Wednesday, January 8, 2014

Ranking College Football Programs

If nothing else, sports enthusiasm is great for generating passionate debate about esoteric topics and hypothetical arguments. Within this realm lies the ever popular, hair-splitting activity of determining who is the greatest of all time. And college football is perhaps the most well attended of these masters of the universe feuds--for there can be only one.

You don't have to search for long to find many lists each with there own methodology. Here is one. Here is another. Here yet another. These all came from a quick Google search and all happen to have the Sooners at the top, but you probably can find lists with different results. 

Thinking about this topic myself but trying not to take too seriously any particular method of ranking (including my own), I have developed a ranking that is different from any I've ever come across. However, I believe it is more elegant and more defensible. My method uses historical average margin of victory to determine who's better, who's best. 

The reason this method may have a lot of credence is margin of victory (MoV) is a powerful determinant in predicting college football outcomes. In fact a simple MoV model adjusted for field neutrality can explain about 63% of the outcome in a typical FBS college football game. Keep in mind that this is not just predicting who will win but also by how much.*

So what did I do? Using the amazing data source from James Howell's page along with some supplemental data from the NCAA and College Football Stats, I compiled all the scores from all the college football games involving a Division I-A school for the past 43 years. Why start in 1970? Because that is when I started . . . but there is some foundation as that is around the beginning of the modern era of college football. Around that time saw the evolution of dynamic offensive strategies, the rise and proliferation of black athletes, the end of one platoon football (1966), and the beginnings of a more intrusive NCAA in the name of competitiveness. 

The results:

Since 1970, Nebraska is the clear leader. 


I include it through the top 26 since some would want to exclude Boise State from the rankings due to limited games played in the top level of college football.

Since 1998 (the BCS era):



Check out the workbook for yourself tweaking the constraints as you see fit. You can change the time period examined, the minimum number of games played to be included in the rankings, and the statistic you wish to sort by. 



*The out of sample prediction accuracy falls off some and the magnitude of the variance of outcomes matters such that the ability for such a model to beat Vegas (>~53% necessary prediction accuracy against the spread) is low (actual prediction accuracy ATS of between 51%-58%).