Key Stat Rankings thru Week 4

[Patler]

Fosco;

I understand what you are doing, but for your model to have any validity, and for your weighting to have accuracy, you need to take about 10 years worth of stats and plug them in. Anything less will be skewed by anomalies. Throwing out the high and low helps, but more data would be better.

[woodbuck27]

Interesting thread. I'd mostly agree with your metrics--although first down efficiency oftentimes is about as useful as time of possession. If this was a weighted formula, then you've got yourself a model.

Again, I could alter and remove the top/bottom 3 teams where the model is incorrect and then weight the ranks (T/O Margin being more predictive) - if you had to add a metric - which would you add besides the above - and how would you weight the metrics? Your expert opinion is appreciated.

Fosco

If I understand your ? Fosco33...

Wouldn't you want to go with number of 1st downs/game as the metric ...along with T/O Margin... as a predictor of:

Which team will win in a certain matchup?

[Fosco33]

Interesting thread. I'd mostly agree with your metrics--although first down efficiency oftentimes is about as useful as time of possession. If this was a weighted formula, then you've got yourself a model.

Again, I could alter and remove the top/bottom 3 teams where the model is incorrect and then weight the ranks (T/O Margin being more predictive) - if you had to add a metric - which would you add besides the above - and how would you weight the metrics? Your expert opinion is appreciated.

Fosco

If I understand your ? Fosco33...

Wouldn't you want to go with number of 1st downs/game as the metric ...along with T/O Margin... as a predictor of:

Which team will win in a certain matchup?

I've got # FDs/game as a metric.

Patler, I agree a few more years of data would help. I'll hunt around and develop a model that people are free to utlilze.

[Patler]

Are Y/P and FD/G really measuring anything different? Both are simply measuring offensive efficiency. I'm not sure how accurate your model will be as a predictor for the outcome of a game between a team with a great offense and no defense angainst a team with a great defense and an average offense. In a game like that a "predictor" needs to factor in team defenses, and the effect of one defense against the other offense.

As a point of consideration, try running the same table but on the defensive side. Y/P allowed on defense, with lowest being #1, etc. Then combine the offensive and defensive rankings for your overall team rank.

[Fosco33]

Are Y/P and FD/G really measuring anything different? Both are simply measuring offensive efficiency. I'm not sure how accurate your model will be as a predictor for the outcome of a game between a team with a great offense and no defense angainst a team with a great defense and an average offense. In a game like that a "predictor" needs to factor in team defenses, and the effect of one defense against the other offense.

As a point of consideration, try running the same table but on the defensive side. Y/P allowed on defense, with lowest being #1, etc. Then combine the offensive and defensive rankings for your overall team rank.

Ok, in looking at that further both Y/P and FD/G look at offensive efficiency but it appears Y/P is a closer predictor to wins than FD/G.

I'll take off FD/G and add a defense indicator like Y/P allowed. I'll also add Def 3rd down %. Since T/O margin affects both offense and defense, I'll keep that in there for this model.

I'll weight the metric's rankings as: .15 (Off Y/P, Off 3rd down %, Def Y/P and Def 3rd down %), .1 for TOP and .3 for T/O Margin.

I'll also run this back through for a number of years - hopefully I can get data enough years back. Then anyone interested can comment further.

[Fosco33]

Are Y/P and FD/G really measuring anything different? Both are simply measuring offensive efficiency. I'm not sure how accurate your model will be as a predictor for the outcome of a game between a team with a great offense and no defense angainst a team with a great defense and an average offense. In a game like that a "predictor" needs to factor in team defenses, and the effect of one defense against the other offense.

As a point of consideration, try running the same table but on the defensive side. Y/P allowed on defense, with lowest being #1, etc. Then combine the offensive and defensive rankings for your overall team rank.

Ok, in looking at that further both Y/P and FD/G look at offensive efficiency but it appears Y/P is a closer predictor to wins than FD/G.

I'll take off FD/G and add a defense indicator like Y/P allowed. I'll also add Def 3rd down %. Since T/O margin affects both offense and defense, I'll keep that in there for this model.

I'll weight the metric's rankings as: .15 (Off Y/P, Off 3rd down %, Def Y/P and Def 3rd down %), .1 for TOP and .3 for T/O Margin.

I'll also run this back through for a number of years - hopefully I can get data enough years back. Then anyone interested can comment further.

I made these changes to my initial charts for 03-06. Thoughts?

[Fosco33]

I took 2003-2005 data and eliminated the top 10 variances (either close (absolute) or far from prediction) and have shown the average # of wins and the range for that type of team by their overall ranking.

For this year's Packer team with a 19 ranking, they'll average 5 wins with a range of 4-6. They'd need to improve up to a 13-16 ranking or better to have a shot at 9 wins and a wild card birth.