Fosco's Team Rankings

[Fosco33]

Waldo, that is a pretty complex system. You must have a background in statistics as I have a lot of background in it and that was tough for me to follow.

Engineering. That is not statistics.

Similar concepts are used to model indeterminate structures (structures that cannot be solved by a basic linear equation, instead a circular calculation is required). Or flow systems like drainage basins and traffic networks, both of which are also circular equations and require best fit approximate solutions.

You ever use the Erlang C modeling for traffic. I used it once to develop a gigantic call center routing system.

[Fosco33]

Interesting. Where do you find updated SOS... I can only find preseason.

I was also considering a season summary vs. recent history (last 4 games). But I'd have to track each individual teams stats (by game) and have an ongoing list. It'd take me way to much time for something that may actually show a very similar trend. It basically takes me 30 minutes or less to pull together my info now - and I adjust the stats in my formula to limit the impacts of straight arithmetic vs. more using algorithims/etc. As I use this to help me smooth out vegas spreads, I still have to account for non-stat issues (momentum, injuries, weather, etc.). If my sole job was to do this, I'd get more complicated/accurate.

Everything is calculated from the basic data set of weekly game scores.

So you have to create a separate tracker - I'm looking for info available online. Oh well.

[bobblehead]

Waldo, that is a pretty complex system. You must have a background in statistics as I have a lot of background in it and that was tough for me to follow.

Engineering. That is not statistics.

Similar concepts are used to model indeterminate structures (structures that cannot be solved by a basic linear equation, instead a circular calculation is required). Or flow systems like drainage basins and traffic networks, both of which are also circular equations and require best fit approximate solutions.

yea, that makes sense...and you did mention engineering in the OP. The beauty is that your method likely works better for considering strength of competition because it will adjust and readjust each teams "strength" as it cycles through....I like it.

[Waldo]

Here is last weeks data run.

I've been looking for a better way of presenting it, this is the first gfx try instead of typed.



Rk = Rank
Pre = Prev
Net = Off - Def, Net Scoring
Off = Offense (opponent adjusted ppg)
O Rk = Rank of offense
Def = Defense (opponent adjusted ppg)
D Rk = Rank of defense
Chg = Difference between most recent 4 weeks and full season
C Rk = Rank of the change, positive to negative

Strength of schedule:

I used the power ranks for this, instead of W-L or quality opponents, rank is used as the data to compute strength of schedule. The average rank of their opponents (16 is average). These are calculated from the power ranks, no part of the power rankings is calculated from this strength of schedule.



P Rk = Previous Rank
F Rk = Future Rank

Estimated Record:

The record estimate uses the power ranks, existing wins, and future schedule to calculate the outcome of all remaining games, and assumes a probability of that outcome occurring.

It is assumed:
If a team is favored by 14+ points, I assumed that they win 100% of the time.
If a team is favored between 7 and 14 points, I assumed that they win 80% of the time.
If a teams is favored between 0 and 7 points, I assumed that they win 60% of the time.
If a team is not favored between 0 and 7 points, I assumed that they win 40% of the time.
If a team is not favored between 7 and 14 points, I assumed that they win 20% of the time.