I use the win-loss record of the team, total games played, conference performance by a team, apply a factor I call "strength of conference" (also calculated), and the points for/against ratio of a team. The R-score number itself is pretty much meaningless except it gives a magnitude of strength or weakness of a team. The latest iteration of my method employs the base-e logarithm so that good teams don't have extraordinarily high R-scores, and weak teams end up with negative R-scores rather than decimalized scores.
The methodology itself, of course, is not publicized. However ALL the rankings are derived from the weekly standings as published everywhere (I import the data from sports illustrated site each Sunday morning). Each team's cumulative win-loss gives a ratio (same method as Colley starts with), then I multiply by ptsF/ptsA ratio, then adjust for the strength of conference (my own invention, I think; idea being that weak conference wins mean less than strong conference wins since the bulk of games played are within conference opponents). Out of conference victories adjust the R-score further, multiplied by #games played (idea being that a 12-0 team has done better than a 10-0 team).
This methodology assumes at the beginning of the season that every team is equal, and that each conference is of equal strength. Performance of teams affects their R-score, and unfairly or not the R-score can change up or down depending on the performance of your own conference's other teams too. Ranks are volatile for the first few games of the season. Expect big shifts week to week until every team has played at least one home game, one away game, one conference game and one non-conference game.
IMPORTANT CHANGE IN METHODOLOGY: A number of critics have correctly pointed out that my rankings are over-rewarding teams with weak Strength of Schedule. I've come up with a SOS adjustment that I think will make these rankings reflect the schedule differences better. As a result, effective Nov 11, 2007, the R-score rankings will not be comparable to those posted in the past because of a major shift in methodology (I hope an improvement). I welcome criticism!
It's been fun to develop, and I've tweaked the methodology once or twice. I originally developed this for the Canadian Football League Power Rankings (only 8 teams, and everyone plays everyone else at least twice); and it's been intellectually a lot of fun.
