All roads lead to A.J. Brown

How to evaluate WRs and fantasy football in general

I’ve gotten some questions this offseason about my reliance on TPRR as a metric, including commentary on related metrics and context that can be applied to TPRR. One critique interestingly posited I spend too much time talking about my favorite players, which I won’t even begin to deny, and I bring up because I’m once again going to use A.J. Brown as a muse to try to discuss some Tacit Knowledge concepts. I’ve done this before, notably at CBS a couple years ago when I used Brown to talk about team-level regression uncertainty, how to think through the intersection of opportunity and efficiency, and other concepts.

That post was the Tacit Knowledge idea in real time, where I said, “Yeah, X is important, but when A and C are both present, Y becomes even more important than X” for like 2,000 words. Those of you who have been around for a while know the point of this newsletter, from my perspective, was to make my fantasy football process available to anyone who wanted to read along. I think it’s super cool so many of you did from the start and more continue to, and I’ve heard from several who have read my stuff since the early RotoViz days that I’ve been at my best here, in this medium. I think that’s in part because I focus more on the stuff I know most intimately to better explain some concept I’m chewing on, rather than try to make sure to balance my analysis to a bunch of stuff I’m not sure really matters to how I’m playing the game. In other words, thought-provoking work comes from inspiration and passion and interest in the topic.

There are so many good fantasy analysts out there that it remains my contention that getting all your information from one source is faulty. There are people who are better and more equipped than me to talk about a variety of topics. My preference remains to try to add actionable advice to what else you’re finding, not to be your all-knowing guide. I hear so frequently about information overload from serious fantasy football players; I want to be viewed as part (but not all) of a well-cultivated set of analysis that helps you win.

That’s not to say I’m not going to try to cover as much as I can, and coming soon here in July will be my Offseason Stealing Signals posts, where I run through lessons learned doing projections. Last year, it took me until August to get into those, but it was important for me to get on them earlier this season. While I recently talked on Stealing Bananas about the pitfalls of projections, going through them is a major part of my process to research and analyze every situation around the league. And walking you through my process is, again, what I’m doing here.

Hopefully those posts, which will be behind the paywall, will lead to stuff like this about Cooper Kupp from last year’s edition. I wasn’t even that sold on Kupp last year; he’s certainly not a guy I was pounding the table for.

In the Signal and Noise section, I highlighted how TPRR was driving this logic, which relates to the latter part of today’s newsletter.

---

Fantasy analysis in 2022

Today we’re going to talk about TPRR as a stat, but also football stats in general, and how I think through what stats I choose, and most crucially how it informs whether or not to take an action on a player — draft/fade, trade for/away, cut/claim/bid — relative to the price. That last part is so frequently missed in what I would call otherwise good fantasy analysis, which is to say that the analysis breaks down the player very well, but then fails to understand whether or not those elements are baked into the market’s evaluation of the player, and whether or not the counterarguments are. Everything comes down to price, always.

Getting to a conclusion about whether to draft or not draft a player has been — for as long as I’ve played this game — and I believe will remain — for as far as I can see into the future — a mixture of an art and a science. I suffered from paralysis by analysis in my early years in this industry; it was thrilling to be lauded for the depth of my work, but applying it did not produce the results I thought it should. In fact, while I still won at a decent clip, I was a bit worse for a few seasons than I had been across my leagues before I started doing fantasy analysis. It was fairly easy to understand why, and thankfully that’s changed and these past few seasons have been some of my most successful ever.

The reason was that before I was doing fantasy content, I cared solely about fantasy as a game, and spent more time in the gray area. But in my first few seasons writing about fantasy, I wanted to be accurate. I wanted my player takes to stand out and I started focusing more on the data I was researching and became at times too certain about things I used to casually recognize I didn’t know enough about to have a strong opinion on. I have since learned to recognize that for all the research I do, more things are just unknowable than I previously wanted to believe. Long-time readers will recognize the Dunning-Kruger Effect, although I followed a slightly modified version where my competence was probably higher before my confidence rose early, because that newfound confidence drove me toward a less effective mix of inputs on the decisions I was making.

Lesson 1 in football is always that uncertainty reigns supreme. Before I was doing year-round analysis, I had to accept uncertainty, now I more intentionally embrace it.

[You’re no doubt wondering when A.J. Brown might come back up, and this point about uncertainty starts to get us there, but I have one more topic to meander toward.]

Similar to how player analysis in the fantasy industry has often fallen into a film vs. analytics battle, there’s a new rift forming in the fantasy industry. Draft structure has finally become mainstream over the past couple years, particularly with the explosion of best ball popularity driven by Underdog’s super fun formats. With the increased focuses on structure and correlation across huge portfolios of teams, player takes — a massive part of the fantasy industry — are being viewed as less important, with many accepting that most of that analysis will be captured by the market as defined by player ADPs. Meanwhile, there is still a huge amount of analysis dedicated to player takes that is naturally at odds with the increasingly more strategic side of fantasy, if the strategic side is willing to accept that most players are draftable around their ADPs.

I’m not trying to pretend this “rift” is any larger than it is, merely to point out two camps seem to be forming — with plenty in the middle. There are many analysts who are intelligently blending the two sides of the equation. The way I would describe them doing this is by limiting player takes to a select group of players the analyst is highly confident are mispriced, and allowing uncertainty to dictate a greater percentage of decisions beyond that select group than they may have a few years ago, i.e. letting ADP and then individual draft trends like whether a player is falling impact draft decisions more. It’s not my contention here that anything is more effective than something else, because I know of successful players that span this spectrum.

---

Headers are a lazy way to shift between disorganized thoughts

I saw this tweet from my old CBS colleague Jacob Gibbs recently.

Jacob Gibbs @jagibbs_23

2021 Target Hog Leaders: (Percentage of targets accounted for when on the field) 37% — AJ Brown 35% — Davante Adams, Antonio Brown 33% — Cooper Kupp, Kadarius Toney, Deonte Harris 32% — Tyreek Hill 31% — DK Metcalf, DJ Moore 30% — Justin Jefferson, Diontae Johnson, Deebo Samuel

This tweet measures something slightly different than TPRR, which Brown was also elite at in 2021. Both stats measure targets earned while controlling for the player actually being on the field (the main difference between the stats would come at the RB and TE positions, where players may be asked to block and thus not have an opportunity to earn a target, while at WR a passing snap is a route almost every time, and only minor differences in outcomes like quarterback scramble rate and sacks should shift things between looking at it terms of target share or targets per route).

So why did I share this tweet instead of TPRR numbers if I needed a massive parenthetical aside to explain it? Mostly because while Brown was very good in TPRR, he was still behind Davante Adams, Antonio Brown, and Kupp, and with this stat he looks even more elite. That’s not meant to cherry-pick; it meaningfully shifted my thoughts, even if some of the difference is probably just a measure that if Brown was on the field, he was often the first read, and there was a greater chance that if he didn’t win on his route, there wouldn’t be a pass attempt at all, because there weren’t a lot of other great options.

And yet, that cuts both ways, right? Defenses knew A.J. Brown was the top option to shut down in the passing game, and he thrived anyway. For some players, that becomes too big of an ask. And while Brown now has more target competition in Philadelphia, he’s shown an ability to dominate targets even among other stars, dating back to college when he doubled up DK Metcalf in receptions (129 to 64) in the 20 games they played together. Doubled!

So what this stat said to me was maybe even as optimistic as I am about Brown, that even I’m not quite aware of the possibilities as he changes teams. Brown is a really interesting player to view from a per-route or per-snap lens, because he hasn’t really yet been used on every snap or every route the way other top receivers are. This was a key element in Kupp’s breakout last year, where I noted this same phenomenon in the screenshot above, and then in 2021 we saw a Kupp who was running routes on nearly every pass attempt.

But what made Kupp elite last year was the efficiency, as well as a spike in TPRR that even with my comments above, I very much did not see at this stage of his career. I probably thought it possible Kupp could move to a 25% TPRR; he jumped over 30%.

As I read the above tweet about Brown, though, my thoughts were essentially, “What if he can still rise to an uber elite level of target dominance? What if he’s a top-three receiver in the NFL in that regard?” This is significant because when we split things out to the three layers of receiving — routes as the opportunity metric, targets as the first “skill”, and then after-the-target efficiency like catches and yards and touchdowns — Brown is often thought of as a player who thrives in that third bucket. He’s efficient. He’s a dominant YAC guy once the ball is in his hands who has scored touchdowns at a high rate because his size and physicality play up in the red zone. If that player has an ability to earn more targets than ever before, I want in.

And in his case, while moving to Philadelphia, there’s this possibility his random 75% snap games from his time in Tennessee — some of which were injury-related, sure, but others were maybe script-related, as Tennessee just did things kind of weird — just don’t exist anymore, i.e. that Brown is running routes on every dropback. We still have to account for long-term trends related to WRs changing teams as well as how comfortable we are with Philadelphia being projected as a low-volume pass offense, and also whether that might be wrong. But this idea of Brown moving from one low-volume offense to another might be underselling how this move could benefit him. The Eagles could throw more or Brown could just run routes like a true No. 1, which could be pretty huge if he’s truly uber elite at earning volume when he’s playing, in concert with his range of after-the-target efficiency outcomes.

---

What those thoughts told me about my process

Setting aside my A.J. Brown analysis, which was predictably glowing about his upside and yes I think he’s undervalued — surprise — that whole line of thinking I described above explains what I’m trying to do with TPRR extremely well.

The specific thing I want to address about TPRR is why I focus on it so much in relation to stats like market share or yards per team pass attempt — or even YPRR itself — which I’ve heard from trusted sources (but not independently verified, I should note) show to be more predictive in large sample testing e.g. looking at r-squared and those measurements. The reason is, for me, projecting forward is a blend of an art and a science, and it’s about what I’m getting out of the stat.

The r-squared of almost any football stat is not perfect, and there is a leftover amount of “unexplained variance” that is going to be relatively high no matter what, because that’s just sort of the deal. The fantasy football industry at large is obsessed with more advanced stats, more intricate ways of looking at things. But at the end of the day, what we need to be as analysts is actionable. No cocktail of advanced stats can clean out the unexplained variance in football, because it’s just there. That’s the art part of projecting. That’s where we talk about ranges of outcomes.

I’m going to make up some numbers, but if you can find me a stat or group of stats or model that has an r-squared of 0.6, but I can find something simple like TPRR that has an r-squared 0.5, the advantage to using TPRR is I can better account for what might be in the unexplained variance. For the stat with a 0.6 r-squared, you can take the stat or model at face value, and you can accept the 40% unexplained variance and probably do pretty well. What I love about TPRR is it’s solidly predictive and I think I can better account for the parts of the equation that aren’t predictable such that my end result is a better grasp of the player. It slices things more cleanly, and leaves out the highly-variable elements like role and team-level volume and after-the-target efficiency.

This is why I had no problem using Jacob’s stat above — it was so similar to TPRR that I could similarly control for the team-level stuff I thought might have impacted Brown, and I could similarly make a projection forward. But if I hear Gabriel Davis’s target share or per-team-attempt stats one more time, I’m going to lose it. No one is betting on Davis near his ADP for any reason other than there’s massive room for his routes to expand. If the routes don’t expand, that’s a losing bet, full stop. Looking at past data that measures his production against the team’s overall passing numbers sticks your head in the sand as an analyst to the reality that he was a part-time player, and the reality that he’s only going where he is because people are reading the tea leaves and saying there is room for him to now be a full-time player. And anything that buckets him into a cohort with similar past data ignores that he’s already an unquestionably unique player to analyze (the simplest way to argue this would be to say that among players he may be grouped with, very few experienced the type of ADP bump Davis has here in 2022, so you’d have to completely throw out a Wisdom of the Crowds element that analyzes the Bills’ actions and Davis’s specific situation to instead bucket him into a group with a ton of players no one in the market ever bought into).

I will say there is room to use Davis’s data to make a contingent argument about if Davis gets those routes, then he’s maybe still not good enough, if that’s the intent of the analysis. But beyond that, where you land on Davis is almost entirely predicated on how you’re projecting his team to act, or what his role will be, or whether he earns those snaps, or however you want to position that. Data basically can’t solve the equation of whether to draft that player, unless again, someone is arguing Davis’s per-route profile for a young player doesn’t present enough upside in an elite passing offense even if he’s a full-time player. Beyond that, it’s almost entirely whether you think he gets the routes bump. (Personally, I want shares of the routes uncertainty because I don’t buy into the if/then I just laid out. And for someone who might note Davis’s TPRR wasn’t really elite last year, I’d argue that contingent on his running a full set of routes, we should probably expect him to be the first read on more plays, as it’s indicative of an organizational buy-in that clearly wasn’t there last year, which is how I would apply some context to the specific TPRR data in this case.)

Things are almost never this cut and dry, as it seems to be with Davis, where there is one so dominant factor. I’ve joked before that every player in the NFL is a delicate little flower, with unique traits and profiles and all of those things. It’s why I hate comps, because no matter how similar, for every player type, there are hits and misses.

But for me to be as accurate as possible, I find there’s a huge advantage to what TPRR tells us. Targets have always been the lifeblood of fantasy scoring. Added research on TPRR like the ability to earn targets downfield, or in two-wide vs. three-wide sets, or thoughts I’ve presented about the strength of the other four players on a team who may be in a route at the same time (something Brown’s numbers probably benefitted from in Tennessee, since no one else could earn targets well) — that’s all context I think I can apply to the numbers I’m looking at, to come to a strong range of what might be possible in various potential roles in the coming situation the player will be in. But a lot of that stuff doesn’t really change — downfield players mostly stay downfield players (and vice versa), offenses often use two-wide or three-wide sets at high or low rates over multiple years under the same coaches, etc. At its core, the simplicity of TPRR is a feature, not a bug. Nearly every way I’ve seen people try to improve it — including my use of wTPRR — makes it harder for me to understand how I want to apply what I’m looking at. The simple, cut-and-dry version of TPRR gives me an understanding of how many targets were earned per route the player ran, which is a fairly predictive skill, and that’s all, and that’s very valuable. I can then adjust for the delicate little flower elements of the player in terms of his past data or his future expectations, and I want to be crystal clear I am not ignoring the great research that’s been done that provides more context into TPRR.

As you think through this, the biggest part of the additional adjustments that are necessary is, again, after-the-target efficiency. This is why I don’t find wTPRR to be as useful as TPRR, because I like applying aDOT’s impact after the target. And after-the-target efficiency is, again, a range. For a guy like Brown, I’ve argued before his range is basically from slightly above average to elite, because of his aDOT and skills. We saw that type of consistently above average yardage efficiency year over year from bigger wide receivers who could earn volume downfield like Julio Jones, Calvin Johnson, Terrell Owens, etc. This is the part of the Cooper Kupp equation I least foresaw — I didn’t realize his efficiency range at his age and with his established background reached up to what he posted, e.g. his career-high catch rate, near-career-high YPT, and monster TD rate. But the breakout stars are typically those players who have elite efficiency booms; these are the bets on talent we’re trying to identify and make based on a ton of factors including prospect profiles and past production.

By using TPRR to control for target-earning ability first — to set a baseline on the most important element of the equation — we can get an early lean against ADP, and then think through potential efficiency outcomes to come to an even better answer. There will be uncertainty in the second part of the equation, and there will even be quite a bit of uncertainty in the first part when we think through someone like Gabriel Davis, whose TPRR skills still represent a fairly wide range of possibilities, not to mention his routes. And there will almost always be a degree of art to the equation, where assumptions are made, both about stuff like team fit and role, but also about the past data and what it tells us about the future.

That’s all I have for today. I wanted to talk through that because I’ve gotten a lot of questions about it, and I want to note once again I think it’s absolutely fine to use team-level stats or other more predictive (in the aggregate) metrics if that’s the easiest way to think through things for you. I still have a backlog of posts I want to get out for you guys, and I suspect they’ll come in a flurry at some point. Until next time.