I have begun to calculate my own power rankings. I will submit my methodology and rationalization in bit but am out the door at the moment. Let me know what you think, there are some obvious outliers if you will at this time.
Code: Rank Team Sum Win Loss % H-W H-L H-% R-W R-L R-% PF PA PD S.O.S. 1 TEN 10.41 4 0 1.000 3 0 1.000 1 0 1.000 25.5 11.5 14.0 0.345 2 NYG 8.66 3 0 1.000 2 0 1.000 1 0 1.000 27.7 14.3 13.3 0.500 3 BUF 8.65 4 0 1.000 2 0 1.000 2 0 1.000 27.3 15.8 11.5 0.424 4 DAL 8.43 3 1 0.750 1 1 0.500 2 0 1.000 30.0 22.3 7.8 0.533 5 SD 5.19 2 2 0.500 1 1 0.500 1 1 0.500 34.5 28.0 6.5 0.541 6 TB 5.04 3 1 0.750 2 0 1.000 1 1 0.500 25.3 19.5 5.8 0.484 7 PHI 4.61 2 2 0.500 2 0 1.000 0 2 0.000 27.5 18.5 9.0 0.550 8 MIA 3.95 1 2 0.333 0 1 0.000 1 1 0.500 20.7 21.3 -0.7 0.508 9 WAS 3.79 3 1 0.750 2 0 1.000 1 1 0.500 21.5 20.3 1.3 0.492 10 BAL 3.23 2 1 0.667 2 0 1.000 0 1 0.000 21.7 14.3 7.3 0.467 11 CHI 2.86 2 2 0.500 1 1 0.500 1 1 0.500 23.5 20.0 3.5 0.417 12 PIT 2.80 3 1 0.750 2 0 1.000 1 1 0.500 19.3 14.5 4.8 0.483 13 CAR 2.28 3 1 0.750 2 0 1.000 1 1 0.500 20.0 17.5 2.5 0.500 14 DEN 1.90 3 1 0.750 2 0 1.000 1 1 0.500 33.3 29.3 4.0 0.484 15 NO 1.85 2 2 0.500 2 0 1.000 0 2 0.000 27.8 25.0 2.8 0.524 16 NYJ -0.09 2 2 0.500 1 1 0.500 1 1 0.500 28.8 29.0 -0.3 0.508 17 SEA -0.26 1 2 0.333 1 1 0.500 0 1 0.000 25.7 26.7 -1.0 0.541 18 GB -0.35 2 2 0.500 1 1 0.500 1 1 0.500 27.3 25.3 2.0 0.458 19 ATL -0.43 2 2 0.500 2 0 1.000 0 2 0.000 22.5 20.8 1.8 0.476 20 JAX -0.94 2 2 0.500 1 1 0.500 1 1 0.500 19.8 21.3 -1.5 0.483 21 ARI -0.99 2 2 0.500 1 0 1.000 1 2 0.333 26.5 25.8 0.8 0.508 22 SF -1.10 2 2 0.500 1 1 0.500 1 1 0.500 23.5 24.3 -0.8 0.483 23 MIN -1.67 1 3 0.250 1 1 0.500 0 2 0.000 17.8 20.5 -2.8 0.508 24 IND -3.28 1 2 0.333 0 2 0.000 1 0 1.000 17.3 22.3 -5.0 0.458 25 KC -4.44 1 3 0.250 1 1 0.500 0 2 0.000 16.3 24.3 -8.0 0.565 26 OAK -4.57 1 3 0.250 0 2 0.000 1 1 0.500 19.5 25.3 -5.8 0.550 27 NE -4.78 2 1 0.667 1 1 0.500 1 0 1.000 16.3 19.3 -3.0 0.500 28 CLE -5.74 1 3 0.250 0 2 0.000 1 1 0.500 11.5 19.5 -8.0 0.593 29 HOU -6.34 0 3 0.000 0 0 0.000 0 3 0.000 18.7 33.0 -14.3 0.458 30 CIN -8.09 0 4 0.000 0 2 0.000 0 2 0.000 13.0 21.8 -8.8 0.559 31 DET -10.66 0 3 0.000 0 1 0.000 0 2 0.000 19.7 37.7 -18.0 0.516 32 STL -19.89 0 4 0.000 0 2 0.000 0 2 0.000 10.8 36.8 -26.0 0.576 Notes: ? Ranks are calculated based on Points Margin, Home/Away and Strength of Opponent for each game and summed over the season. ? Ranks are based on an absolute sum not averaged over games played (teams with less games played will be at a disadvantage). ? The rank differential portion of the calculation is based on the prior weeks ranks. Early season game rank multipliers might be significantly different from their late season counterparts ? Win Loss, Home/Away, PPG and Stength of Schedule are shown as reference only and are not explicitly used in the calculations.
The proof is in the pudding, so to speak. Based upon your rankings, you can make predictions about this weeks games. So your picks are: Buf, Car, Chi, Dal, TB, GB, Ind, Pitt, SF, Giants, Phi, SD, Ten, NO. I would say if you are right on 11/14 then you're on to something. Now, you may want to adjust the rankings for home field in a game and other intangibles. For example, TB at Den--you could subtract 1.5pts for TB on the road, and add 1.75 for Den at home. As an example, this would change your pick to Den. For the model to be good, you would have to calculate these adjustments in a lawful manner. Also, you can apply your theories to past seasons and see how it turns out. Then, adjust the values so that you predict seasons that are over very well. Then, use those values to predict the future. Keep on working on it--it has good potential, in my opinion.
it's early so the dolphins/Pats game skews things alot, a big road win for thedolphins, a big home loss for the patriots, and both had the bye so only 3 games
That is exactly it. Steve I agree with you it is odd and probably means the calculations I use need to be tweaked more than that they are the 8th best team. To give a little background I base the rankings on making a rank calculation based on Point differential, Home/Away (Home wins are worth less than Road wins, etc.) and Rank differential (a 16 beating a 2 is better than a 5 beating a 2, etc.). These are calculated each week and summed from week to week. The calculations are such that the winning team gets X and the losing team gets -X so it will always add up to 0. For example if #5 Team A beats #7 Team B by X points at home, Team A will receive a certain amount of "points" for beating them by X (this amount could add or detract from the total based on if it was a blowout or a close game (blowouts = more impressive)), a certain amount of points for beating a #7 team as a #5 team and a certain amount of points for winning at home. These are all multiplied and added to the previous weeks total which is used to calculate the next weeks ranks.
As mentioned in my previous post, this seems like an outlier. It was the single biggest weekly total by any team in the first 4 weeks because of the huge point differential on the road. For example, Philly beat STL by the most points so far but did it on the road in week 1 (where all teams are assumed to be equal). Given that it is so short into the season and some teams have played more games than others (NYG for instance would probably be #1 if they played 4 games and won the 4th) this will even out over time. Nevertheless I am seeking additional scoring metrics that make sense or others ideas on the best way to value a certain outcome of a game. Thoughts?
Week 5 Power Rankings Code: Rank Team Sum Win Loss % H-W H-L H-% R-W R-L R-% PF PA PD S.O.S. 1 NYG 14.15 4 0 1.000 3 0 1.000 1 0 1.000 31.8 12.3 19.5 0.513 2 TEN 10.87 5 0 1.000 3 0 1.000 2 0 1.000 23.0 11.2 11.8 0.347 3 ARI 10.39 3 2 0.600 2 0 1.000 1 2 0.333 29.4 24.0 5.4 0.514 4 DAL 9.10 4 1 0.800 2 1 0.667 2 0 1.000 30.2 22.2 8.0 0.514 5 CHI 8.82 3 2 0.600 1 1 0.500 2 1 0.667 25.6 17.4 8.2 0.413 6 CAR 7.95 4 1 0.800 3 0 1.000 1 1 0.500 22.8 14.0 8.8 0.494 7 WAS 5.42 4 1 0.800 2 0 1.000 2 1 0.667 21.8 19.6 2.2 0.479 8 MIA 4.79 2 2 0.500 1 1 0.500 1 1 0.500 19.8 18.5 1.3 0.479 9 TB 4.37 3 2 0.600 2 0 1.000 1 2 0.333 22.8 18.8 4.0 0.494 10 PIT 4.07 4 1 0.800 2 0 1.000 2 1 0.667 20.6 15.8 4.8 0.472 11 SD 3.51 2 3 0.400 1 1 0.500 1 2 0.333 29.6 25.8 3.8 0.554 12 PHI 3.02 2 3 0.400 2 1 0.667 0 2 0.000 25.4 19.4 6.0 0.581 13 BAL 2.69 2 2 0.500 2 1 0.667 0 1 0.000 18.8 14.0 4.8 0.493 14 DEN 1.83 4 1 0.800 3 0 1.000 1 1 0.500 29.8 26.0 3.8 0.459 15 NO 1.18 2 3 0.400 2 1 0.667 0 2 0.000 27.6 26.0 1.6 0.526 16 ATL 0.32 3 2 0.600 2 0 1.000 1 2 0.333 23.4 21.4 2.0 0.455 17 NYJ 0.16 2 2 0.500 1 1 0.500 1 1 0.500 28.8 29.0 -0.3 0.507 18 MIN -0.77 2 3 0.400 1 1 0.500 1 2 0.333 20.2 21.8 -1.6 0.507 19 GB -1.16 2 3 0.400 1 2 0.333 1 1 0.500 26.6 25.6 1.0 0.480 20 NE -2.00 3 1 0.750 1 1 0.500 2 0 1.000 19.8 19.8 0.0 0.500 21 IND -2.01 2 2 0.500 0 2 0.000 2 0 1.000 20.8 23.5 -2.8 0.446 22 JAX -2.10 2 3 0.400 1 2 0.333 1 1 0.500 20.0 22.2 -2.2 0.493 23 SF -3.30 2 3 0.400 1 2 0.333 1 1 0.500 23.0 25.4 -2.4 0.493 24 BUF -3.47 4 1 0.800 2 0 1.000 2 1 0.667 25.2 20.8 4.4 0.443 25 OAK -4.15 1 3 0.250 0 2 0.000 1 1 0.500 19.5 25.3 -5.8 0.520 26 CLE -5.59 1 3 0.250 0 2 0.000 1 1 0.500 11.5 19.5 -8.0 0.573 27 SEA -6.29 1 3 0.250 1 1 0.500 0 2 0.000 20.8 31.0 -10.3 0.541 28 HOU -7.57 0 4 0.000 0 1 0.000 0 3 0.000 20.8 32.5 -11.8 0.479 29 CIN -8.92 0 5 0.000 0 2 0.000 0 3 0.000 14.8 23.6 -8.8 0.556 30 KC -9.76 1 4 0.200 1 1 0.500 0 3 0.000 13.0 26.2 -13.2 0.560 31 DET -16.39 0 4 0.000 0 2 0.000 0 2 0.000 16.5 36.8 -20.3 0.526 32 STL -19.17 0 4 0.000 0 2 0.000 0 2 0.000 10.8 36.8 -26.0 0.581 Notes: ? Ranks are calculated based on Points Margin, Home/Away and Strength of Opponent for each game and summed over the season. ? Ranks are based on an absolute sum not averaged over games played (teams with less games played will be at a disadvantage). ? The rank differential portion of the calculation is based on the prior weeks ranks. Early season game rank multipliers might be significantly different from their late season counterparts ? Win Loss, Home/Away, PPG and Stength of Schedule are shown as reference only and are not explicitly used in the calculations.
The anomaly here is obviously ARI and BUF. That game caused the season high in rank adjustment with a blowout win over the then #2 ranked team. As you can see, Arizona went from negative to the third highest rank total in one game. We shall see if Arizona can back it up like Miami which validated its case as a top 10 team with a solid victory over San Diego.
I'd say you have a very good start on your hands Rambo. Obviously you need tweaking, but I like your start. Keep us updated =D
Yeah tweaks are needed and I'm open to suggestions. Maybe I'll post the exact formula to help. It's tough to justify dropping a one loss team from 2 to 24 regardless of the circumstances. It also doesn't take into account injuries which could float teams one way or the other.
Seems to me that you need to find a way to damp the results from one game. The Arizona Buffalo game seems to have skewed things big-time. Is there a way you can throw out a high and a low or weight an anomalous result less than results that fit a trend? For example, we have instruments that are crappy, but they're easy to use. So, we take five readings and pitch the high and the low and average the other three. Using this, we can get fairly repeatable readings, even if there's a propensity for reading high or low. Just a thought. The idea about looking at certain points in past seasons and (knowing what the results will be) seeing how your formula holds up and where it falls down (and if it falls down in a particular spot repeatedly) is certainly a good one in my book. That gives you an idea of where the holes are and if you have repeating holes for the same reason, should point you to a way to improve the formula.
I would be hesitant to throw out results because they are all meaningful. Us beating Arizona by ~30 and Arizona beating Buffalo by ~30 are equally important as a 3 point win in my opinion (as far as weighting them). I do agree that one game will skew too much but am expecting this to balance out as the season progresses. Although that is not a solution or ideal.
Maybe assign some sort of value for injuries. For example, a starting QB might be worth 5 points in the final score while a third WR might be worth 1.5 or 2 or something. Those are totally random numbers, but if you based it on something like WARP (used in baseball to measure a player's value over a replacement level player at that position) it might help.
Are these power rankings intended to rank how a team has done or how a team will do? Past tense versus future tense. Should I be able to use it to predict?
Past tense. Not intended to predict how teams will do agaisnt one another but may unintentionally as an attempt to quantify year to date performance.
Miami is in the top 10 in both offense and defense, it would make sense they rank as high as they do. However a loss to an 0-4 Texans team and Miami could drop quickly.The Texans are the only team in the NFL the Dolphins have never beaten. Sunday will be a tough game for the Dolphins if they even think they have this game is in the bag because the Texans are 0-4, the Dolphins will lose it!!
I see one huge problem here. So lets say that Miami loses to Houston this week(not saying they will, but just for the sake of the argument), then Houston's rank will sky-rocket because they are the #29 and beat the #8. But what if your original algorithm for placing Miami at #8 was wrong, or just didn't have a big enough sample size(as is the case with Miami at 8, imo). The problem with your current formula, imo, is that it relies heavily on an infinite season and we only have 16 games + playoffs.
Miami seemed to be a fluke anomaly before last weeks game as a 1-2 top 10 team but maybe the formula is accurate because they backed it up by beating SD and their loss to us was negligible in the rankings.
