"Power Ratings are numerical values assigned to teams that intend to provide the basis for what the spread should be between two teams at a neutral site. They are updated every week during the season."
That could function as a description of my rankings as well. To be honest, I don't how they derive the rankings at that site, but I figured I would put their rankings and mine to the test and see which set was best at predicting how the market sets the point spreads for upcoming games.
The latest set of College Football rankings at Linemakers is dated November 21, 2012. I've got my rankings archived, so I pulled my rankings from the same day. I then calculated the predicted point spreads for this weekend's upcoming games (Dec. 1). The reason I didn't use the point spreads from last week is that both their rankings and mine had a chance to "peek" at the point spreads, so it wouldn't be a proper test of predictive accuracy.
The table below summarizes the results:
There is variation from game to game, but average accuracy was about the same. The mean absolute prediction error for my rankings was 2.1 points. The mean absolute prediction error for Linemakers was 2.2 points. The Linemakers prediction was closer for a couple of the high profile matchups this weekend (Georgia v. Alabama and Wisconsin v. Nebraska).
Interestingly, if I take the average of my prediction and the Linemakers prediction, that results in a mean absolute prediction error of 1.7. The power of ensemble learning.