14 Sep 2012
As is frequently the case with Google, I was in the middle of researching something entirely unrelated, but randomly came across this working paper by a couple of sports management students at the University of San Francisco. They looked at 92 games over the past three NFL seasons wherein a team was either tied or down by one to three points on their final drive, and succeeded at reaching field goal range (i.e., the opponents' 35-yard line). They then built a regression model that predicted how much time (in seconds) it took to get there.
Because of the limitations of the methods they used (which they fully acknowledge in the paper), the results are more explanatory than they are predictive. However, there hasn't been much work done on this kind of question, so what they found is useful as a first step nevertheless. For instance, here are the descriptive stats for the 92 drives:
On average, these teams started the drive at their own 30-yard line with 2:49 left, and down by a score of about 21-19 with two timeouts remaining. Also, teams tended to benefit from the two-minute warning and one other called timeout (offense and defense combined). In terms of player characteristics, successful drives featured a league-average quarterback and one offensive all-pro. Most importantly, it only took these teams 92 teams an average of 80 seconds to reach field goal range. None of this is earth-shattering, but it's good information to know.
The more interesting stuff come from the regression analysis. Results are below:
The analysis says that, on a team's final drive down three or fewer, the amount of time it takes them to reach field goal range is a function of when the drive starts, where the drive starts, how many timeouts the offense has at their disposal, how many timeouts they actually use, and how good their quarterback is. This reeks of "duh" until you look at the regression coefficients, which show some counterintuitive predictions.
I'll start with the least impressive: passer rating. Obviously, we'd love to see them use a better measure of true quarterback skill like DVOA or DYAR. Putting that aside for a moment, though, the regression coefficient is surprisingly positive (i.e., the better your quarterback, the longer it takes to reach field goal range), but essentially meaningless. That's because, according to the model, having a quarterback that's 40 points of passer rating better -- akin to the difference between Tom Brady and Blaine Gabbert last year -- costs a whopping 10 seconds of drive time. Sorry, with 2:49 left, I'll take Brady over Gabbert every day and 100 times on Sunday even if it means my team reaches field goal range with 1:19 left instead of 1:29. Heck, I might even prefer leaving only 1:19 on the clock considering that it means my opponent now has slightly less time than the average it takes to successfully get in position for a field goal of their own (i.e., 80 seconds).
That strategic nuance leads me into the more interesting counterintuitive predictions, which seem to be somewhat conceptually related (although the VIFs say they're not statistically multicollinear). Namely, the earlier a drive starts and the fewer timeouts an offense actually uses, the more quickly they reach field goal range. So, all else equal, a successful field goal drive starting with 2:59 left that uses one timeout reaches field goal range at the 1:06 mark on average, whereas one starting with 2:49 left that uses zero timeouts reaches field goal range at the 1:36 mark (i.e., 30 seconds earlier). What in the name of Norv Turner and Andy Reid is going on here? Why would starting earlier and using more timeouts be associated with longer field goal drives?
I have my suspicions, some of which I've hinted at in the last two paragraphs, but what do you all think? The paper is very readable for an academic write-up, so give it a look and offer your thoughts in the comments.
12 comments, Last at 17 Sep 2012, 2:09pm by Dan in Philly
Thanks a lot, Dak Prescott. Now more people will think the fourth round is still a gold mine for quarterbacks, but the data says otherwise. The update to our quarterback draft study for 1994-2016 shows little has changed: finding a good QB is really hard.