Monday, September 11, 2006

NFL turnovers are inconsistent from year to year

Are turnovers in the NFL largely random?

Apparently so. The
Wages of Wins blog did a year-to-year regression, and came up with a correlation coefficient of about .15. That means only 15% of a team’s turnover tendencies carry over to the next season. (They do report the r this time, not just the r-squared.)

The also report that teams with more than 10 net turnovers (turnovers forced minus turnovers made) finished with 11 wins. Teams worse than –10 net turnvers won only 6 games, on average.

Which kind of makes sense. According to “
The Hidden Game of Football” (pp. 143-144), a turnover is worth four points. And it takes 34 points to turn a loss into a win. So it takes roughly eight turnovers to make up a win.

The study doesn't give an actual number of net turnovers for the two groups, but it's got to be over 10. Let's assume it's 16. That would be two wins. That works out to 10-6, or 6-10, roughly what the study found.

Which is again evidence of turnovers being largely random. If they weren’t, you might expect turnovers to be an indication of a team’s quality in other ways. If that were true, turnovers would correlate higher with quality, and teams with lots of turnovers would tend to be even worse than the expected 6-10.


At Wednesday, September 13, 2006 3:56:00 PM, Blogger Tangotiger said...

If I understand you right, you are showing that turnovers as a "skill" is uncorrelated to the rest of a team's skill-level. But, you have not shown that the extent of the turnover skill itself.

If I may switch to baseball: if a batter's walk rate was uncorrelated to his SLG, you'd have the same thing. His walk rate is uncorrelated to the rest of a player's skillset, but something quite real. (Just an illustration. A player's walk rate would be correlated to his SLG.)

At Wednesday, September 13, 2006 6:25:00 PM, Blogger Phil Birnbaum said...

Hi, Tango,

Yes, I'm assuming that turnovers can't be completely independent of other skills, unlike walks and SLG.

For instance, if interception avoidance is a skill, it might be ability to get to where the ball is, or ability to throw the ball accurately, or some such ... all of which would show themselves in other areas (such as completion percentage).

It's *possible* that they wouldn't, but intuitively it's unlikely.

For fumbles, though, I'd be more inclined to agree with you ...



Post a Comment

<< Home