Team Efficiency Rankings - Week 15

New England's offense claims the number one spot this week, but their early season defensive woes are holding them back.

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.

1 SD10.830.4421
2 NYG40.730.4793
3 PIT30.720.5172
4 GB20.720.4958
5 NE50.720.56123
6 PHI60.680.5139
7 MIA70.630.56207
8 BAL80.620.511110
9 IND90.600.551411
10 HOU140.580.57427
11 NYJ120.570.55264
12 TEN160.550.55225
13 KC110.540.471514
14 MIN100.530.542112
15 CHI130.530.51286
16 DAL170.510.56628
17 NO150.510.39820
18 CLE180.470.502415
19 BUF190.440.552524
20 ATL200.430.471325
21 TB220.430.411026
22 WAS210.420.551730
23 SF250.400.411816
24 OAK270.400.501617
25 CIN230.380.552922
26 DET280.360.572719
27 DEN240.340.471231
28 JAC260.340.561932
29 SEA290.300.432329
30 STL300.270.393018
31 CAR310.230.463213
32 ARI320.200.433121

And here are each team's efficiency stats.


  • 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

26 Responses to “Team Efficiency Rankings - Week 15”

  1. Anonymous says:

    ATL is number 20?! Outrage and scorn!

  2. Sampo says:


    You might want to check this post (unless that was sarcasm):

  3. Anonymous says:

    As the playoffs approach, would the predictive value of these stats be improved if we used the last n games 5<=n<=8 rather than the whole season?

  4. Brian Burke says:

    Predictively, it turns out that it's a wash. Over-weighting recent games effectively reduces sample size, so there's a trade-off.

  5. Anonymous says:

    Decreasing your sample size is almost surely going to weaken your predictive ability. Sure, some teams are performing better recently, some teams are performing worse recently, and some teams have no discerning trend.

    But unless you can point to a valid reason why, for example, New England's offensive and defensive prowess in the past two weeks is more indiciatve of their overall efficiency and thus more indicative of future performance INSTEAD of their combined efficiency for the whole season, then I don't think it's helpful to only consider part of the season.

    With regard to New England, their true efficiency probably lies somewhere inbetween their performance in the first half of the season and the second half. By only considering the second half, we're most likely overestimating their true quality. By only considering the first half, we're most likely underestimating it. Hence, we look at the entire season.

    Obviously, things like injuries do make looking at certain portions of the season more valid. If team A is #1 in offensive efficiency for 8 games, the quarterback gets injured, and then the team is last place in offensive efficiency for the next 7 games...incorporating the first half of the season to project the outcome of the last game probably isn't a good idea. The problem with adjusting the model to accomodate this is that it probably adds a lot of subjectivity. When do we decide to exclude games? How much worse does a replacement player need to be for us to not consider certain games? Perhaps it's an issue of weighting games differently, rather than an on/off switch of including them or not.

  6. Matt says:

    Can you explain the rankings of the Giants and Eagles?

    Giants ranked #2 have O rank 9 and D rank 3, with opponents GWP 0.47
    Eagles ranked #6 have O rank 3 and D rank 9, with opponents GWP 0.51

    so, same ranking but swapped against more difficult competition makes the eagles 4 spots worse? I'm actually a Giants fan, so I like the love... but I don't quite understand the algorithm behind these rankings.

  7. Brian Burke says:

    Sure. The difference is due to their penalty yards, which are, unfortunately, not tracked for offense and for defense but just for teams as a whole.

    PHI 0.59 Penalty yds/play
    NYG 0.41 Penalty yds/play

  8. Anonymous says:

    Or why Pittsburgh has a higher ranked offense AND defense AND have played a harder schedule, but is ranked below NYG.


  9. Anonymous says:

    I get why weighting more recent games more heavily doesn't improve things, but I have a related question. It seems plausible to me that teams with an above average % of "young" players starting would have a measurable tendency to improve more over the course of a season than a team with an average or below average % of "young" players. Does anyone know of any work that would support that (or not)?


  10. Anonymous says:

    It´s not funny,

    every week some anonymous guy comes over in a harsh tongue to complain about the Atlanta rankings. Why don´t you people go over to Prisco or Espn rankings? There you have "your" team at the top of lists. But if you want that, why don´t you just read the W-L column?

    I hope Brian don´t changes the model, because his rankings are the best so far.

    My prediction: Atlanta will not survive with luck until the Superbowl. At least that is what i hope for, because the best teams should win, not the luckiest ones. If i would want that, i could watch NHL.

    Greetings from Germany, Karl.

  11. Brian Burke says:

    Thank you, Karl.

    In defense of the critics, they have a good point. The model isn't perfect, so it's a good to ask what it is about Atlanta (and some other teams throughout the years) that their efficiency predicts a much different record than actual. The model is a pointer to look at things maybe we wouldn't usually look at.

    Is it their strength of schedule? Is it Success Rate? Is it turnovers? Is it 'clutch' play? Is it special teams? Simple sample error? What makes Atlanta a winner?

    It looks like 2 things to me, maybe 3.
    1)Clutch play, defined by Matt Ryan's insane WPA and its mismatch with his overall efficiency. 2)Turnovers--they're either really lucky or freakishly good with interceptions and fumbles on both sides of the ball.
    3)ST hasn't killed them. I don't see anything that jumps out at me that would say ST is helping them win games, but it hasn't hurt them either. (I'm looking at you San Diego.)

    Good or not, ATL will likely have home field advantage and probably a bye in the playoffs. They may not be the best team on paper, but they can still be the champion.

  12. Anonymous says:

    Hello Brian,

    before i read (most) of your articles, i knew passing efficiency is the most important factor in (NFL/NFL-Europe/WLAF)-Football. That came from my own studies. And your studies confirmed that.

    I had a feeling that because of the short NFL-Schedule, randomness (i.e. luck) and some other factors not yet discussed on this site, has the 2nd most impact on NFL-Success.

    I also had the feeling Clutch-Play is not existing. Studies confirmed that too.

    Turnovers are descriptive, but random and/or situational dependent (very short: Being behind leads to tons of Int´s or seldom comeback wins). Your studies confirmed that too.

    Conclusion: I am a little surprised that you believe in that "Clutch-Thing" now. Atlanta will have plenty of turnovers in the playoffs once they fall behind. Both things should happen, because it´s really time for regression to the mean (i.e. it´s time that luck evens out) :-)

    By saying all that: I truly hope we don´t have another 2001-Patriots or 2007-Giants season. That would really look like NHL or Soccer now. Please not!

    Cheers, Karl

  13. Anonymous says:

    Brian, I don't get how the bucs have a better offense than the colts.

  14. Brian Burke says:

    Bucs O vs. Colts O: TB has a much better running game and fewer turnovers. Turnovers are not completely random, and they are consistent to a measurable degree. Throw in defensive strength of opponent and you've got TB on top of IND.

    Karl-I'm not a believer that clutch is a persistent skill. But I am a believer that it exists. Some players will just happen to have more of their success concentrated in high leverage situations than other players.

  15. Anonymous says:

    Nuances... Brian,

    but i think you know what i meant. And i know what you mean. Clutch-Play does exist to it´s true english meaning... Sorry english is not my mother tongue.

    Turnovers are random in same situations. I tought again you know what i mean. I didn´t want to write a novel.

    Anyway, still think we are on the same page.


  16. Brian Burke says:

    Karl-Ihr Englisch ist viel besser als mein Deutsch.

  17. Anonymous says:

    Clutch exists in the sense that a last minute touchdown to put your team ahead is considered clutch, just as a homerun in the bottom of the 9th inning to win the game is considered clutch. An statistically significant ability for a player to perform better in these specific situations than in any/all other situations is what doesn't exist

  18. Anonymous says:

    Exactly that´s what i meant. I tought it was finally clear with my last post. I mean i didn´t talk in riddles, did i ??

    Back to the original point: ATL belongs to the 20th rank; Brian´s model is correct (from what i can see); there is no need of rankings if they are the same as the W-L-Column...

    After all that, it makes me wonder for what Prisco is paid for. A guy who thinks Y/Att. is going down when it´s going up; that Bushrod would have had no chance vs. Allen; and more nonsense i can´t write down here (it´s way too much).

    I hope for Brian that ATL will have an early exit in the playoffs, so that all the critics stop talking bad about this great site.

    Cheers, Karl.

  19. Anonymous says:

    1. Are QB scrambles and kneels in included in rushing efficiency? Seems like they should be removed. At least kneels. And it seems like scrambles should be given their own category.

    2. Does/would only using only first half efficiency data improve the predictiveness at all by removing garbage yards gained in blowouts or teams plowing into the line just to waste clock? Or would it change the rankings any while retaining the same predictiveness?

    Whether it is luck or some skill not being measured Atlanta is currently 7 games over their expected win total for the last two seasons.

    Last season they finished with a GWP of 0.46 against an opponents GWP of 0.56.

    They finished with 9 wins and would have only been expected to win 6.4

    This season they should have 6 wins based on their game winning probability yet have 11.

    So they are +7 wins over expected in their last 19 games.

  20. Andrew Foland says:

    In any binomial dataset of 32 sets of 13 trials, there are going to be outliers, in fact it would be suspicious if there weren't. So saying "there are outliers!" is not very useful.

    What is useful is to examine the question: are there more outliers than one would expect in a binomial process with the distribution of p_success given in the tables above?

    So I set out to answer this question.

    For each team I calculated an expected-wins-probability-against-their-schedule, given by GWP*(1-Sched)/[GWP*(1-Sched)+Sched*(1-GWP)]. (For reasons I don't understand, this is sometimes called the "log five" rule in baseball.) I then calculated the binomial probability of having observed their actual wins, given that probability in 13 trials. Take the negative log and sum over all teams; that's the log likelihood of the W-L data given the p_success estimates.

    One then performs a simulation. You take the same distribution of p_success, and run many "simulated seasons". For each simulated season instance, you perform the same log likelihood calculation described above.

    Finally, you look at the distribution of log-likelihoods among the simulations, and find where the actual season likelihood falls in that distribution. In particular, you ask, "what fraction of simulated seasons have log likelihoods as bad, or worse, than the observed season?" If this is a very low number, it is an indication that the p_success assigned is producing a bad model. If it's a higher number, it indicates the model is adequate to explain the data.

    (There is a caveat here, in that one should really, truly simulate the seasons, including the fact that every game has exactly one winner and one loser. I did not do so in what's quoted below. But doing so is straightforward, "just work".)

    This procedure produces a number equivalent to what is usually called the "P-value" of the model fit.

    So, long and short, I find that the P-value for the model in explaining this season's win-loss data is about 0.08 . I would characterize 0.08, in general, as maybe a tad low but not at all necessarily indicating a problem. I am fairly sure that this number would be larger for a fully correct simulation, so 0.08 is a conservative estimate in the sense of being toughest on the model. Values above 0.01 to 0.05 (depending on your field of practice) are generally considered acceptable.

  21. Brian Burke says:

    Andrew-Awesome. Not the kind of results I would hope for, but I appreciate your excellent analysis. This has been a down year for the efficiency model, at least so far.

    I wonder how much Atlanta and San Diego alone have confounded the model???

  22. Andrew Foland says:

    If you replace San Diego and Atlanta's actual log likelihoods with the average value for the other teams, the P-value is around 0.5 . So in some sense, yes, they are responsible for much of the model mismatch. On the other hand, one ought next to perform a meta-analysis, asking, "If I always throw out the worst two log-likelihoods, what is the distribution of log-likelihood sums?" It may not be unusual for two teams to generate much of the model mismatch.

    Also, don't forget that I didn't model the correlations among wins and losses in the league correctly, and I'm pretty sure that would increase the 0.08.

  23. Ian Simcox says:

    Andrew - I did have a look at the proper simulation issue not long ago and found that there wasn't a load (I didn't test statistically) of difference in the distribution of wins across teams whether you did a truly random simulation (i.e. 32 teams each playing 16 games, 0.5 WP) or whether you set up your simulation so that each simulation would have exactly 256 wins.

    So I agree that it would be better fixing it so that you had 256 wins in each season, but from a results perspective you won't be a million miles off.

  24. Adam D says:

    I remember reading a quote by the guy who came up with SimCity and The Sims, something to the effect of 'Models are really great for exposing your false assumptions.' That ties into what you were saying earlier Brian...

    I think it's at least possible that the atlanta coaches set up their play calling to get mismatches in clutch situations. Maybe every 3rd and long they run the ball up the middle until the last go ahead drive when they play action. Obviously I'm not a coach (judging by that horrid example), but I think something like that could be going on. As to whether or not it's a good idea to play (or playcall) below your ability in order to catch the defense off guard in the endgame... that's another question.

  25. Unknown says:


    I'm not looking to see rankings mirror the win-loss column, just to at least get to the point where the Falcons are not ranked lower than the Bills, and Brian, I appreciate you recognizing why critics might take exception to that.

    Of the three factors you pointed to Brian, I would say as a Falcons fan there doesn't seem to be any reason to believe Special Teams has artificially inflated our win total. If anything, it almost cost us a couple of games (SF and TB come to mind). A kickoff return did play a big role in winning @TB, but we were ranked 20th in the model long before that.

    Clutch play is something I don't want to get into, although I do have thoughts on the post Sampo linked to and why that certainly doesn't close the book on Atlanta's ranking and actual quality that I will comment on over at that post.

    Turnovers is an interesting one. One thing that the two teams with the best records and atop non-statistical power rankings have in common are that they have been extremely effective (a word I like much better than lucky) at preventing turnovers. Earlier this season the Falcons had a stretch of four straight games without a turnover and I remember hearing that the record was five. Now the Patriots have gone five straight games without one.

    To listen to analysts talk, the Patriots streak of five games without a turnover comes as no surprise because they are so well-coached. I understand that there is a degree of luck involved, but I also think that the ability to limit turnovers very much seems like a skill that can be acquired, as well as something that good coaching can contribute to.

    The results of your model and some of your past comments on the predictivity of turnover rates seem to disagree.

    Serious question: is there reason to think that your model may be underweighting the predictive abilities of turnover rates?

  26. Alex says:

    Can't wait until success rate is incorporated somehow. Indy is the best example of how using it could help - most would say 14th/11th undervalues their offense and overvalues their defense, and they turn out to be like 2nd-best in offensive success rate and league-worst in defensive success rate.

Leave a Reply

Note: Only a member of this blog may post a comment.