Efficiency Rankings - Week 11

One of the more perplexing things about the rankings lately is that the 7-2 Patriots have a net negative balance of offensive and defensive YPA. NE averages 6.5 net YPA but gives up 6.8 net YPA, while their other stats are very average. How are they winning and beating quality opponents (0.60 Opponent GWP)?

Perhaps part of the answer is consistency. Their offensive Success Rate, adjusted for opponent is 2nd in the league. They might only be getting 6.5 YPA, but if they got 6 or 7 yards on every attempt, they'd be unbeatable.

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 0.80 0.41 1 3 2 PIT 2 0.76 0.55 8 2 3 NYG 3 0.71 0.45 4 5 4 GB 5 0.70 0.51 9 8 5 PHI 6 0.69 0.48 2 6 6 MIA 10 0.68 0.59 11 12 7 TEN 4 0.64 0.59 18 1 8 BAL 9 0.62 0.54 19 13 9 KC 8 0.59 0.48 15 9 10 NYJ 11 0.59 0.53 22 7 11 IND 7 0.58 0.53 24 10 12 NE 16 0.57 0.60 7 25 13 MIN 14 0.55 0.51 14 16 14 NO 12 0.54 0.42 16 15 15 CLE 17 0.52 0.56 20 18 16 HOU 13 0.51 0.55 3 29 17 CIN 19 0.51 0.54 25 17 18 CHI 15 0.50 0.45 27 4 19 DAL 21 0.50 0.57 6 26 20 ATL 20 0.48 0.51 13 23 21 WAS 18 0.43 0.52 17 24 22 DEN 24 0.42 0.49 5 30 23 TB 23 0.41 0.41 10 31 24 OAK 22 0.40 0.45 23 14 25 SF 25 0.40 0.43 12 20 26 BUF 27 0.36 0.55 28 27 27 JAC 28 0.34 0.57 21 32 28 DET 26 0.32 0.53 26 21 29 SEA 31 0.26 0.42 29 22 30 CAR 30 0.23 0.47 32 11 31 STL 29 0.23 0.38 30 19 32 ARI 32 0.17 0.44 31 28

And here are each team's efficiency stats.

 TEAM OPASS ORUN OINT% OFUM% DPASS DRUN DINT% PENRATE ARI 4.9 4.2 4.3 1.1 6.9 4.3 3.0 0.41 ATL 6.2 4.0 1.5 0.2 6.9 4.2 4.6 0.33 BAL 6.5 3.8 2.3 0.8 6.0 4.0 2.6 0.38 BUF 5.4 4.4 3.0 1.2 6.5 4.7 0.7 0.31 CAR 4.4 3.6 4.8 2.4 5.8 4.1 3.8 0.41 CHI 5.8 3.8 5.0 0.2 5.5 3.5 4.0 0.42 CIN 6.0 3.7 3.1 1.3 6.1 4.3 3.3 0.36 CLE 6.1 4.2 3.1 1.9 6.6 3.9 3.2 0.39 DAL 7.1 3.6 4.0 0.0 7.1 4.3 2.5 0.51 DEN 7.2 3.2 1.4 1.5 7.2 4.4 1.8 0.48 DET 5.7 3.4 2.6 0.9 6.3 4.7 3.1 0.55 GB 6.8 4.2 3.0 0.5 5.5 4.5 4.4 0.37 HOU 6.8 5.0 2.3 0.2 7.8 4.1 1.5 0.34 IND 6.5 3.7 1.0 0.8 5.9 5.0 3.0 0.36 JAC 6.4 4.3 4.3 0.8 8.1 4.3 2.8 0.39 KC 6.3 4.8 1.5 0.6 6.0 4.0 2.1 0.39 MIA 6.5 3.8 3.7 1.3 5.9 4.0 2.1 0.27 MIN 6.3 4.5 5.4 1.1 6.2 3.7 2.7 0.41 NE 6.5 4.1 1.6 0.0 6.9 4.2 2.9 0.42 NO 6.4 3.8 3.2 0.8 5.2 4.2 2.2 0.42 NYG 7.1 4.6 4.1 1.6 5.7 3.5 3.7 0.45 NYJ 6.1 4.6 2.0 1.5 5.8 3.4 1.6 0.55 OAK 6.0 4.9 3.3 1.3 5.6 4.5 1.9 0.69 PHI 6.8 5.4 1.3 0.6 5.8 4.1 5.0 0.59 PIT 6.9 4.0 3.3 0.9 6.2 2.8 3.2 0.35 SD 8.0 4.0 2.4 1.7 5.4 3.6 2.9 0.41 SF 6.4 3.9 3.3 0.9 6.4 3.7 2.6 0.59 SEA 5.6 3.6 3.3 0.2 6.5 3.9 2.0 0.46 STL 5.1 3.7 2.4 0.2 5.9 4.2 2.5 0.51 TB 6.4 4.4 1.8 0.9 6.5 4.8 4.9 0.43 TEN 6.2 4.3 2.4 1.0 5.8 3.9 3.9 0.54 WAS 6.2 4.3 3.5 1.2 6.5 5.0 2.4 0.38 Avg 6.3 4.1 2.9 0.9 6.3 4.1 2.9 0.43

15 Responses to “Efficiency Rankings - Week 11”

1. Dave says:

Hi Brian,

I just made a comment in the week 10 efficiency rankings to this effect, but I'll repeat it here:
Have you made any changes to the coefficients or formula that you use when calculating Game Probabilities? For week 10, I used the process described in this post - http://www.advancednflstats.com/2009/01/how-model-works-detailed-example.html - and was unable to accurately reproduce your Game Winning probabilities. It could be a rounding issue but a couple of them were way off. Just curious if anything had changed. Thanks!

2. Anonymous says:

Interesting that the Colts have the #24 offense.

3. Brian Burke says:

Dave--Yes. The coefficients are slightly different now. Plus, each facet is regressed to a different degree based on how consistent they are through a season.

4. James says:

Okay Brian, here's the question everyone is waiting for: How long until you start factoring in SR into your prediction model?

5. Brian D says:

Brian:
Why is the Colt offense 24th? Passing is 12th, running 27th, INT 1st and Fumble 11th.

6. Brian Burke says:

It probably has a lot to do with the defenses they've played against. Turnovers are still regressed heavily.

7. Guy says:

Brian: It occurs to me that SR and efficiency are somewhat analogous to on-base percentage and slugging percentage in baseball. OBP is how often a hitter succeeds (getting on base always increases run expectancy, an out always decreases it), while SLG measures the extent of his success (similar to YPA). Both are important and predictive, but OBP is a bit more important.

The common denominator here, I think, is a sequential offense. Baseball teams need to score runs before making three outs, so the ability to string together a sequence of positive events is critically important. Similarly, an NFL team needs to advance 10 yards within the constraint of 3 downs (until they start listening to you more) to maintain possession. And they must (usually) put together several consecutive successful conversions to score. So in both sports, consistency of success becomes very important.

8. Jason says:

I think the Pats defense numbers are a bit misleading in that they've given up large chunks of yardage in what is basically garbage time (270 in the second half against the Bengals when they had already surpassed the final score, not to mention 50 on the half-ending hail mary that didn't reach the end zone). They seem to be trading yards for time, which is working although it's throwing the measurements off.

9. Brian Burke says:

Guy-Absolutely. You need sequential successes. 'Runs', in the statistical sense, are key.

Jason-That's definitely a consideration. I think part of the mystery of why the Pats usually over-perform their stats is that their run success rate. They have been extremely consistent over the past decade despite not being known for a power run game.

10. Guy says:

I'd guess the Pats run SR is partially a function of their pass SR -- the pass "sets up the run" as well as vice versa. So their run success may be as much a function of the strategies forced upon the defense as much as the power of NE running backs.

*

The need for sequential success is why I think you'll eventually discover that running is as important as passing. The higher certainty of success in some situations (short yardage) is tremendously valuable. But it can't be fully captured in expected points. The problem is that football plays -- unlike offensive events in baseball -- are not random. If a team makes a 7 yard passing gain on first down, the expectancy gain assigned to that pass incorporates the high likelihood of a conversion. Why is it so likely? Because two successive runs are extremely likely to net 3 yards. But the "credit" for that in expectancy terms goes to the 7-yard play. Expectancy in a sense "steals value" from future plays.

This can't happen in baseball. A double's run expectancy is a function of the independent probabilities of all future possible offensive events. A baseball team can't conjure up a single just because that's what it needs to score that runner. But in football, to some extent, you can select plays likely to get the yardage you need at that moment. A 3-yard gain at 2nd-and-7 probably has about a zero point expectancy gain, because we already "expected" a first down. Many of the most valuable runs have zero PE because they're essentially being measured against themselves (what we expected). But the high likelihood of a gain in these situations is in fact very valuable.

11. Guy says:

Ooops, I meant "a 3-yard gain at 2nd-and-3 probably has a zero point expectancy gain."

12. Guy says:

Brian: have you ever experimented with using median yards per attempt, rather than mean, in your efficiency metrics? Might remove influence of extreme gains and interceptions. (Though computationally more difficult.)

13. Anonymous says:

Not sure if it's entirely a good idea to remove the outliers like extreme gains and interception returns...they should be factored into the model, since they are a part of the expected value of a play. I understand they increase the variance of the model, but perhaps that's where something like SR can be factored in, so long as it's something a team is capable of reproducing season to season.

14. Brian Burke says:

Because of the way median works, for running, it's almost always 3 yards for every team. A few outliers might be 2 or 4.

15. Guy says:

Doh! Of course, there are no fractional yardage gains in the data. I guess SR is the best option.