Q: What does EPA/play actually tell us about QBs?

A: Mainly, it's saying how good they are at moving the chains, plus how often they're turning the ball over.

EPA adds context on top of those two things, but together they explain ~87% of a starting QB's season EPA/play.

Interestingly, these two stats—first down rate and turnover rate—aren't the most useful tools for projecting a QB's success going forward.

They're just quantifying such large-impact events that they explain a great deal of why a QB's plays had the overall effect they did.

I shouldn't equate them. A quarterback's first-down% explains 76% of in-sample EPA/play and predicts next-season EPA/play with an R^2 of 13%. Not bad!

Considered together, a QB's turnover rate and sack rate explain 38% of same-season EPA/play but predict only 3% of next year's.

A quarterback's sack rates predict themselves year-to-year with an R^2 of 20%. His INT rates predict themselves with an R^2 of 13%. (Fumbles lost being far more chaotic.) Neither one is very predictive of next-season EPA/play. https://twitter.com/Cooper_DFF/status/1265334939236675585

Interestingly, if you want to predict a quarterback's first-down rate next season, you're a bit better off looking at his completion percentage over expectation (CPOE) than the actual rate at which he moves the chains.

CPOE is a pretty effective tool.