Numbers, especially those spit out of a computer, must pass the stupid test. If you think the results are stupid, then the numbers don’t pass the test.
After four weeks of college football, I was working on rankings that only included games from this year. Here at The Power Rank, we take football statistics like yards per play and adjust them for strength of schedule. But when I looked at the results for defense, Texas Tech had the top ranked defense.
Texas Tech?
The same defense that gave up 39.25 points a game last year, 117th in the nation?
Last year’s Texas Tech defense actually looked better by yards per play, ending the season ranked 75th in our defense rankings. (In the rest of this article, offense and defense rankings refer our calculations that adjust yards per play for strength of schedule.) But head coach Tommy Tuberville just brought in his 3rd defensive coordinator in 3 years. Even with a load of talent, a jump from 75th to 1st is almost impossible.
It doesn’t pass the stupid test.
How to Make The Rankings Better
We use past data to calculate rankings that predict the future. This depends on using variables that correlate from the past to the future. For example, yards per play allowed by a defense in the first 6 games of the season has a correlation coefficient of 0.481 with yards per play the rest of the season. In contrast, fumble recovery rates for a defense have almost no correlation between the first 6 games and the rest of the season. Hence, we look at yards per play but not fumble recovery rates.
But what about last season? Surprisingly, yards allowed per play correlates season to season with a coefficient of 0.496, even larger than the early to late season correlation. We also looked at other statistics, such as points per game and pass yards per attempt, both on offense in defense. In general, the season to season correlations were slightly smaller than the early to late season correlations.
But what’s going on behind these correlations?
During the season, yards per play is affected by injuries and the inherent randomness of football, such as that tipped pass that falls into a receiver’s hands for a long touchdown. Changing the yards gained on a play from 0 to 92 will have a dramatic effect on the game average. These factors make the early to late season correlation coefficient much less than 1.
From season to season, teams can have large turnover in players and coaches. I thought this would lead to a much smaller correlation coefficient. It doesn’t. One reason is the identity of the program. Texas has a huge fan base, lots of money, and a long tradition of recruiting stellar athletes from the Lone Star state. Rice has none of those things.
No perfect rankings early in the season
The strong season to season correlations suggests using last year’s results in this year’s rankings. In fact, we have been doing that so far, using all games last year and counting this year’s games twice. We will continue to do this.
But there are still problems.
For example, Baylor is still the top ranked offense. Given that the Bears lost quarterback Robert Griffin III to the NFL draft, this ranking is most likely too high. However, head coach Art Briles has led successful offenses at both Houston and Baylor. Moreover, senior quarterback Nick Florence has already led the Bears to 7.26 yards per play this year, much higher than the 5.45 average over all FBS and FCS games last season. With only this year’s games, Baylor is the 34th ranked offense due to weak competition. They will probably end the season somewhere in between 1st and 34th.
Last Season Does Matter in College Football
Surprisingly, key football statistics like yards per play correlates from year to year almost as strongly as it does from early to late season. While that certainly affects what games we use in our rankings, it also matters to fans. Reputation, tradition and money matter in college football. These factors don’t change overnight. Consider these 3 programs.
- Michigan. The Wolverines had 3 horrible years under coach Rich Rodriguez. But in the 23 years before that, Michigan ended each season in the top 25 of our rankings. In the first year of this streak, a guy named Jim Harbaugh started his first game at quarterback. So it shouldn’t be too surprising that Michigan got back to 12th last year under new head coach Brady Hoke.
- Auburn. The Tigers have almost 2 million fans and pack over 85,000 of them in Jordan-Hare Stadium for home games. Despite the ups and downs of this past decade, Auburn has never finished in the bottom of half of Bowl subdivision teams in our rankings. They won’t fall that low season either, although that won’t make Auburn fans happy with head coach Gene Chizik.
- Rice. This small private school in Texas, which happens to be my alma mater, just isn’t a very good football program. The Owls have never finished in the top half of teams this decade, not even during their 10-3 season in 2008. Don’t expect much from this program.
Programs don’t change overnight. Keep that in mind as the season progresses.
Hello Ed,
I am a new visitor to your website. Your website is very interesting, and I can’t stop reading.
About the weekly college football predictions…where on your website can I find the predictions for the past seven weeks?
Again, fascinating read. Thank you for your time.
John Le
John,
Thanks for the kind words. I’m a little surprised how many requests I’ve gotten for past predictions. I’ll get them up this week.
Ed
Ed,
Have your rankings for college football/basketball and NBA already factored in the home field/court advantage for the home team? That is, after taking the difference between the ranking numbers of the two teams, do we still need to add 3.0 points to the home team?
And is it correct to assume that your weekly college football predictions have already assigned 3.0 points to the home team? Will you also have predictions for the 2012-2013 college basketball and NBA season? Thanks.
John
Hi John,
The predictions on the Predictions page have a home field advantage included already. If you take the difference between the ratings of 2 teams, that’s a neutral site prediction. You would need to add some factor for the home team.
We love basketball. When the rankings have enough games to chew on, we will have predictions.
Ed