ryanratings: July 2011

Sunday, July 17, 2011

Updated look

I've updated the look of the RyanRatings site today. The algorithm remains the same, though I'm playing with a couple of radically different ones as well. I may roll them out as the 2011-2012 season approaches/progresses, in addition, of course, to my now classic ratings. That is, if I can come up with an adequate, clear way of displaying them.

New posts are going to have comments enabled. We'll see how that goes.

I've started up a personal blog, which may, very occasionally, comment on sports and algorithms, though I haven't done much of that to date. The RyanRatings blog—that's what you're reading now—will feature a minimum of editorializing. No worries though—there's plenty of editorializing to go around over on the Slightbit blog.

How it works

For each team, two raw ratings are calculated; one dependent only upon wins and losses (the "W/L rating"), and the other dependent upon margin of victory (the "MOV rating"). The W/L rating for each team is simply the number of games won divided by the number of games played. The MOV rating is more complicated. I assume that a team's expected margin of victory can be modeled by a normal distribution. The distribution's mean is set equal to the average margin of victory achieved by the team over all of its games in the season. The distribution's variance is set equal to the variance of the team's margin of victory in each of its games. Based on this distribution, I calculate the probability that the team will achieve a positive margin of victory against generic opposition. This probability is the team's MOV rating.

The true "base rating" is then calculated by combining the W/L rating with the MOV rating. This combination is intended to dampen the effects of a team's running up the score, while also admitting the difference between a close game and a blowout.

Finally, the team's base rating is adjusted by the strength of its schedule. In the first round of calculations, schedule strength is set equal to the average base rating of each of the team's opponents. The team's first adjusted base rating (FABR) is then calculated by adjusting its base rating up or down depending on how strong or how weak its opposition was, as determined by this metric. In the next round, the FABRs of the team's opposition are averaged to calculate strength of schedule, and the team's base rating is again adjusted accordingly to determine the team's second adjusted base rating. I repeat this process twenty-five times so that the ratings will converge. The twenty-fifth adjusted base rating is the rating reported here.

With respect to overtime games, in calculating the MOV rating, I use the score at the end of four quarters, but in calculating the W/L rating, I treat an overtime win or loss just as I would any other win or loss.

New for 2009: Until this season, I didn't count games against non-FBS competition for any purpose. Starting in 2009, I treat games between FBS and non-FBS teams the same way as I treat games between two FBS teams, except that all non-FBS teams are aggregated for purposes of calculating schedule strength.

Pages

Sunday, July 17, 2011

Updated look