There's a lot of 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, and 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.

A Short-Lived Result

Last week, I published the results of the 2024 rookie receivers through the lens of my prospect model-- always my most-requested piece of the year. I noted how tight the race was for the title of Top Rookie Receiver, writing, "McConkey led the pack based solely on yards, but including touchdowns edged Thomas ahead." Thomas scored a 125.5, while McConkey scored a 124.1.

That was true at the time, but kind of a lot has happened since then. McConkey caught 9 passes for 197 yards and a touchdown in the Wild Card round; the next-best Chargers receiver recorded 2 catches for 16 yards. If we add the playoff data, McConkey's score rockets up to 128.4-- the fourth-best total ever recorded, trailing only Odell Beckham Jr, Ja'Marr Chase, and Justin Jefferson.

But can we do that? Can we just add playoff data to the regular season data? Should we?

Do the Playoffs Matter?

To start, the answer to "can we just add playoff data to the model like that?" is yes, it's my model, I can do whatever I want with it. The "should we?" question is the more interesting one. It has a habit of cropping up from time to time. I remember hotly debating it after Gabriel Davis' 8/201/4 receiving game against the Chiefs to end the 2021 season.

Here are some arguments against including playoff data in the model in particular or any sort of overall evaluation in general:

It's Not Fair to Players Who Miss

Ladd McConkey made the playoffs, but Brian Thomas Jr. did not. This isn't because of any disparity in play between McConkey and Thomas themselves; McConkey had a better quarterback and a better coach; he played with a better defense. In many ways, it's not "fair" for Thomas to lose his lead on a weekend he didn't even get a chance to play because McConkey enjoyed advantages that had already given him an edge all year long.

It's Not Especially Fair to Players Who Make It, Either

On the other hand, playoff games are typically played in worse weather, and playoff opponents typically have above-average defenses. If a receiver plays poorly in a cold-weather game against a Top 5 defense after a strong regular season, it's not fair to them either to see their score fall when most of their peers were already planning their vacation.

Playoff Data Isn't Especially Predictive of Next Season, Anyway

After averaging 2.5 touchdowns per game in the 2021 playoffs, Gabriel Davis averaged fewer than 0.5 touchdowns per game during the 2022 regular season. Again and again, we see a big game in the postseason have a large impact on a player's dynasty value, only for them to disappoint the following season.

Ignore Those Arguments

Life is unfair. (My 12-year-old might prefer "womp womp" there.) Even without the playoffs, some rookies already play harder schedules or in worse weather than others. Some rookies get to play with Justin Herbert; others find their fate tied to Mac Jones. The point of the exercise is not to be fair but to be predictive.

And while playoffs (on their own) are not especially predictive of next season, neither is any other small sample. This is the case for incorporating playoff data in the first place: study after study (after study) finds that using an entire sample is more meaningfully predictive than slicing it into smaller subsamples that might seem superficially more relevant (such as "performance in the second half of the season" or "performance with this particular QB" or "performance with this other player out"). 

Any football game where everyone is trying to win produces good data. Football gives us few enough datapoints as it is, we'd be foolish to voluntarily toss some out. I don't think we should overweight the playoffs-- one game is still one game-- but the subtitle of this column is "Should we include playoff performances when evaluating players?" and the answer is yes. We absolutely should. 

So You'll Include McConkey's Playoff Performance In His Score?

Oh, absolutely not.

McConkey's playoff performance won't be included in future results because no one's playoff performances are included in the results. Perhaps at some point, I'll revisit and add all postseason data since 2006, but I won't apply it to anyone unless I apply it to everyone.

This reticence might not be for the reason you think. I don't think it would degrade the model's performance; quite the contrary, adding data for McConkey (and only McConkey) probably makes the model more accurate.

If Data Gremlins got into my spreadsheet and deleted two games worth of data from every player, and then a Data Genie showed up and offered to magically restore 20 of those games at random, I would take him up on the offer even though it wouldn't hit all players equally. It's not a perfect fix, but as I always say, you shouldn't let the perfect be the enemy of the good. More data is more better.

(Data Gremlins and Data Genies? Sounds like a premise for a bestseller. None of y'all steal it or else.)

What's the Harm in Including It?

If More Data is More Better(™), why not just add the McConkey data and call it a day? In fact, I don't even have to add it-- I already put it in the sheet, I can just leave it there. It's going to take extra effort to go back and delete it again once I'm done writing. Y'all know how much I hate extra effort.

But if the goal is getting this things right, there's one word that should strike fear into the hearts of anyone building a model or even just trying to draw valid conclusions in general: "discretion". (Just typing it gives me the shivers. Not really, though.)

In the very first Dynasty, in Theory-- written more than a decade ago-- I warned about the insidious danger of confirmation bias. I could say I wrote it first because it's the most important thing, but I think that gives me a bit too much credit for the amount of thought that goes into deciding what to write about in any given week. It is a pretty important thing, though.

The trouble isn't that confirmation bias causes you to focus on things that aren't true. It doesn't have to-- there are more than enough true things which, if selectively emphasized, can make whatever case you wanted to make in the first place. The only defense is to avoid selective emphasis entirely.

If I find myself picking and choosing which data to include for which players, then I have lost the plot. The point of building a model is to remove-- to whatever extent possible-- subjectivity from the process. If I am going to abandon that, I should just abandon model-building altogether. I can judge players on memes and whimsy and assign scores by fiat.

So Who Had the Best Rookie Season, Then?

If you think this model tells you that in the first place, then you have lost the plot, too. It doesn't tell you who had the best season; it merely tells you who had the best combination of touchdown-adjusted yards per route run and routes per game per team pass attempt per game.

Given two players with significantly different scores, there's a reasonably good chance that the player with the higher score had a better season than the player with the lower score, simply because gaining yards, scoring touchdowns, and getting on the field are generally pretty good things. But when we're talking about a 124.1, a 125.5, and a 128.4? The model doesn't have the kind of fidelity necessary for me to say there's any meaningful difference between those scores.

(Never mind that the model doesn't consider anything about a player's role, responsibilities, supporting cast, game scripts, or... anything, really. It's a bit of a wonder it even works at all. But fortunately for us, it does.)

Whether McConkey scored a 124.1 or a 128.4, I'd still prefer Brian Thomas Jr. in dynasty. I think his slightly better draft capital matters, I think his size and athleticism raise both his ceiling and his floor, and I appreciate that he already produced in terrible circumstances, which removes a bit of the risk. But you have my permission to prefer either of these two phenomenal rookies-- and even to consider either of McConkey's scores to be his "real" score.

We shouldn't unnecessarily introduce subjectivity at the model level, but some subjectivity in the process of valuing players is unavoidable, and I don't think we should want to avoid it even if we could. The model is the model, but you are not the model; draw whatever conclusions seem worth drawing, hold whatever opinions seem worth holding. The algorithms might be coming for our jobs; we shouldn't let them take our teams, too.