Preseason college football ratings have historically been far less scientific than even the polls. But that doesn’t have to be the case. This spring, ESPN Stats & Information created an automated set of preseason ratings for all FBS teams heading into the 2014 season, designed to predict how strong each team (including its offense, defense and special teams) will be this coming season. This is essentially a preseason version of the Football Power Index (FPI) used last year.
Before any games are even played, historical data on expected points added (EPA) tells us a lot about how teams perform the next year. For example, Florida State’s past few years of performance, number of returnees (players and head coach) and recruiting suggest that its offense in the 2014 season is going to be a stellar 16.9 PPG better than average. Similarly, its defense will also likely be outstanding at +13.6 PPG, with a solid +1.7 PPG on special teams. With an overall preseason FPI of +32.2 -- representing how much better the Seminoles would be against an average FBS opponent (someone like Illinois) on a neutral field -- they project to be the best team in 2014, just as most subjective opinions have them.
The results of the method are interesting, but so are some of the things we learned when building the method. For instance, these are some characteristics we found to be important in making predictions:
• The previous year’s EPA generally matters more than EPA from the prior years, though sometimes the most recent year matters less if the team’s head coach or several starters on that side of the ball are not returning.
• That said, performance from years prior to the most recent one still does have a significant effect. A college football team plays only a dozen or so games in a season, so having additional data from prior years, even if it matters less than information from the most recent year, helps make the ratings more accurate.
• Having more returning starters helps on both sides of the ball, sometimes even if the team wasn’t that good the previous year. A returning starting quarterback helps the most, accounting for a boost of more than three points of offensive EPA per game versus having a new starter (all else equal).
• The presence of a new head coach generally decreases correlation between the previous season and the coming one (relative to returning the same head coach), which intuitively makes sense. A new head coach hurts a team’s projection if it was good in one of the components (offense, defense, special teams) the prior season, but it can help the projection if it was poor in those aspects.
• Recruiting rankings, not just from the most recent class but over the previous couple, are definitely helpful in predicting offensive and defensive performance. There isn’t a big difference at all between having the 12th-ranked and 15th-ranked recruiting classes, but the difference between the 12th-ranked one and the 80th-ranked one can be substantial. We used a survey of various recruiting ranks to avoid being reliant on any single rank. Special thanks to Phil Steele for providing us information on returning starters and head coaches for 2014 and historical seasons.
From a big picture perspective, the method behind these rankings is fairly similar to how a knowledgeable writer might go about constructing his or her own top 25: look at which teams were successful last season or have been good over the past few years, account for what each team has coming back, and assess teams’ recruiting classes over the past couple of seasons.
All we did was enhance the details of that big picture using the systematic approach of ESPN Analytics. See the 2014 Preseason Football Power Index ratings here.