clock menu more-arrow no yes

Filed under:

New and Improved Ratings, Etc.

New, 7 comments

The Ratings

Well, they're definitely new and when you start out really low then you're probably going to improve. So I've got that going for me.

Which is nice.

The previous version of the ratings had a couple of issues that bothered me. The first was that the composite ratings would occasionally have a team ranked higher or lower than they were in either the NMV or PMV ratings set. That kind of result was illogical and was frankly the result of what I consider to be some ad hoc input previously incorporated into the ratings. The ad hoc aspect of the ratings calculations has been removed in this new version and all inputs are either logically defensible or driven by the data.

The second issue was that no aspect of the ratings was either intended or suitable for predictive analysis of future matchups. My original intent when getting into generating computer ratings was actually to produce them for Texas High School Football. And with the high school football ratings I intended only to rank teams based on their performance over the course of the season. Consistent sources for accurate score information was too tedious to track down every year, so I no longer publish those, but I did become more interested in Sagarin's predictor model and the previously published Massey predictions.

So the new version of the ratings will contain both the "Season Rankings" section which I consider to be the better one to look at to rank teams based on the season performance while the "Power Ratings" section is intended to predict future outcomes based on that past performance. The Season Rankings version is still based on the previous idea of assigning a game outcome value that substitutes as the ratings' guess for the probability that the winning team will win a rematch. The PMV ratings are calculated first where that probability can vary from just above 50% to just shy of 100%. The NMV ratings are calculated next using the PMV win correlation as the game outcome value for the winner of each contest. The actual margin of victory from each game is not considered. The final rating is a composite of these two ratings.

The Power Ratings is the new aspect and determining how to calculate this part of the ratings was based on discussions on various boards. In more than one place recently the Big 12 offense versus SEC defense debate has been raging. In support of the Big 12 teams' defensive abilities, several fans have compared the number of points and yards allowed by Texas and others compared to those teams' average outputs. The first time I saw this kind of analysis it was on the old GoBig12 board and the late PhxHorn was using them to forecast the Cotton Bowl against Mississippi State. The only shortcoming then, of course, was that doing the calculations by hand made it too time-consuming, just short of impossible, to account for a teams' opponents and their opponents' opponents. What the Power Ratings do is essentially the same analysis but they use the score of every game played and iterate as many times as necessary until the ratings converge to an acceptable level.

So, here are the All Teams ratings and the D-1A only ratings. The left half of each output is the Season Rankings and the right half is the Power Ratings. In order to predict the margin of an upcoming game, simply subtract a team's "Pow" number from their opponent's and account for the overall homefield advantage as necessary. In order to predict the score of a specific team in the contest, start with their "Off" rating and subtract the opponent's "Def" rating. Add half the homefield advantage to the home team's score and subtract half from the visiting team's score. All of these calculations have to be done from the "All Teams" page.

The classification-specific pages have normalized values for all of these power rating figures. This will allow me to compare units across seasons, e.g., offer an answer to the question of which offense was the greatest of all time. I will be calculating historical seasons as time permits. Something to keep in mind is that the Offense and Defense ratings are based on point totals. Therefore dominating special teams units can help both units or just one in the case of a dominant kick returner, for example. Let me know if you see any errors or have any questions.

Quick Title Belt Update

Texas won the belt on Saturday.