Last year, I’d periodically find time to look at multi-week samples of my favorite receiving stat, TPRR, when the sample got a little more robust than just a one-game outcome. I’d hoped to do that this week, but got a little sick, so I decided we’ll do it in this space. It’s a little more complicated than the idea behind this input volatility concept, but it allows us to make some longer-term assumptions that might not be accurately reflected in the shorter-term projections, so there’s definitely an overlap here.
Before jumping in, I wanted to talk about this tweet, which is super cool.
One of the things with fantasy analysis is there are so many ideas for analyses that become commonplace quickly, and everyone has slightly different variations of them, and it gets lost a little bit what the original or best source might be (or frankly whether we are using the analysis correctly, or should even care what it tells us). And I’m not here to argue that copying a form of analysis is some form of plagiarism or anything, but rather that in this industry breakthroughs aren’t rewarded as much as whoever can promote their version i.e. most iterations are more or less the exact same stat/approach, sometimes with some super small alteration that amounts to window dressing but gets emphasized as something major.
Relative to this dynamic, there are very few examples of people taking a concept and meaningfully improving it in a way that makes the analysis better and more usable — finding a way to better show what the concept is trying to show, and giving us better paths to analyzing it. Those types of situations push the overall space forward in terms of how we might want to consider not just that idea but various other stats, and when something like that is achieved, it should be celebrated.
I think Anthony’s done that with his fantastic take on EP models that I really enjoyed this week, and hats off to him for it. What we have here is an EP model presumably built on all the common factors — line of scrimmage data, air yards, etc. — but then rather than assigning a rounded value to that opportunity, there’s a simulation method of 10,000 outcomes for each target, which creates a range of possibilities on the individual play level that can then be summed to create a range for what that player’s EP could be. And then the (awesome) cherry on top is dropping the caret where the player’s actual output lies, giving us this really clean visualization of the size and depth of the opportunity, plus how the player has performed relative to those expectations.
I immediately sunk my teeth into this and had a ton of thoughts on how it should be applied. Let’s blow up that chart and talk about some of my takeaways, but this thing is so good because reasonable minds can disagree with how I’m interpreting. The first few are team-related, since this is based on simulations and this visual immediately allows us to understand some of the team-specific outcomes so far.
Bad QBs/schemes
One of the first things that stood out to me was the team correlation in some spots. Here were a few that were immediately evident.
No one on the Steelers with meaningful volume has hit a 25th percentile outcome for what we’d expect on this sum of their individual types of targets. That’s pretty clearly a Mitchell Trubisky issue, and he either needs to play better or the team needs to turn to Kenny Pickett before we can feel good about any of these guys hitting their weekly ceilings. Among the takeaways, I’m more willing to give a pass to a guy like George Pickens in terms of his long-term outlook and dynasty value based on his early-season production. Diontae Johnson also looks like a guy whose production to date has more or less been the floor of what he is as a volume-earner, and there’s more room for upside than substantially worse production as we look over a longer timeline into the future.
Robbie Anderson and Ian Thomas approach or are just beyond the 50th percentile of their outcomes, in large part because Anderson hit for the 75-yard touchdown in Week 1 and Thomas hit for the 50-yard reception that same week. For both of them, that one big play accounts for more than 50% of their receiving yardage through three weeks, and we see other players that have hit for splash plays in this limited three-week sample that are well beyond the 50th percentile range. DJ Moore is below the 20th percentile, and Shi Smith is laughable at the first. The splash play guys being only “average” overall is pretty great evidence the rest of their target sample includes significant inefficiency, and the teammates show it’s a systemic thing in Carolina.
Corey Davis is another guy who hit on a long touchdown that is a clear efficiency booster but still sits around the 50th percentile overall, and Garrett Wilson and Elijah Moore are both below the 15th percentile. Tyler Conklin has run hot in a passing game that has otherwise been pretty poor efficiency-wise relative to its overall volume. That volume is likely to come down with the switch from Joe Flacco to Zach Wilson, but there has to be some hope the per-target efficiency could offset that some. Notably, Garrett Wilson’s whole range of volume has been far stronger than Moore’s.
Team-level regression candidates
Contrary to the above, there are some teams that look poor across the board but the context around their rosters suggests some potential actionability (these things always overlap, and the Jets for example could fit in this category depending on your stance on Zach Wilson).