Let’s do a midseason check-in on my preseason prediction model
A lot of football analysts have made predictions about the career arc of Buffalo Bills quarterback Josh Allen. Eye-test analysts have been a mixed bag of prognostications. Data-based predictions have NOT been a mixed bag, with seemingly the entire field of mathematics not liking Josh Allen’s chances. And until this year math was the undisputed champion.
After Allen’s rookie year I was too impatient to see how his second year would go so I took a look at similar QBs in the decade prior to Allen to see how they changed from year one to two. After year two played out in reality, I looked like football Nostradamus. That’s nearly the end of my tooting my own horn. I did it again to predict year three and now that we’re ten games in let’s take a look to see how Allen is doing. Note: A more robust look at methodology can be found in the earlier articles.
Here’s a quick refresher on reading these charts. 2019 and 2020 Josh Allen numbers should be self-explanatory. The three bars marked “average” represent the range of change that was normal for the sample group and applied to Allen’s 2019 stats. The best- and worst-case bars take the top and bottom performing player for each stat and apply the amount of those changes to Allen’s 2019 base as well.
This is the last time I get to toot my horn. Josh Allen ended 2019 with a 2.0 percent interception rate. On average, quarterbacks got worse in this metric in their third year, although any decrease up to and including 1.2 percent would have been considered normal. For Allen that means he’d have had to fall under that 0.8 percent bar to make the model “wrong.” He’s remained pretty flat so far. To be fair, significant improvement was unlikely as this was a stat that 2019 Josh Allen looked really good in already.
Yards per game
Perfect average improvement would have been a “whopping” five yards more per game. Josh Allen improved by 99 yards per game. That’s higher than five yards. It’s also higher than the top of the range that’d be considered normal by 45 yards. Allen didn’t improve as much as Matthew Stafford did but, being objective, the model was designed to predict a “normal” change and Allen’s change was abnormal.
Yards per attempt
Yards per game can be directly impacted by play calling. Increase passing attempts and yardage volume will go up too. So I prefer yards per attempt. Josh Allen is again beat out by Matthew Stafford but also again he’s above the normal predicted range. That’s also not a small amount over that range.
For touchdown percentage, change the name of the best-case scenario to Andrew Luck but otherwise the narrative remains the same. Josh Allen broke through the normal range of change by a decent margin.
And here’s the only one aside from interception rate where the narrative is different. Here, though, it’s a good different. Of the quarterbacks drafted in the decade before Josh Allen, none of them improved their completion percentage as much as Allen has going from year two to three. For anyone doing this same project for another quarterback Allen is now your best-case player for this stat.
Unlike most analysts, I don’t mind being wrong. First, when I said that fans should be rooting for Allen to have a greater-than-normal improvement this year I was also including myself in that statement. As a fan it’s way better to be wrong on this one.
Second, I included the best- and worst-case scenarios for a reason. Most data-based analysis recognizes the existence of outliers and I wanted to give you the full range so we would know it if we saw it. Make no mistake, even though he’s not the new best case for every stat, Allen is looking like an outlier.
It’s not a shock to see Stafford as the best case in two categories because the two are closely related. Stafford was nowhere near the best case for the other stats. His 4.1 percent improvement in completion percentage was less than halfway there for that stat and he regressed in TD percentage and INT percentage. Who cares about Stafford right? You should. Only one quarterback, Ryan Tannehill, improved in every one of the five measures above. And Tannehill’s improvements were thoroughly average. Putting those facts together means there wouldn’t have been any evidence to suggest a quarterback would make massive progress on four out of five categories.
Now of course there’s six more games to play and Allen could fall off but it’d take a steep cliff to come back to the averages the model predicted.