https://preview.redd.it/h3jdp82qys1c1.png?width=1214&format=png&auto=webp&s=2a018ed99a2bf55cba74524209eafb9f1f89dce9
The most efficient rushing team in the league, the Baltimore Ravens, still only averages -0.03 EPA per rush play. Averaging across the whole league, the average rushing play is -0.09 EPA and the average dropback play is 0.06 EPA.
Taken at face value, teams should abandon the run and just pass. This of course would be too simplistic as one could argue that the threat of a run helps unlocking the passing game and improves the EPA.
However, another way to look at this is perhaps EPA is just a flawed metric and is either too simplistic or is missing a key nuance in its modelling. Perhaps there's a flat EPA adjustment we need to apply to all plays that would make rushing EPAs positive? Perhaps too much weight is given to the explosive pass? Perhaps we need to adjust the era data from when teams rarely played two high safeties to counter today's passing league?
Nevertheless, I wonder if more and more OCs in the league are using EPA and other advanced analytics and coming to the conclusion you might when looking at this data that passing is far superior to running and ending up with too many teams trying to pass it on too many downs, abandoning the run and putting too much pressure on their average QB?
My guess: Passing EPA and rushing EPA aren’t independent. That is, rushing some optimal amount “earns” you more passing EPA than you “spend” in EPA by rushing, such that your net EPA is better than if you only pass.
Think of it like investing in a business. Invest a few dollars upfront in marketing, better employees, etc. (rushing) to improve your profit margins later (passing).
Of course, as with any business, if your profits increase by less than you invested, you’re going to lose money. Same here. If you run too much, your net EPA will decrease. So teams have to figure out that optimal balance.
Yeah there's going to be a different graph two different curves, one for dropback epa and one for rush epa with the x-axis being % of plays are dropbacks, and you're most likely going to see a big dip in passing EPA as the proportion of dropbacks increase.
You could have a third curve showing dropback_epa*(%dropback)+rushing_epa*(1-%dropback) and look for maxima on that to figure out your ideal proportions