionally bad or (in many cases) both. If that was a fluke, it'll even up throughout season. If particular team constantly plays well (or constantly plays bad), then trashing will occur again, garbage time will happen again, so I think it is actually good to include garbage tiem from previous games in prediction of future outcomes.
I am surprised that Detroit isn't much worse in the defensive ratings. Their O is much improved, and they have held leads, won a game, and have won TOP, however, their D is worse than last year (and last year's 0-16 lions had the worst D of all time).
This year, the Lions let qbs comp 72%, get a rating of 118, andhave given up 12 passing TDs. Their run D is giving up more yards per play than last year (although there are fewer runs against the lions this year).
So, it is shocking to see them in the mid 20s in the D categories, and not in the 30s.
Brian, I tried to duplicate your numbers, using the 7 coefficients you had listed some time ago (I don't recall a Def int perc rate coeff as is listed above). I checked a few teams, and are u sure the vikings (among others) are correct?
I know they have played a weak schedule, and I have basically what you have for their OPP GWP (40%). using the stats you have above with these coeffs for each of them reading across (0.46 0.25 -19.4 -19.4 -0.62 -0.25 -1.53) skipping the d int %, I got an average rating of -2.32, with minnesota -1.74. this gives them (correct me if I am wrong) an e to the (-1.74+2.33) odds of winning (1.79 to 1), thus giving them about a 64% chance of winning before taking into account schedule strength. I must be doing something wrong because opp GWP of 40% should not take them to 48% GWP. or can any readers verify or disupte my findings? thanks
keep up the great posts..
The coefficients are a little different now. I added 2 more seasons to the database. I also reintroduced def ints at a mitigated level.
All in all, MIN is nearly average or below average in all their team eff stats except interceptions, which are not as predictive as the other stats. They have the 6th weakest schedule so far, so that puts their GWP at 0.48, just slightly below avg. I think that makes sense, but I agree it doesn't jive with a 4-0 record.
One other thing I should mention, which would effect MIN more than other teams--Early in the season, I regress each team stat. For example, we shouldn't expect IND to sustain a 10 YPA average all year. It's bound to fall back down to Earth from here out.
Some stats regress more than others. The turnover rates are the least consistent and most random, and they get regressed the most. So right now, MIN's strengths are their low off int rate and high def int rate, which are very unlikely to be sustained going forward. That, in addition to the weak opponent strength, is why MIN doesn't get ranked higher.
Brian...
just to clarify..are you including def.int now?
are forced fumbles still out?
could you post the new coff. when you get the chance?
2) How long into the season do you regress stats
to account for high variance?
thanks
Def ints are in, with reduced weight. Forced fumbles are not in.
Each stat is regressed differently. Some "converge" on steady state variances quicker than others. For example, running stats take 5 weeks, but def ints take 9 weeks.