As with the rest of these recaps, the ideas and analysis presented here are from Adam Steele. I’m just a dunce with admin rights.

We’ve seen a steady decline in quarterback play across the last month, and this week hit a new low. The unweighted average QBR for week ten qualifiers was a dismal 44.7. That would have ranked 27th in the league last year! This isn’t a surprise as offense tends to decline in the second half of every season as defenses jell and the weather starts to make an impact.

Here are this week’s numbers: [continue reading…]

Adam Steele is back, continuing to refine his methods in real time, and we get to benefit from it. Thank you Adam, for working it out publicly.

After tabulating the numbers for week nine I realized I needed to make one more tweak to the formula. Since EPA per play and QBR are both agnostic to volume, QB games with a low number of plays were disproportionately clustered at the top and bottom of the rankings. Obviously it’s harder to maintain an extreme performance over a larger sample than a smaller one. My solution was to regress EPA/P by adding 20 plays of 0.1 EPA (roughly league average) to everyone’s stat line before calculating their z-scores. This fix strikes a nice balance between efficiency and volume.

Onto this week results: [continue reading…]

Adam Steele is back with a brand new convention, and we thank him for keeping the site humming.

With the NFL trade deadline taking place this week, I decided to make a trade of my own. I’m ditching DYAR in favor of EPA per play. After Davis Mills placed seventh in DYAR by turning into Dan Marino down 38-0, I knew I had to switch to a metric that filters out garbage time.

Thanks to Ben Baldwin and his nifty site rbsdm.com, it’s easy to query EPA/P with various amounts of garbage time removed. After some experimentation I settled on a 4% filter; plays which occur when win probability is below 4% or above 96% are thrown out. The vast majority of plays are still counted but nonsense like the Davis Mills experience is rightfully ignored. To wit, Mills drops from -.149 to -.474 EPA/P with this filter applied.

Here are the week eight numbers: [continue reading…]