The ratings are listed below 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. A full explanation of the methodology can be found here. This year, however, I've made one important change based on research that strongly indicates that defensive interception rates are highly random and not consistent throughout the year. Accordingly, I've removed them from the model and updated the weights of the remaining stats.RANK TEAM LAST WK GWP Opp GWP O RANK D RANK 1 
3 0.76 0.53 20 1 2 
5 0.76 0.54 7 5 3 
1 0.76 0.56 1 17 4 
2 0.75 0.52 12 4 5 
4 0.69 0.52 6 10 6 
9 0.68 0.56 3 13 7 
7 0.68 0.43 16 2 8 
6 0.67 0.51 5 12 9 
11 0.66 0.43 2 19 10 
8 0.65 0.51 11 11 11 
15 0.64 0.52 14 7 12 
12 0.63 0.56 18 6 13 
10 0.63 0.52 19 3 14 
13 0.62 0.48 9 18 15 
16 0.61 0.51 21 8 16 
14 0.56 0.51 4 22 17 
17 0.52 0.55 13 20 18 
20 0.51 0.50 24 9 19 
18 0.49 0.49 8 27 20 
19 0.48 0.46 15 24 21 
21 0.47 0.43 23 14 22 
22 0.44 0.51 10 28 23 
24 0.38 0.48 17 25 24 
23 0.34 0.43 28 21 25 
25 0.33 0.48 27 16 26 
26 0.32 0.49 25 26 27 
29 0.29 0.52 22 30 28 
30 0.28 0.60 31 23 29 
28 0.27 0.57 26 29 30 
27 0.27 0.55 32 15 31 
31 0.15 0.53 29 31 32 
32 0.14 0.57 30 32
To-date efficiency stats below. As always, click on the headers to sort.TEAM OPASS ORUN OINTRATE OFUMRATE DPASS DRUN DINTRATE PENRATE ARI 7.2 3.3 0.023 0.025 6.4 4.0 0.025 0.38 ATL 7.6 4.2 0.023 0.015 6.1 4.9 0.021 0.29 BAL 5.8 3.8 0.031 0.025 5.1 3.4 0.048 0.41 BUF 5.9 4.2 0.035 0.036 6.2 4.2 0.021 0.29 CAR 6.9 4.8 0.032 0.015 5.6 4.1 0.023 0.35 CHI 5.5 3.9 0.026 0.016 5.7 3.5 0.038 0.31 CIN 4.2 3.4 0.032 0.027 6.6 3.9 0.018 0.30 CLE 5.0 3.9 0.031 0.025 7.1 4.5 0.054 0.32 DAL 6.9 4.3 0.037 0.030 5.3 4.0 0.017 0.49 DEN 7.1 4.5 0.029 0.020 6.9 4.9 0.013 0.34 DET 5.3 3.7 0.034 0.040 7.8 4.9 0.011 0.38 GB 6.5 4.0 0.025 0.023 6.1 4.8 0.040 0.50 HOU 7.2 4.4 0.039 0.029 6.9 4.5 0.029 0.34 IND 6.6 3.4 0.023 0.010 6.0 4.2 0.033 0.32 JAX 5.7 4.2 0.021 0.018 6.8 4.1 0.032 0.43 KC 5.3 4.7 0.028 0.020 7.3 4.9 0.026 0.30 MIA 7.1 4.1 0.014 0.016 6.1 3.9 0.026 0.37 MIN 5.9 4.5 0.041 0.023 6.0 3.2 0.025 0.38 NE 5.9 4.5 0.023 0.018 6.7 4.2 0.031 0.24 NO 7.6 3.8 0.031 0.019 6.3 4.0 0.027 0.41 NYG 6.1 4.8 0.023 0.020 5.5 3.9 0.035 0.45 NYJ 6.0 4.8 0.039 0.024 6.1 3.7 0.027 0.28 OAK 4.9 4.2 0.027 0.037 6.4 4.7 0.035 0.45 PHI 6.3 4.0 0.030 0.014 5.2 3.4 0.031 0.34 PIT 5.8 3.6 0.027 0.023 4.3 3.2 0.037 0.42 SD 7.5 3.8 0.026 0.020 6.2 3.9 0.021 0.38 SF 6.0 3.9 0.034 0.042 6.3 3.8 0.023 0.40 SEA 5.0 4.4 0.032 0.016 6.9 4.1 0.012 0.28 STL 5.2 3.9 0.040 0.027 7.6 4.7 0.018 0.39 TB 6.0 4.0 0.020 0.020 5.7 4.3 0.050 0.41 TEN 6.2 4.4 0.022 0.020 4.9 3.7 0.036 0.44 WAS 5.7 4.5 0.013 0.021 5.8 3.8 0.028 0.36 Avg 6.1 4.1 0.029 0.023 6.2 4.1 0.028 0.37
Week 15 Efficiency Rankings
published on 12/16/2008
Subscribe to:
Post Comments (Atom)
Nice - the Eagles recently seem to have justified models having them highly all year.
One (unrelated) thing I was kind of curious about is, with your win probability graphs and such, if you could calculate some sort of crude WPA? It seems like that's what stats are essentially trying to get at anyways - incorporating the differing values of first downs/turnovers/varying events, just in a less direct way. Obviously, this wouldn't necessarily be a good stat for comparing individual players, but even something like, say, total running WPA vs total passing WPA on first down would imo be very interesting.
Have you looked at subdivisions of the season (last 3|5|x games; last 80% of games; etc) to see if that increases the game projections? Thought is that teams seem to go through ups and downs during the season. Romo goes out, and Dallas becomes horrible for several games. The Redskin's tackles get hurt, and they seem to be a different team. Do the stats back that up, or is it just my mind playing tricks?
Miles-So far, nothing really improves accuracy, but it doesn't hurt it either. You can look at week 5 or 6 and see that the model's accuracy isn't much different than in week 15.
One thing to mention is that there is an upper limit to any prediction system accuracy. Sometimes there are upsets, and there are plenty of 55/45 toss-up type games. I have done extensive research on that, and I think 80% is the theoretical maximum achievable.
Alex-Yes, eventually we could get to a player-level WPA. I agree that any player-level stat in football is relatively meaningless, but that hasn't stopped me in the past!
Out of curiosity, have you checked to see how the GWPs correlate with point differential as well? I always thought this may be a better measure of team strength than wins - if a team is 1-1 with a thirty point win and a three point loss, that says they're likely a better than average team. Looking at the data in terms of that seems to show a strong correlation, though there are still some surprises (i.e., Atlanta having a poorer PD than the Jets and being substantially higher)