Here are my power rankings generated by a mathematical program that takes into account this seasons results so far. Be warned that they will not be totally accurate until later on in the year when more match data is available. With only 3 weeks worth of results the rankings dont look very accurate with a few surprises (Chicago ranked 2nd!) but they will start to take shape later in the year. When comparing the chances of two teams playing each other a total of 7 must be added to the home teams rating to allow for home field advantage. An example would be Philadelphia@Kansas City next weekend. Philadelphia are rated at 32 compared to KC 30. Because KC are at home we add 7 points which makes their rating 37. So the tip here would be KC by 5. These predictions become more accurate as the season progresses.




1 Cincinatti 47
2 Chicago 34
3 Pitsburgh 33
4 Philadelphia 32
5 Kansas City 30
6 Jacksonville 30
7 Washington 29
8 Indianapolis 29
9 Atlanta 29
10 Cleveland 26
11 Seattle 25
12 Tampa Bay 24
13 Carolina 23
14 Dallas 21
15 New England 21
16 Minnesota 21
17 NY Giants 21
18 San Diego 20
19 Denver 20
20 Miami 19
21 Oakland 18
22 NY Jets 17
23 Green Bay 16
24 Detroit 16
25 Buffalo 14
26 Tennessee 13
27 New Orleans 11
28 St Louis 11
29 San Francisco 7
30 Houston 6
31 Baltimore 5
32 Arizona 5

Hi ATC


Thanks for your ratings and I really appreciate you taking the time to give them to us. wonder if you could explain how these are calculated in a brief manner.

Just looking at them (I know its early days and distorted by a small sample) but they seem very offensivly slanted. For example you have Cinci way out in front. No argument that their offensive numbers are good but I have them in the middle of the pack defensively thus my ranking has them 5th. I have Tampa Bay about where you have them on offense but they are my best defensive team. Thus on my ratings they are 4th.
Mine might be too defensively slanted.

Anyway, thanks again for the post. I post this just as a discussion for after my dismal performance this weekend I probably shouldn't say anything. But, I am always interested in a different well researched approach such as yours. And I think, any discussion is better than none.
thanks again

Thanks for taking an interest in my power ratings Karla.
The Ratings are produced by taking into account the final scores in all games played this season and home ground advantage only. This information is imported into my computer program which then processes the data to come up with a rating for each side which will best fit the data(a form of least squares regression) It basically compares the strength of a teams win on how strong their opposition are according to the data. You will find that once the results from week 4 are added the ratings will change significantly. They wont take proper shape until about week 10. The numbers dont have any meaning but the difference in ratings between teams do. It predicts Cincinatti to be a 42 point better team than Arizona (47-5=42) which is obviously not the case but this is caused because of a lack of data. You will find that the values will become more compressed as the season unfolds. If Cincinatti was playing Arizona at home the predicted margin would be 49 and if it was away it would be 35 (allowing for a 7 point home field advantage which has been factored into the model). Predicted probabilities for matches can be evaluated from the ratings but again wont become robust until later on in the season.

Cheers
ATC

Hey ATC, nice job with the stats.

Coupla questions if ya don't mind.
1. What kind of test statistics did this model produce, specifically sigificance level (allowing for small sample) and R-squared.
2. Could you improve the model by breaking the collective data into smaller individual parts? Could you perhaps add more variables? Qtr by qtr: scoring, rushing, passing, defence stats, sacks, picks, turnovers - those sorts of things.
3. You say in your earlier post, KC vs Eagles that you would add 7 points to your KC total as they are playing at home, but then later say that home ground advantage has been factored into the model. Could you clarify please? The reason i ask is if home ground advantage has been factored in then the KC/Eagles game would score:

KC: 30
Eagles:32
Rather than 37/32.

{Afterthought****...Looking at early betting markets, the Eagles are slight underdogs at +2. Using this model (30/32) taking the Eagles on the road with the points would be the way to go from a betting perspective....

I would think this is a more accurate reflection of the game at face value. I don't mean to seem critical, far from it. This is an excellent idea, your model intrigues me and i would love to hear more.
Thanks.

__________________
I like Bing Lee.

Hi Mad,

The program I am using is a form of least square regression where i input equations into my program in the form a-b=c. An example of this would be
Denver - Kansas City = 20 . This was the score differential between these two teams in Week 3. After week 3 there is 46 equations of this type which are inputed into the program. The computer then assigns numbers to each of the 32 variables (the 32 NFL teams) and then alters these values to minimise the error. Home ground advantage is taken into account as the average winning margin of home sides this season (about 6.5 at the moment). So the above equation would be changed to Denver - Kansas City = 13.5. If for example the model fitted Denver with a rating of 40 and Kansas City 20 then the error margin would be 20-13.5=6.5. The program minimises these error values across all equations. As you could imagine the R-squared value would be low and the p-value would be high suggesting a bad fit but im sure these values will improve as more equations are added.

Hope this makes sense.

Cheers
ATC

Hi ATC

Couple of questions if you don't mind.
Is the program something you bought or did you design it yourself?

Are the scores the only thing you input?

ta
Karla

Hi Karla,

I designed the program myself as a way of rating teams within a biased schedule. Obviously it is unfair to look at simple win loss ratios for each team at a given time when comparing them because teams play teams of different strengths. An example of this is say Oakland who have won 0 from 3 but have been beaten in close games by 3 reasonably good teams. My program allows for this uneveness in the draw and rates teams according to the strength of their opposition based on the score differential and the Home field advantage. It becomes an accurate predictor as the season progresses.
It is possible to add more factors but at this stage im keeping it simple to only take into account final scores and home field advantage. Assumptions being made are - 1) that a teams strength is constant throughout the season
2) Home field advantage is the same everywhere for every team 3) A scoreline of 10-6 is the same as 38-34 and 4) A teams performance is measured on their winning margin. Obviously a lot of these assumptions are not true.

Cheers
ATC

On a side note while I'm thinking about it, the Philadelphia Eagles are very beat up right now. QB McNabb has NUMEROUS injuries and doesn't look right and the highest paid place kicker in the NFL, David Akers is out with a groin injury. The Akers injury happened on th opening kickoff of the Oakland game and the dumb Philly coach (Andy Reid) kept Akers in the whole game. Now, Akers did kick the winning FG, but he may have made the injury way worse and the rcovery time alot longer by playing. JUST WANTED TO LET EVERYONE KNOW ABOUT THESE INJURIES BEFORE MAKING A DECISION ON THE GAME.