Saurabh Rane

Methodology was pretty rudimentary. Starting point was prior season EPA, prior strength variance ratio, normal-normal model for posterior & then let it rip.

Tons of room for improvement, e.g. QB/coach changes + Vegas wins in prior

As a reference, here's stable EPA/play in 2024

Surprisingly, it takes team EPA/play an average of ~4 games to loosely stabilize (within ~6 spots of final ranking) & ~8 games to truly stabilize (within ~3 spots of final ranking)

Going to try to be more x-consisting about posting here, so here's some W1 NFL thoughts

EPA, after 1 week, is incredibly noisy & may not be tell us much. BAL (probably) isn't the best offense ever

In an effort to cut through low sample size (& to learn Bayesian stats), here's a stable EPA/play