The toughest schedules so far this year belong to AFC East teams, Miami and New England in particular. Jacksonville has also faced a tough slate of opponents to date. Aside from the NFC West teams that get to play each other twice, the softest schedule so far belongs to San Diego.
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 | SD | 1 | 0.83 | 0.42 | 1 | 1 |
2 | GB | 3 | 0.76 | 0.53 | 3 | 8 |
3 | PIT | 2 | 0.75 | 0.51 | 6 | 2 |
4 | NYG | 5 | 0.71 | 0.47 | 8 | 6 |
5 | PHI | 4 | 0.67 | 0.50 | 4 | 7 |
6 | MIA | 6 | 0.67 | 0.58 | 12 | 9 |
7 | NE | 9 | 0.65 | 0.57 | 2 | 23 |
8 | NYJ | 10 | 0.62 | 0.53 | 20 | 5 |
9 | BAL | 8 | 0.62 | 0.50 | 14 | 12 |
10 | TEN | 7 | 0.58 | 0.56 | 19 | 3 |
11 | KC | 11 | 0.58 | 0.43 | 11 | 14 |
12 | MIN | 13 | 0.58 | 0.54 | 16 | 11 |
13 | IND | 12 | 0.58 | 0.55 | 18 | 10 |
14 | CHI | 14 | 0.58 | 0.51 | 27 | 4 |
15 | HOU | 15 | 0.54 | 0.55 | 5 | 30 |
16 | NO | 16 | 0.53 | 0.40 | 10 | 19 |
17 | DAL | 17 | 0.52 | 0.55 | 7 | 27 |
18 | CLE | 18 | 0.47 | 0.51 | 21 | 17 |
19 | BUF | 23 | 0.46 | 0.57 | 23 | 21 |
20 | ATL | 19 | 0.44 | 0.49 | 13 | 25 |
21 | WAS | 21 | 0.43 | 0.55 | 17 | 24 |
22 | CIN | 20 | 0.42 | 0.54 | 28 | 18 |
23 | TB | 22 | 0.41 | 0.41 | 15 | 26 |
24 | DEN | 25 | 0.37 | 0.49 | 9 | 32 |
25 | SF | 26 | 0.36 | 0.39 | 24 | 15 |
26 | OAK | 28 | 0.33 | 0.49 | 26 | 16 |
27 | JAC | 24 | 0.32 | 0.57 | 22 | 31 |
28 | DET | 27 | 0.31 | 0.56 | 29 | 22 |
29 | SEA | 29 | 0.28 | 0.44 | 25 | 28 |
30 | STL | 30 | 0.24 | 0.38 | 30 | 20 |
31 | CAR | 31 | 0.23 | 0.48 | 32 | 13 |
32 | ARI | 32 | 0.17 | 0.44 | 31 | 29 |
And here are each team's efficiency stats.
TEAM | OPASS | ORUN | OINT% | OFUM% | DPASS | DRUN | DINT% | PENRATE |
ARI | 5.0 | 4.1 | 3.7 | 1.1 | 6.9 | 4.5 | 2.9 | 0.46 |
ATL | 6.2 | 4.1 | 1.2 | 0.2 | 7.0 | 4.2 | 3.9 | 0.33 |
BAL | 6.6 | 3.7 | 2.2 | 0.8 | 5.7 | 4.2 | 2.7 | 0.35 |
BUF | 5.7 | 4.5 | 3.1 | 1.2 | 6.4 | 4.6 | 1.2 | 0.30 |
CAR | 4.5 | 3.9 | 4.9 | 2.4 | 6.0 | 4.1 | 3.6 | 0.43 |
CHI | 5.9 | 3.8 | 4.6 | 0.2 | 5.5 | 3.6 | 3.8 | 0.42 |
CIN | 5.7 | 3.7 | 3.4 | 1.3 | 6.3 | 4.5 | 3.5 | 0.35 |
CLE | 6.1 | 4.2 | 3.4 | 1.5 | 6.5 | 4.1 | 3.9 | 0.38 |
DAL | 7.0 | 3.8 | 3.6 | 0.5 | 7.0 | 4.3 | 2.5 | 0.46 |
DEN | 6.9 | 3.5 | 1.4 | 1.5 | 7.4 | 4.3 | 1.8 | 0.48 |
DET | 5.6 | 3.6 | 2.7 | 0.9 | 6.6 | 4.6 | 2.6 | 0.55 |
GB | 7.2 | 4.0 | 2.4 | 0.6 | 5.5 | 4.5 | 3.9 | 0.36 |
HOU | 6.6 | 4.9 | 1.9 | 0.5 | 7.5 | 3.9 | 2.2 | 0.34 |
IND | 6.6 | 3.6 | 2.3 | 0.7 | 6.0 | 4.9 | 2.6 | 0.34 |
JAC | 6.0 | 4.5 | 4.9 | 1.0 | 8.0 | 4.3 | 2.6 | 0.34 |
KC | 6.5 | 4.9 | 1.2 | 0.5 | 6.1 | 4.0 | 2.1 | 0.37 |
MIA | 6.4 | 3.8 | 3.7 | 1.4 | 6.0 | 3.8 | 2.6 | 0.28 |
MIN | 6.2 | 4.4 | 4.8 | 0.7 | 6.2 | 3.6 | 2.5 | 0.39 |
NE | 7.0 | 4.3 | 1.4 | 0.2 | 6.9 | 4.2 | 3.4 | 0.39 |
NO | 6.8 | 3.9 | 3.3 | 0.7 | 5.8 | 4.2 | 2.0 | 0.37 |
NYG | 7.1 | 4.6 | 4.3 | 1.7 | 5.5 | 4.0 | 3.2 | 0.43 |
NYJ | 6.1 | 4.5 | 2.2 | 1.4 | 5.6 | 3.4 | 1.8 | 0.52 |
OAK | 5.7 | 4.7 | 3.7 | 1.3 | 6.2 | 4.4 | 1.8 | 0.62 |
PHI | 6.6 | 5.4 | 1.3 | 0.7 | 5.8 | 4.1 | 5.1 | 0.59 |
PIT | 6.9 | 4.2 | 2.6 | 1.1 | 5.7 | 3.0 | 3.2 | 0.48 |
SD | 8.0 | 4.0 | 2.4 | 1.5 | 5.3 | 3.5 | 3.6 | 0.39 |
SF | 6.0 | 4.2 | 3.4 | 1.0 | 6.2 | 3.6 | 2.5 | 0.53 |
SEA | 6.0 | 3.5 | 3.1 | 0.4 | 6.8 | 4.2 | 2.1 | 0.45 |
STL | 5.4 | 3.8 | 2.2 | 0.2 | 6.1 | 4.4 | 2.0 | 0.47 |
TB | 6.2 | 4.3 | 1.5 | 0.8 | 6.2 | 4.7 | 4.6 | 0.43 |
TEN | 6.2 | 4.4 | 3.3 | 1.0 | 5.8 | 4.0 | 3.4 | 0.55 |
WAS | 6.2 | 4.0 | 3.3 | 1.0 | 6.6 | 4.9 | 2.4 | 0.35 |
Avg | 6.3 | 4.2 | 2.9 | 0.9 | 6.3 | 4.1 | 2.9 | 0.42 |
Atlanta drops a spot? Someone's not gonna like that. :)
Don't the Saints have an even lower strength of schedule than the Chargers?
So is the point of this post to say that schedule is not included or that it is. I have to hope it isn't and suggest (if I am right) that perhaps your probabilities would be improved if you did include strength of schedule.
As someone who has graduate training in statistics, and who has followed your probabilities since last year, I cannot understand why you are so inflexible with your model. It performed well last year--but not well enough to justify how badly it is doing this year.
There is no way, Brian, that you think the model is performing well. Your posts on Fifth Down every week seem to be apologizing for the previous weeks predictions. So why not try changing the regression a bit? I mean, do you really think Miami is a better team (more efficient) than New England or the Jets?
If you want to keep your readers I suggest some admission that your model is not very good this year. I know, from previous comment arguments, that you are quite defensive, but us loyal readers can only take so much.
Heather-Yes, opponent adjustments are included.
And the model is performing just fine. Last week it outpaced the consensus favorites again.
Keep in mind there will be down years, however. That doesn't mean the model needs to be reflexively bent to fit the off-year. It is updated each new season with fresh data.
Thank you for your advice on how to keep my readers.
Actually, consensus favorites were 11/16 last week and your model was 10/16. That's not better.
I haven't run the numbers, so I take no position on whether the model has or has not been bad this year. Choose your favorite statistical test and your favorite level of confidence, run it against the data, and see if it the predictions have, in fact, been statistically inconsistent with the outcomes. Post the results.
But having been someone who has provided graduate training in statistics and modeling, I'd say that as a general rule, inflexibility in the face of noise is no vice.
I've been playing a yahoo pick-em game using Brian's game probabilities. I wanted to test how accurate his heavy favorites have been. So I pick in every game where one team has a 65%+ chance of winning. So far 73% of my 45 picks over the past 5 weeks have been correct. This is pretty close to the expected success rate. Given that I have graduate training in statistics, maybe I can do a more detailed test of his predictions, but in this small sample they look pretty good.
However, the picks fail sometimes. SD beat NE when SD had a couple of rare, but amusing fumbles.with some of the most innovative fumbles you'll ever see. ATL just beat GB, but only by 3 points- kind of a random outcome.
Also, back in week 10 his biggest favorites did pretty bad, going 3-4. However, 3-4 or worse is a fairly likely outcome in a binomial distribution where p=75%.
I read FO and your site. FO seems to put a lot more value on turnovers and special teams.
My gut tells me that the answer is probably somewhere in the middle.
Have you ever compared your results in terms of predicting winners?
The teams you and FO disagree most on are
Atl FO 9th, You - 20th
Minn FO 23rd, You - 12th
Dall FO 28th, You - 17th
jmaron-I don't track FO's winners.
Heather-Not sure what consensus/Vegas picks you follow, but I always use the USAToday's opening lines. It went 9-7, missing on PHI, DEN, IND, OAK, WAS, SEA, and ARI. You can always hunt for the line that's most unfavorable for the model though, after the fact.
The efficiency model is doing just fine. But like I mentioned, there will be up years and down years. It's wise not to chase the data and overfit the coefficients to a small sample of cases. Someone like Heather, who claims to have "graduate training" in statistics, whatever that may mean, should know that.
To be honest, predicting winners is just a means to an end for me. I want to understand the game and what helps teams win. The model is tool to that end. Predicting winners is just a happy by-product.
But the work goes on. Please read my recent post "Predictivity" for my latest research into predictive football modeling.
I want to understand the game and what helps teams win.
That's why I come here, because you do. OTOH, using a site like this (or FOers or Wages of Wins or whatever) to try and pick winners or beat the spread is just foolishness. There's a theoretical upper limit on picking winners of about 70%, and a 16-game season itself isn't enough for reliable results (not to mention even smaller samples of 8 games, 11 games, etc). So model prediction results always will vary. Anyone who really wants to know who's most likely to win on Sunday should just check the Vegas line or the Pythagorean ratings at PFR.com. Done.
But "running game success rate" solves a long-time puzzle, why coaches run so much, and why it can work even if average yds per rush are low, when so many more yards comes from passing. (Lombardi's Packers really were a powerhouse running team *even when* rushing a next-to-bottom 3.5 yds/carry.)
And WPA brings a brand new quantitative measure to MVP arguments, and makes objectively clear the difference between players' quality of performance and their reputation for making dramatic (good or bad) plays that results from things happening at key times.
These really add to understanding the sport. And they are both born natives to here. Kudos. As long as things like this keep coming, I'll keep coming back.
Bill James started this tide of quantitative sports analysis 30-odd years ago, and all the immense work done by the army of baseball sabremetricians since then hasn't been to beat their local bookies, it's been overwhelmingly just to better understand baseball, and to enjoy the game more, and for the fun of figuring things out, and then maybe ultimately to help improve the way the game is played. Now this tide is spreading to other sports.
As far as I am concerned, in *that* traditon, this has become the premier site for the NFL.
I thought the predictivity post was a really useful one and it gave me some great ideas for what can be done in a general sports modelling system in the hunt for the holy grail, a variable that correlates heavily both with itself and with winnings.
Just on Heather's point, I did review how your probabilities are doing so far this season the other week and they seemed to be ok in general, except the heavy favourites weren't winning as often as you'd expect. I suspect that's to do with regression to the mean of the various efficiency stats than it is a general fault with the model.
As far as I can tell, the model is built on season end efficiency models, but the inputs you use are how the teams are doing 'to date'. In nearly every case, midway through the season the best teams will have higher efficiencies in week 8 than they finish with, and the worst will have lower.
Thank you, Jim. Very kind.
Ian-Thanks. You are correct, except that the 'to-date' variables are regressed individually to match their typical season-ending variance. It's a regression within a regression. No matter how much I crank down the model's confidence, it still seems to produce some very heavy favorites. But you might notice there are no (or extremely few 90+/10- matchups this season.)
One more thing to keep in mind. This is a really dumb model. Seriously. It has no idea that Rusty Smith is starting at QB for the Titans or that the Redskins are missing LaRon Landry. If you are playing in some sort of pick 'em league, I don't recommend using these probabilities blindly. They're a great starting point, though, and a great place to find some underdog picks.
In that vein, adding some sort of man-in-the-loop criteria to the model (like when 3rd string QBs are starting) would easily give the model a better record than anything comparable.
Picking up on another user's comment from last week's Falcons post, I wanted to give some further thoughts on the Falcons this year that will certainly be biased, although I will try to preserve some level of balance.
Turnover rate-While the Falcons may regress a bit to the mean on turnovers, I think they are probably as good as any team in the league at preventing turnovers, and the regression might not be drastic. Ryan has actually gotten better as the season has progressed, culminating with his current streak of 4 (right?) consecutive games without an interception. If his ability to avoid interceptions has improved as the year has gone on, then his interception rate might "regress" back to its current value overall for the year. As for fumbles, the Falcons have lost all five fumbles that have been forced against them this year, so their stellar performance in that area seems well-earned as well.
Home record-Agree with the other commenter that the Falcons are not the unstoppable force at home that the record suggests. We are not necessarily built for the dome and the home crowd is mediocre. As he said, looking forward our home and road records will come more into line, although this says as much about how the Falcons are currently being underrated on the road as it does about how we are being overrated at home.
Luck-Before Hartley missed the field goal, we had the unluckiest of all bounces off of Decoud's leg when we were receiving a punt at midfield with the lead in the fourth quarter, so Hartley's kick should have never happened. Before Roddy White's pushoff (which many objective "experts" agree was not illegal), we were already in field goal range and there's no reason to believe based on this season or that game we wouldn't have gotten back into field goal range with over 20 seconds left. San Francisco was probably the "flukiest" of all the Falcons' wins, but that doesn't change the fact that Roddy made a great play, and didn't the same thing just happen on Thursday for NO-Dallas?
Injuries-This is probably where we have been the most "lucky". We have had a few significant injuries throughout the year but we have probably been as fortunate as any team in the NFL on this front.
Smart football-Aside from limiting turnovers, the Falcons also lead the league in fewest penalties and least yardage lost from negative plays. Those factors both impact our success themselves and also point to a savvy required to win football games that goes beyond efficiency stats. Along that same vein, yes, the Falcons have played close games all year, but many have essentially been close by design. With the exception of San Francisco, if you look at the games against New Orleans, Baltimore, Tampa, and Green Bay, the Falcons controlled those games. Many fans complain that the Falcons get too conservative on both offense and defense with a lead, which I'm sure would have an effect on the efficiency stats being used. While I too would argue that we play with fire a little too much with this strategy, the fact remains that by predictably running the football and playing a prevent-type defense, the Falcons limit teams' ability to come back and create situations where at worst we will have Matty Ice leading a drive with a chance to win the game for us in the end. Again, this is winning football that might not be fully reflected in efficiency stats.
Final thoughts-I agree with the previous commenter that the Falcons are a good team but not necessarily a great one, but I think that the Falcons are as "good" as any other team in the league this year. That said, there are about ten teams that are very evenly matched, so if the Falcons were ranked tenth and not twentieth in the model I likely wouldn't feel compelled to post such a long-winded comment. As it is, though, hopefully the case of the Falcons and the potential explanations offered by myself and other users can point to some potential improvements in the model.
I think that the Falcons are as "good" as any other team in the league this year
I don't think people should beat too much on Atlanta. My Jets have been lucky to win two in OT (once after the other team fumbled in scoring territory), once on a last moment pass interference on a desperation bomb attempt, and once when at a critical moment Sanchez under-handed the ball onto a DLman's facemask and the guy dropped it. They could easily be 6-5. There's a lot of luck in the game.
But it's not luck that these teams are in a position to be lucky -- not so far behind luck couldn't help. The formula to be a top team is to win a bunch of one-sided games, don't lose any one-sided games, and be lucky in the close ones. The combination of good and lucky is the winner. Take the 17-0 Dolphins. They were very lucky in that they had one of the weakest schedules in history (only two opponents with winning records, both 8-6) and won 6 one-score games, including by 1-pt and 4-pts over the 4-win Bills. Maybe that didn't look so impressive at the time. But they did go 17-0 and collect the ring. Nobody today says they didn't earn the perfect season.
Re: Falcons, especially regarding the comparison to FO. While they are ranked 20th in GWP, they are also ranked 10th in NET WPA and 12th in NET EPA (which you have to derive from the advanced team stats page). That is pretty close to FO's DVOA ranking them 9th (excluding special teams).
Comparing EPA/P on Atlanta's offense to YPA/YPC: Rushing: 5th in EPA/P but 17th in YPC.
Passing: 9th in EPA/P but 15th in YPA.
This is with mildly above-average success rates on offense. How can such a middling offense in yards per play add so many expected points per play? I'm not sure, but I think that is the key to figuring out "What's the deal with the Falcons."
Also, since someone posted above about the difference in ratings between FO and ANS on Dallas and Minnesota:
Dal: DVOA*: 29, NetEPA: 22, NetWPA: 25
Min: DVOA*: 21, NetEPA: 23, NetWPA: 21
It looks like rather than FO and BB disagreeing, that GWP and DVOA simply measure different things and that EPA measures something closer to DVOA.
I'd be curious to see the team efficiency rankings with each team's best game (largest win margin) and worst game (largest loss margin) excluded from the data...seems that almost every team has at least one exceptional win and loss that skews the real average data.
On a related note, it would be curious to also exclude 4th quarter stats for teams that are up or down by a significant margin at the beginning of the 4th quarter (21+ points?)...a lot of meaningless yards are piled up during this period that skews the data.