Kiko13 College Football Rankings Theory

Last Update: 02/25/03 06:39 PM

Theory Behind the Rankings

For starters, I'd like to say that I am not all that happy with my rankings. In the fall of 2000, I came up with a statistical model over the course of about two weeks, it seemed to work okay and I haven't had time to really do a lot of research to come up with a better one yet. I have comments about my system after I explain the theory.

I don't want to go into deep detail about a statistical model and bore you to death so I will give you the basics of what my model is doing. I have 5 different models which all use different information. The 5 models all include the team playing, who they're playing against and on who's field they are playing on. Each model rates the teams on 5 different things:

1. Difference in the score of the game (awards teams for beating another team by a lot).

2. Square root of the difference in the score of the game (as the difference goes up the team is not awarded quite as much as they were when the game was close).

3. Offense (awarded for scoring a lot).

4. Defense (awarded for giving up little).

5. Winning (awarded for finding a way to win regardless of the score).

Each team gets a rating for the above 5 categories. Then they get an average rating by giving categories 1-4 a weight of 1/8 and category 5 a weight of 1/2. Therefore half of a teams rating is based on the score of the game and the other half is based on whether they won or not. Since the models include a team's opponent for each game, the ratings automatically take into account strength of schedule.

Comments:

The ratings can sometimes be overpowered by strength of schedule. Personally I'm all for a team playing a tough schedule, but at times it can take it into account too much. I once added a team to the database where their schedule included the top 10 teams and lost by 1 point to each of those teams. Even though they were 0-10 they ended up being a top 10 team themselves. I also think I may be weighting the score categories too much. Seems like teams that run up the score do better.

Like I said, I have not had much time to do a lot of in depth research. Hopefully some day I will and will be able to come up with a better model. Although for as little time that I have put into coming up with my system, overall I think it does a pretty decent job. It just needs some fine tuning.

One thing I noticed during the 2002 season, is that it seems difficult for a team to do well in the rankings which is defensively oriented. For instance, Ohio State won 7 games by 7 points or less in 2002. However, they never lost. There is something to be said about finding ways to win regardless of how it is done. A team which has a very strong defense, but has a somewhat weak offense will win a lot, but may not have a large margin of victory. Suppose 2 teams have the same record and a similar strength of schedule. What team would be considered better? A team that is extremely powerful on one side of the ball while the other side is average or a strong balanced team, but neither side being nearly as strong as the first team's powerful side? Using the 2003 Fiesta Bowl as an example, it looks like having one strong side would be better. Of course that is only one game and the question is being raised because rankings in general are determining what teams are better than other teams that haven't even played eachother. Because of things like this I am considering going the BCS route and eliminating margin of victory.

Please e-mail any comments or questions about this page to kyle@kiko13.com.