College Football Analytics: Predicting The CFP Rankings

Michigan State is out. Who takes their place?

Whether you like them or not, the College Football Playoff Rankings are the staple of debates in November and will give us our playoff participants at the end of the year. Each week we get to talk about who is in, who got snubbed, and who does and doesn’t control their playoff destiny.

The new rankings are revealed in a 30 minute show every Tuesday, so for tonights reveal we will attempt to predict what the committee will do with the rankings. This is.. easier said than done to say the least. Nobody seems to know for sure what the committee will use as their criteria, and it seems to change every week.

Leaving our prediction to one single number feels incomplete. We need to give ourselves some wiggle room by creating a prediction range. To do this, we are going to do something called quantile regression. When you implement linear regression, you’re predicting the mean, or average, of your target variable. Quantile regression simply pivots to the median, and allows you to creating a percentile range around the median prediction. To put it in simplest terms: We are going to create a floor and ceiling for each team for their college football playoff rankings.

NOTE* Remember, we are not trying to create power ratings or predict which teams are better. We are simply trying to predict what the committee will do tonight.

The Stats Behind The Model

So what exactly are we going to use for these predictions? We’re going to use things that you would expect the committee to use when they rank the teams. A team’s prior College Football Playoff ranking is far and away the most important variable in our model. Outside of a couple outliers, the committee really doesn’t overcorrect on their previous rankings. After prior ranking, a teams win percentage and the point differential of their last game come up next in terms of importance. Offensive and defensive performance (In the form of success rate and EPA/Play) really didn’t play a factor in the predictions.

Just as a quick reminder this was the previous top ten and the initial rankings this year. Michigan State unfortunately fell victim to the Purdue Spoiler-makers, while Wake Forest lost an offensive shootout to North Carolina. If you want to dive into the advanced stats of those games, you can read my weekly recap here.

Week 11 Predicted CFP Rankings

For the sake of a clear picture, here is just the predicted top 15 this week. It is virtually guaranteed that the top 2 of Georgia and Alabama will remain the same, even with Alabama struggling with a down LSU team. Oregon slides up to the number 3 space after Michigan State loses, but it is a close margin behind Ohio State. Cincinnati is still on the outside looking it, but the model believes their is a scenario where the Bearcats find themselves in the top 4.

Overall there really isn’t much shakeup to the top 15 outside of the losses. It will be interesting to see who ends up filling that third and fourth spot. We still have plenty of football left to be played, so shakeups will eventually happen.

Here is the entire top 25. As you can see at the bottom of the list, some teams that lost (Minnesota, Fresno State) are actually projected to leave the top 25. The end of the rankings are when we see the most variance, and the model can’t quite predict out of the pool of unranked teams. However, we can use the variables in the model to get an idea of where the committee might look in order to replace any teams leaving the top 25.

These teams seem to be the most likely based off of what they have done this season, and what they did this past weekend. As you can see, there is a little bit of a theme going on… all of these teams reside in the Group of 5. In the initial rankings last week, there were 4 Group of 5 teams ranked (Cinci, BYU, Fresno State, San Diego State). Couple that with undefeated Cincinnati being left out of the initial top 4, and you feel the group of 5 rage boiling. Fresno State will most likely be exiting from the rankings, so it would be fitting if the committee threw the G5 a bone and put another G5 in their place.

They have a very good cast of teams to pick from, including an undefeated UTSA Roadrunner team. Any one of these teams would be excellent additions to the Group of 5, and I am predicting that the additions to the rankings will come from this list.

There we have it! Predictions for the night. If you watch the coverage, hopefully the rankings don’t make you want to break things. Just as a reminder, we are trying to predict what a bunch of humans will do based on a criteria that seems to change with every week. I can’t imagine these predictions will be too far off, but the randomness is what makes college football special!

If you want to dive in to the data like I do, check out @CFB_Data and @cfbfastR on Twitter, where you can learn how to get started in the world of College Football data analysis!

If you want to see more charts and one off analysis, follow my twitter page, @CFBNumbers