Saturday, December 27, 2008

Bowl trouble

Eight games into the bowl season and our modified Golden rankings are 1-7 in predicting outcomes. Now, a couple of these games had teams that were within just a few places of each other, but the results are still troubling.

So the question is, how do we modify the rankings next to get a more accurate prediction model? When you look at our seven losses, you see that five of the seven winners in those games had higher schedule strengths than the team that was predicted to win. This suggests that perhaps a tweak to the formula to add more emphasis to the opponent opponent's wins. Perhaps 0.4 points per opponent opponent's wins as opposed to the current 0.2 points. This is something that Golden had mentioned he might do with his version of the rankings, so it is worth checking out after all the bowl games are done.

Another modification that I had considered involves the FCS teams. Right now wins over FCS teams are discounted based on the winning percentage of FCS teams vs FBS teams versus the winning percentage of FBS teams against FBS teams (.500 by definition). it occurred to me that the FCs winning percentage should be compared to the road winning percentage of FBS teams, since FCS teams never get home games against FBS teams. I will also investigate this after the season.