By Kurt Bullard
It’s hard to believe we’re only two years removed from what might have been the greatest single-season performance from a quarterback of all time: 5,477 yards, 55 touchdowns. And yet, the Broncos enter the postseason as perhaps one of the weakest 1-seeds in recent memory, mostly because of the question mark that the QB position has become for the team over the season. Manning only threw for 2,249 yards in 10 games, tossed 17 interceptions against only 9 TDs, and most dramatically brought fans to question whether Brock Osweiler was indeed a better person for the job right now.
The biggest criticism of Manning is that his arm strength has significantly declined in recent years, and that the future Hall-of-Famer can’t make the throws he used to. I was curious to what extent this was true, so I decided to run a logistic regression to see how much the distance of a throw affects whether or not Peyton completed it.
Using TruMedia data from 2006-2015, I was able to see the air yards for each Peyton Manning throw. I was more concerned with air yards than actual yards gained, because it showed how difficult of a throw it was—that is to say, a screen that goes ten yards is different than a ten-yard out route. I then created a dummy variable for throws that happened in 2015.
I ran a logistic regression on the air yards, the aforementioned dummy variable, and the interaction term, the latter of which measures how the effect of air yards is different in 2015 than the rest of his career.
I then ran a backward stepwise regression, which did not remove any of the variables. The regression produced the following model.
The coefficients refer to the effect of the log odds of a completion—while an extra air yard previously only dropped the log odds of a completion by 0.062, an extra air yard this year dropped it by 0.09. And, although the interaction term is not significant according to the regression, the backward stepwise regression kept the term in (meaning it produced a lower AIC than had it been left out), suggesting that the variable added predictive power to the model.
The following graph shows the translation of the log odds into actual percentage odds.
From the graph, it’s clear that Peyton could use a little extra protein from some chicken parm this year. Past Peyton completed 20-yard passes an average of 44% of the time. Present Peyton only does it an average of 34% of the time. That’s a noticeable difference from one of the all-time greats. The model also gives him about a 20% chance of completing a 30-yard pass in 2015, which are the same odds of him completing a 40-yard pass in the past.
Additionally, Peyton has been less aggressive than he was in the past. The following table shows the average air yards on Peyton’s pass attempts from each year starting in 2006, with 2015 obviously being the most passive Peyton has been in attacking downfield.
So when Peyton takes the field in Denver in a few weeks, don’t expect the Peyton of old. Expect the new, beleaguered one.