ESPN: QBASE 2023
NFL Draft - Football Outsiders' QBASE projections for the QB Class of 2023 are here, led by Alabama's Bryce Young. The projections are lower than you might expect for Anthony Richardson and Will Levis. Click the link below for the ESPN+ version of QBASE, which will also run on FO in a few days.
11 comments, Last at 17 Apr 2023, 2:09pm
#1 by JacqueShellacque // Apr 16, 2023 - 10:08am
More important than QBASE 2023 should be looking into whether QBASE from the past has actually predicted anything, or whether the model is unfalsifiable (and therefore unscientific). For the model to mean anything, it absolutely needs to be able to tell the difference, within some confidence range, between this QB from a past QBASE, currently an elite NFL starter:
Mean Projection in Years 3-5: 656 DYAR
Bust (< 500 DYAR): 45.7%
Adequate Starter (500-1499 DYAR): 27.0%
Upper Tier (1500-2500 DYAR): 17.9%
Elite (>2500 DYAR): 9.4%
and this QB from the same year QBASE, currently barely hanging onto an NFL roster spot who's bounced around:
Mean Projection in Years 3-5: 419 DYAR
Bust (< 500 DYAR): 53.8%
Adequate Starter (500-1499 DYAR): 26.1%
Upper Tier (1500-2500 DYAR): 14.3%
Elite (>2500 DYAR): 5.8%
otherwise the concept of confidence intervals seemingly built into these percentage forecasts is being abused. And no, you can't simply eyeball the lower 'good' numbers and higher 'bust' % for the journeyman and say the model therefore works. The purpose of models should be to create a process by which a decision-maker will be more likely to make a good choice than a bad one. So the question is, does QBASE do a better job than a roulette wheel would of selecting those QBs that are more likely to succeed in the NFL (and that's even setting aside the question of what 'success' means).
#6 by richallen7 // Apr 17, 2023 - 3:05am
surely the whole point is that it's not an exact science and that there are a huge range of outcomes for each prospect. QBase is just trying to get under the skin and see who's more likely to make it. If you've been reading these over the years you'll know that they adjust the model all the time to make it more predictive overall. You can't reasonably expect it to "know" all these things when there's so much uncertainty. It just doesn't work that way, not least because a fair bit is unquantifiable (landing spot being the big one).
#3 by MarkV // Apr 16, 2023 - 6:48pm
I believe that the biggest problem QBASE has is that scouts and front offices adjusted fairly rapidly (but unevenly) to it, so the quarterbacks that it flagged as huge bust risks got downgraded on draft boards, and its safe picks got pushed up. The result was that as a purely retrospective when it first came out it was very potent, but then swings wildly as both FO and scouts tried to adjust to each other.