Stealing Signals, Week 8, Part 2
Late games, SNF, MNF
I know I said no intro today, but I got a super cool comment I want to share. I really appreciate all the comments, including from those of you I hear from often, but while I can see the subscriber and reader numbers, I do run into the imposter syndrome I talk about sometimes when I think about longer-term subscribers, where I think anyone who has been reading me for any length of time has to have gotten annoyed and moved on by now.
I also always say I love the comments that come from people that clearly understand my approach, and give me different thoughts on how to analyze from a similar lens, which is just such invigorating food for thought in what is otherwise an often lonely endeavor writing this newsletter on my own (since leaving teams of fantasy analysts where I got that type interaction more naturally).
Anyway, Brett’s comment was a really fun one to read this morning:
Hey Ben! Long time reader, but first time commenter. I just wanted to say that I loved the intro, and it's exactly the type of content that keeps me coming back. The metrics you rely on are [in my opinion] the easiest to understand and apply concurrently with other stats or the eye test. To throw back to some previous content, it seems like most people who make models have the goal of eliminating the need for tacit knowledge, which in my mind is almost impossible for something this complex.
“It seems like most people who make models have the goal of eliminating the need for tacit knowledge” is one of the most profound things I’ve read in a while, and after I just wrote 8,000 disorganized words veering through a ton of things I wanted to hit on (but knew I didn’t have time to structure well, so I just went with the flow), it’s just so cool to see one of the key things I was trying to communicate conveyed so succinctly.
I also agree with the way Brett finished that sentence, “…which in my mind is almost impossible for something this complex.” Yes, absolutely. That doesn’t mean we don’t try, but the tacit knowledge and domain knowledge here absolutely does matter. A big point I was trying to make is even coming at this from a data-centric approach as a young analyst, and personally trying to fight the idea that you need inside football knowledge to do this well for many years — which is a huge reason I comment on how some of the specific models are misunderstood or fall short, because it’s learned experience and I’ve been there and misunderstood that same stuff before, for years — at this point in my career I’ve never been more sure that you do need solid domain knowledge to do football data well.
(And there are a lot of analysts — even some that have been around for a long time — that have just never acquired that domain knowledge. On the flip side, there are some data geeks who so clearly do try to acquire that, and some that you see really learn the game and they do profound stuff. Additionally, I’ve heard from some of you younger-than-me analysts in the past 12 or so hours, and I totally understand the feeling of inadequacy that you’re never going to understand it all, but I promise with an open approach and diligence, you are growing more than you realize. And maybe a little patience, which is rich coming from me, because I was trying to break this whole industry from Day 1, getting way out over my skis at times. And of course there is still so, so much that I don’t know, but somewhere on this journey — probably about the time I wrote the tacit knowledge piece — I internalized that learned experience is what it is, and it builds and that does sometimes just take time. But if you’re looking in the right places and trying to learn, the payoff is honestly there. And I mean, there’s a ton of great football analysis out there, so it’s not really that hard to “know ball” in 2023, except probably not even 10% of the analysts doing football work actually do, and are able to synthesize it well, so when people do get there, their work shines, and that’s absolutely achievable.)
Then there was a second part of Brett’s comment that got to the collaboration point I was making!
I will include one challenge as well - in my mind, we should be exercising extreme, extreme caution when ascribing 95+ percentile outcomes to two separate players (especially Tyreek and AJB, who are in very similar situations). I personally find it more likely that there is some underlying reason - whether it's becoming more of a focal point in their second year in the offense, a shift in league-wide offensive philosophy, or something different. It's not a very actionable observation, but these are the kinds of things I try to watch for since I believe being on the early end of larger trends is one of the biggest edges to be had.
I love this. It’s exactly how I think. So first, on the 95th percentile thing, I completely agree, but also (and this is semantics) my thought process was to be at the 95th percentile is a 1 in 20 outcome. And defining the ranges of outcomes for these players coming into the year included negative outcomes, obviously, as well as injuries and teammate injuries and all that disaster stuff. But then it also included a whole bunch of “good not great” outcomes where there was production but not on these historic levels.
When we get to where Tyreek Hill is the fastest player to 1,000 yards ever, that’s not something I would have predicted as better than a 20/1 chance, right? Even as someone whose projection on Hill did show some real statistical upside given the RPO offense and the way I was willing to push the consolidated target share (so basically all I really needed to get even further to Hill’s ceiling was to say, “Jaylen Waddle is a nonfactor or is hurt”).
But even as my pretty clear WR2 in projected receiving yards — I had Justin Jefferson at the top, but Tyreek within 100 yards of him and then no one else within 150 yards of Tyreek’s 1,616-yard projection — my 95.1 yards per game projection is frankly nowhere near the 126.8 he’s actually put up so far. Hell, that number is well clear of his 100.6 from last year.
So in terms of preseason range of outcomes, and looking at what he’s done statistically to this point, not necessarily where he’ll finish, I stand by the 95th+ percentile outcome, despite fully agreeing with your take that we should reserve that. (The A.J. Brown story is similar — he’s at 117.4 yards per game, after a career-high 88.0 last year, which by the way both Hill and Brown had receiving yards per game totals last year that were good enough to be in the top 160 such seasons in NFL history, and this year they’ve reached heights that are currently the second- and seventh-best per-game totals of all time.)
My semantical defense of dipping into the “95th+ percentile and maybe 99th percentile” well aside, I loved the point that basically says, “it’s not the 95th or 99th percentile going forward.” I’m right with you there Brett — I definitely think there’s an actual explanation for why this happened, and I think the only thing we differ on in terms of using that 95th percentile commentary is I had been meaning it to apply to a preseason range of outcomes rather than where we sit currently.
Knowing what we know right now, I absolutely think there’s something to both of these guys doing this in the same season. And I have kind of hinted at the RPO stuff a couple times, but I like what you’re saying about a shift in league-wide offensive philosophy, as well as other stuff. It does very much feel like we’re moving back to where No. 1 WRs can get to legit production ranges, something that fell off when the league went to more three-WR sets and got more WRs on the field generally (I’ve referenced this before, but Brandon Marshall once had a 40% target share for a full season; on top of that, Marvin Harrison, Calvin Johnson, and Julio Jones all had 200-target seasons in their careers; elite No. 1 WR stuff like that had petered out in recent seasons but it does feel like we’re seeing a leaguewide shift toward more of that in 2023, which is super exciting honestly because it was annoying for a couple years when it seemed more difficult for teams to get the ball to the best WRs).
I also really like the idea of the second year in the offense thing, with both Hill and Brown obviously being traded in 2022 and bucking the trend of WRs underwhelming in new offenses — something we tried to debunk a little last year as players this good hadn’t ever really been moved until more recently — but still having a full season and offseason in a new offense and with a new QB to potentially grow even further.
All of this stuff from Brett was too great not to share, including his final note that being early on larger trends is one of the biggest edges we have, so I had to share this morning. Also, Leonard Fournette is joining the Bills and starting on the practice squad; I don’t have a huge reaction and think it’s probably worst for Latavius Murray, although James Cook needs those passing snaps and Murray has been playing too much there, with Fournette being an even bigger threat to steal some of that stuff. I don’t think he’s a real threat to Cook as the lead back, but perhaps is more of a threat to Cook’s share of the backfield than Murray has been, which has already been a nuisance.
On the flip side, it very well might be another Dalvin Cook situation.
Also, the trade deadline hit today. Joshua Dobbs looks like a positive for Minnesota’s pass catchers, and Ian Rapoport said on NFL Network that Jefferson is “coming back sooner rather than later” which is awesome to hear confirmed. It’s still a pretty big bump down from Kirk Cousins, especially given Dobbs will need to learn the offense and you’ll see more conservative play-calling for a bit as a result. And ultimately, his mobility means you’re never going to get near the level of volume in the passing game that had Cousins leading the league in attempts and completions at the time of his injury. But the confirmation Jefferson is close — he’s one week away from being eligible — and the addition of at least a fringe NFL starting QB, are both massive positive notes for a Vikings’ team that also traded an interior offensive lineman Tuesday in a move that certainly doesn’t scream, “We’re going to be run heavy the rest of the way.”
Lastly, I grabbed a slightly different visual from Sam Hoppen this week for the HVT section at bottom, and added 15 quick-hitting bullet points about backfields, so just wanted to highlight that there’s a chunk of interesting RB tidbits buried down there.
Let’s get to the games. Data is typically courtesy of NFL fastR via the awesome Sam Hoppen, but I will also pull from RotoViz apps, Pro Football Reference, PFF, Next Gen Stats, Fantasy Life, the Fantasy Points Data Suite, and I get my PROE numbers from the great Michael Leone of Establish The Run. Part 1 of Week 1 included a glossary of important statistics to know for Stealing Signals.
You can always find an audio version of the posts in the Substack app, as well as easier-to-see versions of the visuals at the main site, bengretch.substack.com.
Seahawks 24, Browns 20
Key Stat: Jaxon Smith-Njigba — 0.0 aDOT, 66% routes (lowest since Week 3)
It’s not accurate to say that Seattle prioritized undrafted rookie Jake Bobo (2-2-23, 1-3-1 rushing) over first-round pick Jaxon Smith-Njigba (4-3-36-1), but it sure felt like it. I’ve gotten a lot of questions about Bobo, who remains a limited-skill-set player that the Seahawks are nonetheless using in specific packages. His role is disproportionately weighted to the red zone, which puts more of a fantasy spotlight on it than Bobo’s actual upside deserves.
JSN meanwhile did slip back to a route share of 66%, a particularly disappointing development after he was over 80% not just last week but the week prior, when both starting WRs were active. JSN also saw his aDOT fall back to right at the line of scrimmage, and probably his furthest downfield target came on an out where he shut down the route, Geno Smith expected him to keep going, and the pass went right to a defender for a nearly disastrous pick-6, if not for a drop. JSN did bounce back and turn a WR bubble screen into the game-winner, and Geno noted in his press conference that route was based on alignment, where JSN had to make a read and adjustment and did well to do so (great example of the first-read stuff from yesterday’s intro where it became a quick-hitter due to defensive alignment, but wasn’t specifically called that way). JSN’s been productive since the bye, averaging over 9.0 yards per target in all three games, and I don’t expect him to be in this 66% routes range in many more games this year, but it is definitely a short-term concern.
The one game where Bobo didn’t have a situational role was last week with DK Metcalf (14-5-67) out, but Metcalf returned to his normal role here, with Tyler Lockett (9-8-81-1) seeing fewer targets but showing more efficiency. What else is new.
There actually was some new usage in the backfield, where Kenneth Walker (8-66, 2-1-4) played very well but ceded more work to Zach Charbonnet (5-53, 2-2-21) than we’ve seen. Charbonnet actually finished with more snaps at a season-high 57%, with Walker down to a season-low 43%. I don’t think Charbonnet is the lead going forward or anything, but it may be closer to a 50/50 split. Charbonnet did look very good, the best we’ve seen him look, and it felt like Seattle just wanted to reward that in a dual “hot hand” and “get your rookie going” sort of way, but not to punish Walker or anything, just due to Charbonnet’s limited role to date. I liked it; it felt like good coaching. Perhaps the bigger note thought is Charbonnet doubled up Walker in routes, and Walker was down to 29% after being at 48% or higher for each of his past three games.
Jerome Ford (9-37, 2-1-2) had an odd day, where he got an early touch but then went to the sideline for a long stretch, and it felt like he was either reinjured or had just been given a token touch but the team didn’t trust his ankle. But then after just the one first-half touch on the game’s opening drive, he got nine more in the second half. Between there, Kareem Hunt (14-55-1, 1-1-12) and Pierre Strong (10-41, 1-1-41) both got a lot of first-half work, with Hunt again getting the short score on the team’s only RB green zone touch of the day. His effectiveness in short yardage is likely going to keep those opportunities coming for him. But then Hunt didn’t get any run in the fourth quarter, when Ford got back into the mix, which was really hard to parse. The three backs wound up very level in snaps, and even closer in routes. Strong did stay involved throughout the game.
Amari Cooper (11-6-89) and David Njoku (8-4-77-1) led the way in the passing game, with Elijah Moore (5-2-30) lagging a bit targets-wise, and no other WR or TE seeing multiple targets, creating the type of concentration we like to see, if not necessarily elite production.
Signal: Jaxon Smith-Njigba — fell back to 66% routes, but also third straight game of at least 9.0 yards per target, continues to produce; Kenneth Walker — 29% routes (lowest since Week 3 as Zach Charbonnet got more involved, had been at 48% or higher in past three games); Kareem Hunt — another successful goal-line carry on team’s only RB green zone touch
Noise: Jake Bobo — 3 TDs in past five weeks (red zone specific role, but limited upside and only 26% routes with WRs healthy here); Zach Charbonnet — led the backfield with 57% snaps (previous season high was 44%, and don’t expect he’s taken over the backfield from a productive Kenneth Walker, but it does feel like we’ll see something closer to a 50/50 split); Browns RBs — all three backs played similar snap shares, but there was a weird rotation, with Jerome Ford getting one early touch then nothing until nine more touches in the second half, Kareem Hunt leading the backfield most of the day but then not getting a fourth-quarter touch, and Pierre Strong mixing in throughout
Ravens 31, Cardinals 24
Key Stat: Trey McBride — 86% routes, 14 targets, 0.83 WOPR