Bill Barnwell is the managing editor of Football Outsiders and perhaps its most regular contributor. I agreed to asking him questions about the Seattle Seahawks in exchange for a free .pdf copy of Football Outsiders Almanac 2009.
Field Gulls: Football Outsiders projects Seattle to rebound this season and win 9.9 games. That projection relies on Seattle’s injury rate regressing towards the mean. How did you determine the replacement value of a player and how much Seattle’s starters were worth above replacement?
Bill Barnwell: Well, that's one of the things it takes into account, although it's certainly not the only one. You can't truly calculate individual replacement value for players like offensive linemen or non-"skill position" players. We simply separate players into simple sections of starters and reserves because it's certainly preferable to not doing so.
FG: Did you group Seattle’s replacements into a "starter" box and therefore value the team losing Sean Locklear the same as it losing Walter Jones? How do you compare the value of Logan Payne and Nate Burleson?
BB: Yes, we consider losing any starter to be as valuable as losing any other starter. Of course, that's not ideal. We're working on ways to quantify individual player value in an attempt to improve the reliability of the data, but we haven't found anything we're happy with yet. There is a very tangible difference in data reliability, though, when we do separate "starters" and "reserves".
As for Payne, we designate players like him who inherit starting jobs because of injury as "NEW STARTERS" and assign them a partial value for their injuries.
FG: Vince Verhei states "The ten most injured offenses from 2004 to 2007 averaged 6.2 wins and a -3.6% offensive DVOA; the next year, those numbers jumped to 7.2 wins and 0.9% DVOA." What was the average record the following season of all teams that finished with 6 wins? Could this be explained by regression towards the mean?
What was the offensive DVOA the following season of all teams that finished a season at 0 to -5% DVOA? Could this be explained by regression towards the mean?
BB: You're conflating cause and effect here with these two questions, John. Teams regress to the mean because their injury rates improve, not vice versa. Teams don't regress to "the mean" because it's a nice mathematical concept; there are tangible reasons, things like fumble recovery rates and uncontrollable special teams factors and, in a big way, injuries.
. . .
That last answer seems a little evasive to me. I know I didn't conflate cause and effect and I'm pretty sure I didn't confuse cause for effect and effect for cause, if that's what he meant to say. What I was asking is simply how much of the team's in question rebound in performance is because of improved health and how much is it because of the remaining factors that cause a regression towards the mean? I think that could be reasonably tested by seeing how much all teams that win six games improve the next season and how much all teams that score within the given DVOA range improve the next season. If injuries were a major factor, we would expect a significantly better improvement from the teams impacted by abnormally high injury rates.
I sent him the above response, and if and when he responds I will update it.
I usually run a review of Football Outsiders annual book, but haven't had time this year. I'm sure many of Field Gulls readers have already bought a copy, and if you're interested in exclusive stats and essays from some of the most prominent football analysts, buy a copy.
Update: After a little back and forth, we have an answer.
BB: The r2 of our correlation would suggest that something like 20-22% of a change in a team's year-to-year performance is due to the effects of injuries. That assumes that our metric is a perfect indicator of injuries, which it obviously isn't, but I don't doubt that our metric could get even better. So I'd estimate that's a reasonable guess.
. . .
R2 as defined by Wikipedia means:
In statistics, the coefficient of determination, R2 is used in the context of statistical models whose main purpose is the prediction of future outcomes on the basis of other related information. It is the proportion of variability in a data set that is accounted for by the statistical model. It provides a measure of how well future outcomes are likely to be predicted by the model.
So, assuming the metric is a perfect indicator of injuries, it can account for about a fifth of a team's future performance. Which is significant, but also underscores the fact that Seattle needs more than health to rebound. Football Outsiders believes the Seahawks have what they need to be a playoff contender and even a Super Bowl contender.