College Football Rankings: Do the Computers Know Best?

Abstract

The bowl-championship-series (BCS) committee uses 10 ranking schemes, including eight computer rankings, to select college football teams for bowl-championship-series bowl games, including the national championship game. The large financial benefits of participating in BCS bowl games make it imperative that the selection process accurately select the best teams. I evaluated the performance of the 10 ranking schemes the BCS committee used during the 1999 and 2000 seasons to select bowl teams. I found that almost all are equally accurate, but the Seattle Times scheme clearly underperforms the others. In addition, two proposed changes to the BCS selection formula, (1) to prohibit computer ranking schemes from considering the margin of victory in their rankings, and (2) to include explicitly the outcomes of head-to-head games among teams being considered for BCS bowls, could do more harm than good and could decrease the likelihood of the committee selecting the best teams for the BCS bowls. (Recreation and sports. Statistics: data analysis.

Football Result Prediction with Bayesian Network in Spanish League-Barcelona Team

“…Koning et al [13] and Kuonen [14] specifically address tournament-specific issues, and/or directly model football tournaments. Martinich [15], Amor, et al [16], and Yang et al [17] investigated other sports such as American football, major league baseball.…”

Diagrammatic perturbation methods in networks and sports ranking combinatorics

“…Now we have an iterative prescription for calculating ψ of any diagram: from the negative of the logarithm of 1 plus the terms that represent itself and its subgraphs in Eq. (22), subtract the ψ's of the subgraphs. As an example, let us use this rule to evaluate ψ[Θ ij Θ km Θ lq ].…”

“…One often evaluates the quality of an algorithm by counting the violations in the final ranking it produces [22]. Any MVR will perform the best in this respect by definition, but in a recent study of a particular MVR that was mentioned earlier was shown to have performed as well as or better than other ranking methods in other tests as well [6].…”

Reweighting the Bowl Championship Series

“…Most statistical investigation of the BCS has involved proposing new algorithms for ranking teams, discussing what should and should not be included as inputs to those systems, and metrics for comparing computer ratings (Callaghan et al, 2007;Martinich, 2002;Coleman, 2005;Mease, 2003). A thoughtful discussion of the statistical and contextual issues surrounding the development of a ranking or rating system can be found in Stern et al (2004).…”

“…The appeal of this approach is that it both informs us about the performance of existing components and provides a policy prescription. For example, although Martinich (2002) compared the accuracy of the 10 components that made up the BCS in the 1999 and 2000 seasons, it is not clear how, were it so inclined, the BCS should have incorporated the results. Should only the most accurate components be used for the overall BCS rankings?…”