(Editor's note: Yup, another promoted fanpost!)
This article sprouted from an odd seed: I noticed that Russell Wilson pre-2021 and Geno this year had the same massive overperformance at completion percentage above expectation (CPOE) compared to their EPA/dropback (if you don't believe me, look here and go to the QB tab). Judging Geno's 2022 performance seems to me imperative to the ongoing discussion of extending him or letting him walk. To that end, I selected this list of QBs for comparison:
- Patrick Mahomes in 2022 (for high range context)
- Davis Mills in 2022 (for low range context)
- Justin Herbert in 2022 (for mid range context)
- Geno Smith in 2014 (the previous season in which he got significant snaps)
- Russell Wilson in 2020 (the last "regular" season from him)
- Tavaris Jackson in 2011 (the only other pre-Russ, during-Pete QB)
Those last three plus Geno in 2022 I have anonymized as A, B, C, and D (not necessarily in that order). After all the charts, I'll... nonimize them.
For those who really care about my calculations, I'll explain them in depth at the bottom of the article. The summary is that I have interpreted all metrics to have the same scale for better comparisons. I have also flipped two of the charts upside down (hits and sacks) so they have the same higher numbers-are-better meaning as the others. Zero is league average and is reinforced with a thicker line. Anyway, on with the charts!
Our anonymous B takes a lot of hits. So that seems like a strong candidate to be Russ. Mahomes takes the (expected) lead here, showing that he's pretty much good at everything.
Interestingly, all our masked QBs fare much, much better at taking sacks than taking hits. This time C is our clear worst player with Mahomes again defining what awesome looks like.
This is where things get weird. Pete's QBs seem to be really great at this metric. There does indeed seem to be something to his offensive system(s) that favors CPOE. Don't get me wrong: it's better to have it than not. But it does not seem to have a strong correlation to quality. For instance, the 2022 leader at this number is Jacoby Brissett at 2.45 and the worst is Kenny Pickett at -2.1. I don't see anyone falling over themselves lauding the former nor flaming the latter so this metric seems to need extra context and interpretation to be useful.
Okay now we're getting to a more comprehensive gauge. There are no perfect measuring sticks but this is a pretty good one overall. Mills forms the caboose and Mahomes forms the engine for this train.
EPA is one of my favorite metrics and it looks roughly like a slightly exaggerated version of the NY/A graph above. Mills is bad and Mahomes is just real darn good.
Alright, enough farkeling. Time to unmask:
- blue = A = 2020 Russ
- yellow = B = 2011 Tavaris
- green = C = 2022 Geno
- red = D = 2014 Geno
Here are all the metrics in one dense chart with the QBs disunreanonymized:
Hopefully you at least guessed that D was G-G-G-Geno and the Jets based his terrible CPOE. Tavaris really took the hits at an alarming rate way back when, and really all Pete's QBs seem to be mediocre about taking sacks.
Based on this, that Geno really is a different QB than before. His performance this season significantly outscores his past self at the more important metrics of NY/A and EPA/dropback. He's still kinda prone to hits and sacks but those are tolerable when the primary signs are positive. His CPOE makes a startling improvement here, for whatever that's worth.
The one clear number Geno lagged Russ was at EPA/dropback, and that's not a good one to trail. I favor EPA over NY/A so there's wiggle room for whatever conclusion you'd like.
So what did we get from Geno in 2022? Based on the above we got a league average-ish starting QB which was amazing value for his contract. Is that worth big money? Or several years? Or both? I can see him possibly reaching neutral value on a big contract and being a lower risk than a draftee or other replacement. Can the Hawks afford to let him go? That's a lot to risk and having won the lottery this year does not change that it is still remarkably unlikely to win next year.
And now for the statistical details on my process, the part no one's been waiting for!
I use the play-by-play data from nflfastR and per season I remove non dropbacks, 4th downs, two minute drills, and win probabilities above 90%. I group the dropbacks by defense to determine their averages. Next I find all the starting QBs by removing ones with less than 200 dropbacks. Then I group the dropbacks by starting QB, determining each week's stats. I offset each of those numbers by the opposing defense's seasonal averages (e.g., Geno facing SF's defense gets a numerical boost because SF has a great defense). Finally, I combine all the opponent-adjusted weekly stats into seasonal averages per starting QB.
At this point I have good data but it's difficult to compare between metrics. So I convert all metrics from their native units into the number of standard deviations from the mean. For the stddev and mean I only use starting QBs from the same season to makes sure I account for NFL offensive inflation and that I ignore non-starting QBs. Anyway, this converts them all to the same scale and makes their numbers really useful. So a score of 1 means 68%, 2 means 95%, 3 means 99.7%, and the negative scores have similar but opposite meanings.
Oh, also, I used the "qb_epa" column in the play by play stats, not the "epa" one. I understand that the former removes negative effects from receivers' fumbles as that does not reflect on the passer. It's sort of like baseball's ERA being a superior gauge of pitchers than RA would be.
Let me know in the comments if you have any questions or requests for other articles. Thanks!