Originally posted on 10/15/2011:

For those of you using Formulawiz's Marg Diff system you may have wondered why we use the setting of 10 games back to generate the data. It appears to be the right setting. I ran a few scenarios and thought I would share the results.

Normal settings for collecting Marg Diff data is to leave it at 10 games back and leave the team settings on "Home" and "Away" which are the default settings (do not use Overall). I tested setting the games to 3 weeks back, 5 weeks back, and finally 10 weeks back with the selection set on "Overall" for each team instead of "Home" and "Away". Note: the 3 and 5 weeks back data had the tabs set for "Home" and "Away", like normal.

  • Back tested CFB data for weeks 9/24/11, 10/1/11, 10/8/11. That's as much data as I could stand collecting and entering into a spreadsheet (remember this is a manual process).
  • Used the standard filters: 3 pt overlay, 15 pt spread limit, no small colleges


Results: Net improvement in Win% Vs the "normal" settings
These are the average results for all 3 weeks of data (9/24/11, 10/1/11, 10/8/11).
  • 3 week back: -1%
  • 5 weeks back: -6%
  • 10 weeks back: -5% (remember this one used OVERALL setting instead of H/A)


Although none of the tests showed an improvement in results Vs the normal settings, hopefully it helps answer some questions about using 10 games back and the "Home" and "Away" as well as keep someone else from having to spend 6-8 hours checking it.

If anyone else is running tests to tweak the system to improve performance, I hope you will share it on the thread so we don't duplicate work.

Good Luck,