Seth Walder
@sethwalder
ESPN NFL Analyst. DMs open.
Big move over the weekend, with Healy becoming the latest front office member with a quant background to reach AGM title. He's well respected around the league in analytics circles. Here's a line from my analytics staffer survey in 2024:
How did the Myles Garrett trade affect the Rams' projections? From SportsCenter this morning:
ESPN's Football Power Index is out! The Rams are No. 1, the Seahawks are No. 4 and the Patriots are...No. 14! Projections, most likely Super Bowl matchups and more in today's story: www.espn.com/nfl/story/_/...
Sometimes it’s hard to know if a selection is an “actual” reach or if the league was simply higher on the player than the outside world realized. So for these six players, I asked!
Who will be available at pick No. 23 (Eagles)? Live update from the Draft Day Predictor through Pick 15:
62% of Jets users picking at No. 2 selected Arvell Reese over the last seven days in our draft simulator. www.espnanalytics.com/draft-sim-re...
Here’s a scatter plot of predicted yards per route run and actual yards per route run. Note that I used padding at the 5th percentile for players who didn’t average up to 200 routes per year.
That is both because it is down on the first three but also high on the latter two relative to their expected draft position. Here’s the top 50 projections from 2021-2026. Some players, like JSN, are missing due to not playing enough in final or penultimate college season.
Small project: Spent some time over the last week seeing if I could create a wide receiver model that incorporated data from Telemetry sports that could better predict a player’s yards per route run in a player’s first four years than the draft itself.
This is a year-old chart from Ben Baldwin on twitter, but sharing because it's such an important chart. It demonstrates the league does not sufficiently bake in positional value in the draft. If it did, those lines would all be roughly the same.
And here’s that same view for the class of 2026. Omar Cooper Jr. notably on the “faster than his 40” side and Makai Lemon on the opposite end.
That way, we can compare actual 40 times to play speed adjusted 40 times! Anyone to the right of the center line is predicted to be faster than their actual 40; anyone to the left, slower.
New ESPN draft simulator feature: how often users select each player at each pick! Can see here that Jets users select Makai Lemon at 16 more than anyone else, and far more often than the draft day predictor forecasts. www.espnanalytics.com/draft-sim-re...
Update on calibration of my free agency contract projection model. Feel pretty good about this!
Initial results comparing my FA contract model's predictions to actual results for contracts OTC has so far. Model is based primarily on media rankings. Very happy with these results! A testament to @greggrosenthal.bsky.social, @profootballfocus.bsky.social, Matt Bowen and Daniel Popper, too!
Let's try this again: Just ahead of free agency, here are the top 90 APY predictions from my media-rankings-based free agency contract model. Updated to remove signed free agents, add newly-added free agents, and reflect latest rankings.
Just ahead of free agency, here are the top 90 APY predictions from my media-rankings-based free agency contract model. Updated to remove signed free agents, add newly-added free agents, and reflect latest rankings.
Here are past out of sample (leave-one-year-out) predictions for edge rushers. This is a particularly clean position, they aren’t all like that.
New project I’m excited about: a model that forecasts free agent contracts based primarily on media rankings. The model utilizes rankings from ESPN (Matt Bowen), NFL.com (@greggrosenthal.bsky.social), the Athletic (Daniel Popper) and PFF.
Talked about the Tytus Howard trade and why I think it’s a misallocation of Cleveland’s resources
the final and official darnold meter chart of super bowl LX darnold-meter.com
That's an extremely obvious error by Mike Vrabel. Needed a 42% chance of converting to justify going for it. We estimated they had a 67% chance of conversion. Even if the model knew it was a long 1 to go, it still would have been miles ahead of the 42%.
darnold-meter.com is super bowl ready. new "since kickoff" button added