How well can analytics predict the 2015 college football season? Can you make any conclusions before watching some games?
It seems like you shouldn't put much faith in preseason numbers. Randomness in football can instantly flip the emotions of the millions of Americans who connect their sanity to the actions of teenagers.
For example, consider the Auburn Tigers. In 2013, they had a magical season in which tipped hail mary passes fell in the hands of their receivers and missed field goals were returned for touchdowns. (Tigers fans didn't like it when I called this luck.)
Randomness swung the other way for Auburn in 2014. They went 2-2 in games decided by a touchdown or less (5-1 record in 2013). The tipped passes and special teams touchdowns evaporated.
Persistence of teams in college football
Despite this randomness, it is possible to make accurate preseason predictions in college football. Teams tend to persist from season to season. Rice will never have the tradition or financial resources of Auburn. Statistics like margin of victory in games tends to have a high correlation from season to season for FBS teams.
Because of this persistence of college football teams, quants can make accurate preseason rankings by including a team's performance over a number of years. Bill Connelly uses five years while I use four in my regression model, but you most likely get decent results using more than one or two seasons.
In my regression model, I also consider turnover margin the past four seasons and returning starters from last year. The model produces rankings based on a rating, or an expected margin of victory against an average FBS team. After adding 3 points for home field, these rankings make a prediction for each game.
Since 2005, this model has predicted 70.4% of game winners. Over the same time period, the higher ranked teams in the preseason rankings has won 60.1% of bowl games. For comparison, the team favored by the closing line in the betting markets has won 61.5% of games. These markets take advantage of insights from the regular season that the preseason rankings do not consider.
To check out The Power Rank's preseason college football rankings, click here.
Based on The Power Rank's preseason rankings, let's look at three overrated teams compared to the USA Today Coaches poll.
There's a huge gap between the Coaches (7th) and the analytics (25th) for Auburn.
I understand the thinking behind a 7th ranking for Auburn.
- Offensive guru Gus Malzahn will figure out the offense with new QB Jeremy Johnson.
- New defensive coordinator Will Muschamp, an excellent coordinator at Texas before he became the head coach at Florida, will ignite a defense that hasn't lived up to its recruiting potential.
Moreover, my preseason model considers a four year window. For Auburn, this includes the last two years of the Gene Chizik era, which do not reflect the current status of the program. The 25th in my rankings is definitely low.
However, there are many reasons to believe 7th is too high for Auburn.
Auburn lost QB Nick Marshall, an efficient rusher and passer, the top two rushers and receiver Sammie Coates on offense. Malzahn has proven his worth as offensive maestro, but even he couldn't get much out of Auburn's offense in 2011 after Cam Newton left.
There's also no certainty that Muschamp can make this defense elite. This unit ranked 41st in 2014 and 50th in 2013 by my yards per play adjusted for schedule calculations. At least Muschamp has many returning starters and graduate transfer CB Blake Countess from Michigan, who is excellent in zone but struggles with man coverage.
My preseason model predicts 7.0 wins for Auburn in 2015. This is low, and a better prediction is closer to the 8.5 wins set by the markets.
In 2013, the Seminoles put two very different units on the field.
On offense, QB Jameis Winston led one of the elite units in the nation. They ranked 5th in my rankings based on yards per play adjusted for strength of schedule. They would rank even higher without the debacle against Clemson when Winston didn't start.
On defense, Florida State struggled in 2014. They ranked 34th by yards per play adjusted for schedule strength, a terrible rank for a unit that sent 4 starters to the second and third round of the NFL draft.
Looking forward, the offense will regress without Winston. However, this unit shouldn't fall too much, as the coaching staff usually does a great job at QB development. First, they have to pick either Sean McGuire or graduate transfer Everett Golson, who led Notre Dame to the BCS title game in 2012, as the starter.
The defense should get better, as the unit ranked in the top 5 in both 2013 and 2014. However, Florida State needs serious improvement on this unit to be 8th best team in the nation as the Coaches Poll predicts.
The Power Rank's estimate of 17th seems more reasonable for Florida State (this is the same rank Bill Connelly gets with his model). My model expects 9.3 wins, very close to the 9.5 wins set by the markets.
Last season seemed like a disaster for USC in Steve Sarkisian's first season as head coach.
They lost early at Boston College, as the Trojans couldn't stop the running game despite the presence of DE Leonard Williams, the third pick in the 2015 NFL draft. They also lost by 18 to rival UCLA and had an uninspiring 9-4 record.
Despite these poor performances, USC ended the season 19th in my team rankings that take margin of victory and adjust for strength of schedule. It was similar to their 20th ranking in 2013 and 19th in 2012.
Since my model considers a four year window, it's not surprising that USC starts 2015 at 19th.
2015 is a key year for Sarkisian. If he has another 9-4 year with the elite talent on the USC roster, then he might last a few more years before USC moves on.
On the other hand, if USC wins the Pac-12 South and looks like the 10th team as predicted by the Coaches Poll, then Sarkisian might remind Trojan fans more of Pete Carroll than Lane Kiffin. With QB Cody Kessler and wealth of five star recruits on the roster, this jump is certainly possible.
My model has USC 19th and predicts 7.6 wins, quite a bit less than the 9 set by the markets. Still, it's hard to find 4 losses on USC's schedule. The highest likelihood is at Oregon (19% win probability). After that, they have Notre Dame (35%), Arizona State (41%), UCLA (49%) and Stanford (54%) as toss up games.
If USC splits the toss up games, then they need to lose one more game to go under 9 wins. If USC plays above their 19th preseason ranking, this seems improbable.
Check out the win totals report
If you're interested in the win totals for other teams, check out The 2015 College Football Win Totals Report. In addition to win totals for most FBS teams, I also discuss two teams with value in the markets as well as two other overrated teams for 2015.
To get this free report, click here.
Ed Feng is a data scientist and writer, mostly on college football. His content appears on his site The Power Rank as well as outlets such as Grantland, Deadspin and Bleacher Report.