Bowl Season

Our ranking system is not meant to be predictive. Instead, it is meant simply to rank teams by the success of their overall season. We'll be coming out with our post-bowl rankings soon, but first, we do annually take our rankings and use them in the ESPN and Yahoo bowl prediction contests, just to see how they measure up by using the rankings to a) predict the winner of each game - although we know there will be upsets, we "predict" that the favorite according to our ranking system will win each game, and b) assign "confidence points" to each game by using our ranking system to determine the likelihood of an upset and giving more points to the least likely upset - the likeliness of an upset determined by the difference in ranking of the two teams.

Using this simple system, once again, our rankings have proven to be reliable. Yes, 14 of the 35 games ended up as upsets according to our pre-bowl rankings. This is no surprise. Upsets happen every week in our ranking system. Plus, this system does not take into account team motivation - is the team disappointed by the bowl they ended up playing in? - or factors such as coaching changes. No, we did not win either of the contests we entered. There are many thousands of entries in each of these contests.

However, we did perform very well. How well? As expected, most of the upsets were between teams we had closely ranked, meaning we did not rate those games as being worth very many points, so most of the 21 games the system predicted correctly were high value games. Out of a possible 630 points, the ranking system scored 440 points, which is quite good. How good? It outperformed 96% of all entries in the Yahoo contest and 98% of all entries in the ESPN contest. The system especially performed well on games on or after New Years Day - or in other words the bowl games that appear most likely that both teams would be happy and motivated to be playing in. In fact, the system was only wrong on two games on or after New Years Day - both of which were rated very low in confidence points (1 point and 3 points respectively) and therefore are not considered surprising upsets in any sense. In fact, in one stretch, the system correctly picked 9 games in a row correctly, including all of the New Years Day bowl games.

What do we take from this? Not much. While the system seems to be based on a sound premise - the idea that teams with more successful regular seasons will typically perform better in the bowls as well - once again, this ranking system was not meant to be predictive. 14 of the 35 games still turned out to be upsets. But the system did seem to accurately reflect the likelihood of an upset in these bowl games based on the teams' performance during the regular season. Last year, one independent organization compared the BCS system computer rankings to bowl team performance and found that those computer rankings predicted only about half of the games correctly - roughly the same as you would expect if you picked the games by flipping a coin. Well, we outperformed that low standard. But that is no surprise. We feel like the premise of our ranking system is much more sound than any computer ranking system that is not allowed to take into account important aspects such as score differential in games. We do feel that these results tend to support the premise of our ranking system - just as they did last year.

But then again, we didn't think we really needed that justification in the first place. The results of our ranking system speak for themselves. Final rankings for the 2010 season coming soon. (Remember, Auburn was ranked behind TCU in the end of the regular season rankings.) Was the win over Oregon enough to push the Tigers past the Horned Frogs in the final rankings or will we award our national championship to a different team than this year's BCS/human polls? Stayed tuned and see...