Jessup Pope (JP) College Football Rankings: Week Beginning November 8, 2015

Here are the JP College Football Rankings after the 10th week of the season. (See our first post for a brief explanation of our methodology.)

Our top two teams, Notre Dame and Clemson, remain unchanged and Alabama’s big win over LSU moved them from 6th to 3rd place. Notice that ours and the College Football Playoff (CFP) rankings agree on 3 out of the 4 top. Yet here is one way we can see a dramatic difference between our ranking system and the more traditional polls and rankings.

In our rankings last week, Michigan State and Michigan were 9 and 10, respectively. This week, Michigan State, who lost, moved up 2 positions and Michigan, who won, moved down 4 places. This is because the former narrowly lost on the road to a decent Nebraska team (#19) whereas the latter won at home against a very weak Rutgers side (#104). Similar movements occurred all throughout our rankings. Another major difference between ours and more traditional ranking systems is that ours is forward-looking, meaning it is meant to predict future outcomes whereas more traditional systems are primarily designed to reward past performance. This is why a 3 loss USC is ranked 4th in our system (#3 last week) yet they are outside the top 25 of the AP poll. Baylor remains unchanged from last week at #60. A good performance against a very strong OU team (#5) this Saturday will probably see them rise, even with a loss. The top 50 teams are listed at the end of this blog post.
Relationship with courses we teach
One might reasonably ask what this has to do with the courses we teach. It turns out, quite a lot! For example, one of the courses I teach, data mining, initially involves separating data into two datasets or partitions. We then run our algorithm on the first partition to teach it. Next, we examine how well it forecasts the results in the second partition. This is exactly how we came up with these rankings. The benefit of this partitioning process is that when we say that we can beat the Vegas line 56% of the time, it means we are beating it on new data, not merely data that our algorithm used to learn. It is actually very easy to “predict” results that you are using to learn; what is hard is predicting or rather forecasting results that are new. In a later post we’ll discuss some of these ideas and their relationship with what we teach in greater detail. We will also examine other research applications of this ranking system.