Lindy's Five Essential Websites (Non-Major Media) for 2013
[+] Team Summaries

Sunday, October 7, 2012

Remodeling CFBTN

I am constantly adjusting the model that underlies the predictions and analysis on CFBTN. From time to time the model gets more than a tweak. As I get access to new data or develop a new approach, I run a few tests to see if they will make the model more robust and more efficient. If so I integrate them into the model. If not, they are discarded in the dust bin of history.

Beginning in the offseason and continuing through the season I've worked in a few other changes to the model. Until now, the posted results reflect the old approach, but beginning this week I will be posting the results from the new and improved model. While you may not notice much difference from the outside, I felt it was important to detail where some of these changes would appear. I should also note that I tested tweaks to the model in many other areas, but only in these respects were the predictions and analyses improved.

- This last week I adopted a new method for estimating a team's tempo, or how rapidly they run plays on offense. Ideally I could be out there with a stopwatch to time each play, but this new approach is the next best thing.

- The offense and defense unit ratings are now adjusted for possessions per game. No longer are offenses rewarded and defenses punished for getting more possessions per game. The team power ratings are unaffected by the change (the change to the offense and defense is zero sum).

 - With the change in the tempo metric, estimates on the number of plays run by each team per game will also be affected. This, in turn, will affect total yardage. This could significantly affect the estimate on the total points scored, but will have minimal impact on the estimating margin of victory. These changes will also impact fantasy projections.

No comments:

Post a Comment