BCS mess

» Posted by Vince on December 11, 2007«

The big question once again is who should be in the BCS championship game. I fully admit that I don't know. The point of this post is that nobody can know. The data do not support any conclusion.

However, we do know that the voters (collectively) in the two human polls have no integrity because LSU hopped over VT even though both teams won conference championships. Moreover, VT beat a tougher opponent more convincingly. Clearly, there is a contradiction between the last two regular season polls: both cannot be true. This begs the question should we believe either poll? It is very hard to treat the human polls with much seriousness after this Kerry-like flip-flop.

What I would rather point out is that there is insufficient data to rank teams. There are 100+ Div 1A teams, each plays 12 opponents. Thus, there is no direct data linking a team with nearly 90% of the other teams. Moreover, this information is clustered because teams play in conferences making it even harder to compare teams in difference conferences. Finally, the better team does not always win a game. Therefore, you cannot rate teams merely by the outcome of a single game.

The table below shows the inter-conference games between the BCS teams. (Not including ND because there is no point in piling them on this year.) Each conferences record against another conference is shown in the corresponding row. For example, the 1-0 in the first row indicates that the ACC won the only game with the Big 10 (Duke over Northwestern).

The point to take away from this table is that there are very few inter-conference games. For example, SEC didn't play the Big 10, the ACC didn't play the PAC 10, and the Big 12 didn't play the Big East. Only 3 out of 15 inter-conference match-ups had more than 3 games (interestingly, all involve the ACC). Simply put: there is insufficient inter-conference data to draw a definitive conclusion. The data is not statistically significant.

Let's look at one problem. LSU and VT had identical records (11-2; BCS conference champs) but LSU stomped VT in the second game of the season so it is reasonable to rate them higher than the VT. Except that 4 of 6 computer indexes rate VT higher. There must have been some other evidence suggesting VT is stronger than LSU. They did not play a common opponent, so we must assume that VT's wins are of higher overall quality. VT's wins are against ACC teams and weak non-BCS teams, therefore, the computers must have concluded that the ACC is a stronger conference than the SEC. One wonders if Clemson had not beaten South Carolina on a last second FG would the indexes have ranked LSU ahead of VT? On the other hand, I cannot explain why Sagarin ranks VT above LSU and ranks the SEC first and the ACC distant fifth.

A similar problem occurs when comparing Oklahoma and LSU. They didn't play each other and there were only 3 inter-conference games. Basically, those two teams are incomparable. Nevertheless, this is better than comparing Ohio State and LSU. No Big 10 team played an SEC team. The closest comparison we can find is OSU beat Kent State, which lost to Kentucky, which beat LSU. However, what to conclude from that is anyone's guess. There is a saying that no information is better than bad information. The BCS might benefit from considering this.

There is more to say, which will come in later posts.