Week 4 Bayesian Quarterback Rankings

Joe Burrow and Brock Purdy projections rising, despite #QBLosses

The big, fundamental change to the rankings this season is the integration of Adjusted Quarterback Efficiency (AQE) numbers. This produces rankings that align more closely to what the typical football observer or data-based analysts would assign based on a combination of observation and statistics.

For the Week 1 projection, I weaved the AQE figures for 2023 and 2022 into the mix. In Week 2, I discovered the addition of prior years’ charting from FTNData, enabling us to go back to 2019 and calculate AQE. Because we’re shifting the historical data for several years in the new projections, the projection movement from Week 1 wouldn’t be primarily based on last week’s quarterback performances, but mostly on revisions to 2019-2021 efficiencies.

You can find all the previous weekly editions of the Bayesian Quarterback Rankings here, and the backlog for Adjusted Quarterback Efficiency is here.

COMPARING GRADES AND EFFICIENCY

PFF grades aren’t part of the analysis, but I find it helpful to make not of how they align with EPA per play, as many contextual elements of quarterback play (drops, interception-worthy throws, easier throws that become big gains, etc) are part of the grading methodology, but aren’t accounted for in EPA. At the same time, I think EPA does a vastly superior job of weighing what is and isn’t important in points-based results.

The plot below is a bit different than previous iterations of this post, substituting AQE for unadjusted EPA per play, and you might notice that the data has less dispersion (i.e. something like a higher r²) than using straight EPA. Even so, AQE doesn’t perfectly align with PFF grading, and you can decide which measure is more representative of fundamental quarterback play. (hint: it’s AQE!)

I took Bryce Young and Skylar Thompson out of the plot, even though they met the 40-play threshold. The former is on the bench, and the latter might not start, plus he was so bad as to mess up the scaling of the whole visualization. Maybe I should remove Deshaun Watson for pulling down the EPA scale, but I think he’s locked into several more starts considering his contract.

There are quite a few quarterbacks AQE prefers to PFF grading. Some are because they’ve been so good by any metric that grading isn’t able to catch up (Josh Allen and Malik Willis), some because of larger positive adjustments in AQE (Brock Purdy and Joe Burrow). Allen and Purdy are consistent production-over-grades guys, which is reflected in their strong Bayesian rankings below.

On the flip side, Derek Carr and Geno Smith have been good producers who grade a lot better than their strong numbers. They generally have been grades-over-production, mostly because of grading’s bias (imho) towards accuracy, especially downfield, over doing other things that drive efficiency, i.e. points. Smith normally suffers from bigger negatives, like interceptions and sacks. That’s true, again, this season, so my ranking on Smith is going to trail a lot of media ones, who perceive his play more similarly to grades, loving the big-time throws and discounting the mistakes as a function of poor protection. Smith definitely has had poor blocking (Seahawks are 29th in PFF pass blocking), but his propensity to make bigger mistakes has been present for the last few years.

Justin Fields and C.J. Stroud are two more quarterbacks who don’t look as good by AQE as watchers’ perceptions, which will explain why they haven’t had the best Bayesian rankings movements so far this year. This disconnect was present for Stroud as a rookie also, but his AQE was still excellent for a first-year quarterback, just not as relatively good as his grading.

PROJECTED ADJUSTED EFFICIENCY

These results are the ranking for the go-forward projections of adjusted quarterback efficiency starting this week. I also included the AQE rankings for each quarterback over the last five seasons (minimum 250 dropbacks) so you can see the evidence going into the projections. All of these ranks are now based on AQB, including the 2024 numbers.

Older data is decayed over time, so the 2023 and 2024 AQE data matters more than those from pre-2020. That said, older data can’t be fully discounted, or else you miss bounce-back performers of great quarterbacks returning to form, like Aaron Rodgers in 2020 and 2021.

My starter assumptions on Thursday morning going into Week 4 are Justin Herbert, Jordan Love, and Skylar Thompson. It’s already been announced that Gardner Minshew will remain as the starter for the Raiders.

“Percentile” is the mean (“best guess”) projection as a percentile of historical franchise quarterback results (min 2K career dropbacks).