There's a lot of really strong dynasty analysis out there, especially when compared to five or ten years ago. But most of it is so dang practical-- Player X is undervalued, Player Y's workload is troubling, the market at this position is irrational, take this specific action to win your league. Dynasty, in Theory is meant as a corrective, offering insights and takeaways into the strategic and structural nature of the game that might not lead to an immediate benefit but which should help us become better players over time. (Additionally, it serves as a vehicle for me to make jokes like "theoretically, this column will help you out".)

Breaking the Code

Here at Dynasty, in Theory, I have a code: nothing practical, nothing actionable. We have a lot of really strong dynasty articles on Footballguys dedicated to giving practical advice for managing your teams. My focus is more on the weird, interesting, or conceptual. Some might accuse me of naval-gazing. (I certainly wouldn't argue the point.)

But to quote a fictional pirate, the code is more what you'd call guidelines than actual rules. So let's break it with some really practical, useful information!

I've been working for a while on a model to evaluate rookie receivers. In part, this is motivated by a lack of competition; there is a ton of effort devoted to valuing prospects before they reach the NFL but much less effort devoted to revising those evaluations once they're here, so it takes a lot less effort to find something of value and stand out from the crowd. Which is good, because I'm incredibly lazy and want to get the maximum return on the very least amount of effort possible.

But it turns out (for wide receivers, at least) that, despite conventional wisdom that players need a couple of years before we can really be sure of who they are, rookie seasons are actually shockingly predictive of the overall course of their careers. And best of all, they're predictive regardless of draft position. A model that tells you that a Top 10 draft pick with a monster rookie year like JaMarr Chase might actually go on to have a pretty good career isn't especially helpful. Not because it's wrong, but because you didn't really need a fancy model to tell you that.

But my model has a great track record at identifying completely off-the-radar players before they really break out. From 2006 to 2021, the Top 10 scores for players who were drafted outside the first two rounds belonged to Marques Colston, Terry McLaurin, Keenan Allen, Mike Williams (Tampa Bay version), Stefon Diggs, Cooper Kupp, Doug Baldwin, Tyreek Hill, Hunter Renfrow, Denarius Moore, T.Y. Hilton, and Amon-Ra St. Brown.

That's not a 100% hit rate (though it'd be foolish to ever expect perfection, and if a model ever does give a 100% hit rate, you can be confident that it's probably overfit). But I have startup dynasty ADP since 2014, and almost all of those guys were extraordinarily cheap to acquire after their rookie seasons. In that span, McLaurin was the 27th WR off the board, Allen was 9th, Diggs was 34th, Kupp was 30th, Hill was 34th, Renfrow was 65th, and St. Brown was 22nd. If you acquired all of those players at prevailing market rates, you probably built yourself a dynasty.

Crucially, once you have a player's score, knowing their draft position adds very little predictive power, meaning rookie performance is almost entirely new information that's not already accounted for in draft capital.

I've run my model early on this year's rookie class. With two weeks left, these scores are not yet final and will change a bit, but they're likely fairly close to where they're going to finish. I want to show the scores, but I also want these scores to serve as a jumping-off point for a larger discussion on the use and abuse of models when it comes to making decisions.

And since this column cares so little about the practically useful stuff, I'll even share the full scores for this year's class and every qualifying rookie from 2006-2021 before the paywall. If you're not a Footballguys subscriber and you've come this far, I appreciate the click and want to make your visit worth your time still. All the good analysis is going to be behind the jump, though.

I'll provide the full results below, but in the meantime, this thread shows how the 2021 rookie class scored and then provides screenshots of all 163 qualifying rookies from 2006 to 2021. The model is tuned so that scores of 100 are roughly average for the sample, with higher scores being better and lower scores worse:

Hey all, just wrapped up all 2021 data in my rookie model. As a reminder, this strictly looks at performance in a player's rookie year (both usage rate and production per route). But it winds up being super-predictive, especially good at identifying off-the-radar outliers. pic.twitter.com/Z9Pfp3zwSI

— Adam Harstad (@AdamHarstad) January 10, 2022

And this thread shows the early look at the 2022 class along with the four closest comparable players from the past, so you can see where this year's rookies rank in contrast:

Couldn’t resist a sneak peak at how the rookie WRs are faring in my production model. Remember, this score is strongly predictive of career outcomes.

Robinson/Burks aren’t anywhere near the minimum qualifying threshold. Watson is a hair short but should get there this week. pic.twitter.com/oKjpYzQEpy

— Adam Harstad (@AdamHarstad) December 27, 2022

I'll probably write much more about the model over the offseason, but for now, let's talk a bit about models in general and this model in particular.

Everything You Didn't Know You Ever Wanted to Know About Models

To start with, here's every qualifying rookie since 2006, along with where they were drafted in the NFL draft and how well they scored in my rookie-year production model.

(As an aside: if the numbers in this chart differ slightly from the numbers above, that's because each score is based on the average of all rookies, so when more rookies are added to the model, those averages change very slightly.)

Now, to explain a bit of the mechanics behind the model and address some common questions and criticisms.

What is the Model?

The core of the model is yards per route run (or YPRR), which I've studied for years and have found to be very predictive of career outcomes. Yards per route run is exactly what it sounds like-- the number of yards a receiver has gained divided by the number of routes a receiver has run. In my opinion, this is the only true "efficiency" stat for receivers. Many people like to use yards per target (or YPT), but YPT is a bad statistic for reasons both conceptual and practical that I'll detail in a bit.

I'm further adjusting YPRR by adding a bonus for every touchdown. I've tested the model in the past and found that scoring a large or small amount of touchdowns as a rookie does tend to carry predictive signal for the rest of a player's career.

There are several different ways to calculate "routes run". Some sites only count routes run on plays where a pass is attempted. Other sites count routes on any play where it's clear that the offense's intention at the snap was to pass the ball. (This means it counts routes on sacks and scrambles even though the ball was never thrown, but it doesn't count routes on draw plays or designed quarterback runs.)

I'm using the latter definition of a "route run", and under this definition, any value over 2.0 is extremely good. Of course, if a receiver runs one route all year and catches a 13-yard pass on it, he'll have a YPRR of 13. We need some way to ensure small-sample guys like this don't dominate the model.

I have two means of dealing with this. The first is a qualifying threshold; receivers need to run at least 250 routes to qualify for the model. On average, we see around 10 rookies a year reach that total. All of the rookies listed above have cleared that target except for Christian Watson, who is at 225 with two games to go. He's averaging 26.3 routes per game over the last six weeks, so I'm including him because it's very likely he qualifies (barring injury).

The second way I protect against small samples is by including a "usage rate" term. Currently I'm using (routes per game) per (team pass attempt per game). This means if a receiver averages 30 routes per game and his team throws 40 passes per game, his "usage rate" is 75%. The final model score consists of two-parts TD-adjusted YPRR and one part "usage rate". I've found that adding the usage term to downgrade players who were playing a part-time role and upgrade players who never left the field improves the model performance.

Why Yards Per Route Run?

There are two primary reasons. The first is conceptual: any "efficiency" stat should be "units of production divided by units of opportunity". Many think that the target is the unit of opportunity for the wide receiver; you can't gain yards if you aren't targeted. But earning targets is a skill; if a bad receiver and a good receiver are both running a route on a play, the quarterback is more likely to throw to the good receiver than the bad receiver. Role players might post huge numbers on a per-target basis, but they're only earning a target when they're comparatively more wide open.

As an example: Rashid Shaheed and Chris Olave are rookie teammates who are both playing well. Shaheed averages 63% more yards than Olave for every time he's targeted... but Olave is targeted 45% more often when he's on the field. (Plus he's on the field much more, to boot.) It would be silly to suggest that Shaheed is having the better season when Olave is performing as the #1 receiver and Shaheed is a situational deep threat. Their yards per route run totals are within 12% of each other, but Olave's usage rate is significantly higher (and Shaheed doesn't hit the minimum qualifying threshold, anyway).

The second and more important reason to use YPRR is simple: because it works better. If I rebuilt my model using YPT instead of YPRR (but kept everything else the same), the rookie receivers who would benefit the most are Kenny Stills, Mecole Hardman, J.J. Arcega-Whiteside, Hank Baskett, Henry Ruggs, Gabriel Davis, Tre'Quan Smith, Dante Pettis, Chester Rodgers, DeVante Parker, Terrance Williams, Equanimious St. Brown, Malcolm Mitchell, Robert Foster, Mike Wallace, James Washington, Tyler Lockett, Anthony Miller, Nelson Agholor, and Harry Douglas.

Wallace and Lockett notwithstanding, that is not a list of receivers you wish you had been more invested in for dynasty. It largely fits with the conceptual case: they're mostly situational deep threats who played a tiny b