Some metrics are like getting a football stuck in a tree.
But sometimes they get it wrong.
This year, they accepted the talk “Identifying an accurate metric for football efficiency”. Tim Chou, a football coach, discovered yards per point as the new way to evaluate football teams.
There are many problems with this premise.
First, yards per point is not a new concept. Phil Steele, who publishes a yearly preseason magazine, has been writing about it for years.
Second, yards per point is not a good metric for football. Randomness plays a big role in its value, mostly due to turnovers. Phil Steele uses it as a metric to determine teams that got lucky the previous season.
Over at Bill Connelly’s Football Study Hall, I discuss the problems with yards per point as a football efficiency metric. Kansas State’s offense from last season is a perfect example of these problems. To read the article, click here.
After seeing this article, Jeff Fogle at Stat Intelligence wrote a post in which he remembers questioning yards per point back in 1986. This is neither a new or particularly useful concept.
Interesting.
I think YPP is at least a good predictor if you’re looking at “one” stat to predict a winning percentage. I think a YPP differential is better, when you take YPP for offense and a YPP for defense and subtract the two, as I’ve done here: http://pac12statcentral.com/testsite/temp/scatter/public_html/ (Plot y as winning & and x as YPP differential). It provides a 78% R^2
Another good “single” predictor is Pass efficiency differential margin, where you take a differential of a team’s offensive and defensive pass efficiencies. This obviously has drawbacks (especially for teams that are pass heavy or don’t pass at all). Anyway, the link I shared will let you try out some different combinations of stats to see what correlates with what. It is one of the new interactive features I will have on my site this fall. knowhuddle.com
Jim,
Good stuff, thanks for sharing. Please let us know when your site is up.
Yards per point is a good predictor… of the past. There’s no doubt that it correlates strongly with winning percentage. The problem is that I’m looking for metrics that predict the future.
Thanks to both of you for providing tools such as this for those of us who lack the mathematical acumen to create them ourselves.
I went to the link and played with it for a while and had two thoughts, one for each of you.
Jim – is there a reason the drop downs are ordered the way they are? It would make it easier to use if they were alphabetized.
Ed – do you think this would help, say, halfway through the season or do you think the sample size would be too small? Also, another thought, the non-conference games don’t give an accurate picture compared to in-conference games for the BCS conferences. I’m glad this is the last year I’ll have to use BCS to describe anything!
Craig,
Sample size is always an issue in college football. I think the tool is useful in finding correlations in past seasons. Then we kind of assume that they will continue to hold this season.
Hi Guys. I’ve been doing a lot of research with regard to yards per point and trying to derive a predicted points scored in a game from this statistic. So far the randomness factor of turnovers, etc… tends to leave the results a bit further away from where I want them to be. I have a complete statistical analysis for every college football team (FBS) and I’m able to accurately predict total yards for each team, within a margin of error of about +/- 25 yards a game for each team. Does anyone have any thoughts on a better metric than yards per point that might be more useful in predicting scores? I’m always looking for new and efficient ways of evaluating games.
Tom, thanks for stopping by. I would try yards per play to attempt to predict points scored. This is on my to do list. Let’s compare notes when we both do it.