By Kurt Bullard
It’s been a week or so since the famous decision by Jack Del Rio to go for two against the Saints on the road, trailing by one point with only 47 seconds left in the contest. Derek Carr would proceed to drop the ball right into Michael Crabtree’s hands on a fade route to the back-left corner of the endzone. After the defense was able to stave off any Drew Brees-led comeback, Oakland found itself with a victory in its season opener.
It wasn’t without controversy, however. According to ESPN’s win probability model, the Raiders’ chances of victory would have been maximized with an extra point instead of a two-point conversion attempt. As we know, Del Rio chimed in on the discussion, reminding everyone that he made the “right call.”
The one downside to the ESPN win probability model in scenarios like this one is that it doesn’t take into account unit vs. unit matchups. While it incorporates overall team strength, it can’t take into account the specific matchup of Raiders offense and the horrid New Orleans defense.
After deciding to go for two, the Raiders’ ability to convert in this specific matchup would very much swing the outcome of the game. So, I thought it would be interesting to see whether or not there were any way to get a more accurate model for two-point conversion rates given a specific situation.
To do so, I looked at all explicit attempts for two-point conversions—that is, no fake-kick attempts—over the past three regular seasons. By my count, there were 221 attempts in that span. I then ran a logistic regression for conversion successes against play type (pass or run), the offensive team’s offensive DVOA from the end-of-season, and the defensive team’s defensive DVOA from the end-of-season.
That regression produced the following model:
No predictor by itself is significant, and the residual deviance is not largely reduced by the three predictive variables. However, it is interesting that the pass coefficient is pretty negative. Over the past three years, 55% of run plays have converted, while only 46% of pass plays have ended up in two points. However, only 20% of conversion attempts have been running plays, with teams predominantly opting for the air attack in these situations. The sample size isn’t large enough to deem the effect significant, though.
Using last year’s DVOA metrics to measure the offensive and defensive strength of the Raiders and Saints, respectively, the model predicts that a pass attempt in this situation would have converted about 51% of the time. So, a conversion in this situation on a Carr throw was more or less a toss-up.
This model is pretty simple, and therefore does not offer deep insight into the world of two-point conversions. But it does confirm that converting in these situations is very much a random outcome. In this case, Del Rio and the rest of the Raiders just happened to come out fortuitously.