A Method for Calculating Power Rankings (and projecting final standings)
| 15
|
by Derekcs
Team quality is not exactly a transitive property (i.e. A is better than B, B is better than C, does not necessarily mean A is better than C. C might just match up really well with A). So how would you decide who is the best team in the NFL? My idea was simply this: if you pitted each team against every other team in a home and away game, who would be expected to win the most games? Obviously, actual results can vary widely, and don't necessarily reflect the entirety of a team's quality. But based on the stats, we can get some idea of who's more likely to win. With logistic regression, we can produce the probability that a team is going to win a given game. The power ranking is simply the expected win totals (the sums of those probabilities for each team) normalized for a 16-game season.
The inputs used for this regression were rushing yards per carry, passing yards per attempt, sack rates, third down conversion rates, interception and fumble rates on offense and defense.
The power rankings as of week 2 of the 2007 season are:
Rank Team Expected Wins
- NE 12.7821
- DEN 12.3057
- PIT 11.9186
- IND 11.2294
- HOU 11.0758
- DAL 10.9747
- TB 9.6725
- CAR 9.4341
- DET 9.1868
- WAS 9.1101
- ARI 9.0378
- MIN 9.0016
- JAX 8.8980
- PHI 8.7440
- SEA 8.4853
- BAL 8.2007
- TEN 7.4589
- CLE 7.3766
- GB 7.2951
- CIN 6.8305
- KC 6.8023
- CHI 6.7402
- STL 6.7007
- NYG 6.2975
- SF 6.2020
- MIA 6.0230
- SD 5.9910
- ATL 5.5020
- OAK 5.1785
- NYJ 4.9122
- BUF 4.2329
- NO 2.3994
A couple things to point out here. First, KC is ranked higher than the Bears because Devin Hester's talents are not included in the model. Otherwise, Chicago has not really been any better than KC. Second, passing is valued much more highly by the model than rushing, and offense is valued somewhat more than defense by the model. So you might not like Detroit over Minnesota, but as I've pointed out here and here, good pass offenses are more likely to take a team to the postseason, and the strength of Minnesota's defense has been against the run.
Here's another moment where I point out how eerie Jon Kitna's prediction of the Lions winning 10 games is. Using the probabilities generated by the model, I looked at each team's remaining schedule and projected their final win totals based on the number of games they are considered more likely to win. Detroit and Minnesota are projected to lead the NFC North with 10 wins, with Chicago bringing up the rear (5 wins).
Two caveats about these projections:
1. SMALL SAMPLE SIZE. I don't think San Diego is a 3-13 team. They haven't performed all that well in the first two weeks, however. (Opponent adjustments will help once the sample size increases by a week or two).
2. THIS MODEL DOES NOT CONSIDER ALL POSSIBLE FACTORS. No, no one expects the Saints to go winless or the Pats to go undefeated. A good proportion of the games are considered close calls, but in the case of the Pats, they're all in their favor. The reverse is true for the Saints. Luck and special teams and injuries and other factors not considered by the model will swing many games the other way.
AFC East BUF 2 MIA 5 NE 16 NYJ 2
AFC North BAL 9 CIN 6 CLE 8 PIT 14
AFC South HOU 15 IND 14 JAX 8 TEN 6
AFC West DEN 14 KC 7 OAK 3 SD 3
NFC East DAL 14 NYG 3 PHI 9 WAS 11
NFC North CHI 5 DET 10 GB 8 MIN 10
NFC South ATL 4 CAR 9 NO 0 TB 12
NFC West ARI 11 STL 4 SF 3 SEA 11
