Toward a Comprehensive Model for Rookie Receivers

Stephen Gill

Recently, I finished creating and writing up an article on my first generalized linear model dedicated to better using past information to predict future receiving production. GLMs, in layman’s terms, help you model weird data, such as the probability of a success versus that of a failure. In that specific example, there’s a pretty intuitive success/failure that we’re interested in modeling in dynasty: breakout rates.

Sure enough, I’ll be covering rookie wide receiver breakouts today. For some time, we’ve known about factors (draft capital, situation, college production, athleticism) that matter when we try to predict what rookie prospects are going to see success and which ones won’t. However, to this point, we don’t have a great idea of how to weigh those variables together.

Enter this model. If you’d like to know more about it, or look at who it has liked or disliked in the past, you might want to read that introduction. Otherwise, the general takeaway is this: After a lot of time putting together various combinations of variables and seeing how well they work together in predicting receiving success (plus some diagnostic checking to see what really helps and what doesn’t), I came upon just five measures that, together, are very effective in predicting whether a rookie will ever have a breakout season.

Those numbers are: the round a player is drafted in (surprisingly, a little better than the actual pick), number of college games played, number of breakout seasons in college, maximum market share of yardage in college, and whether the player was drafted to an offense with a returning offensive coordinator.

It’s certainly unexpected that that specific set of variables works best — and 84% of the time, correctly gives a higher probability of success to players who ended up breaking out than those who didn’t. But that is indeed the case, and I went into full detail about those predictors in the model’s introduction, so I’ll leave my full explanation there. Now, let’s get on to the fun stuff: How does the model like this year’s crop of receivers? Here’s every drafted receiver with college production data available, plus Preston Williams:

word image 49

There’s… a lot going on here. Some stuff is more surprising than others, so I’m going to pick out the things that I find surprising here.

Not very optimistic?

With just two players given over 50% probabilities to have breakout seasons, it seems like the model is super down on this class. But, one should keep in mind with those raw probabilities that less than 30% of receivers end up having a breakout season. In fact, the overall expected number of breakouts from this class is about 5.3, compared to 4.7 from 2016, 5.4 from 2017, and 5.5 from 2018. One narrative that was common in the draft process (at least in my brain) was that the class made up for its lack of top-end talent with depth. These results back that idea up, as 12 different receivers finished with good-not-great breakout probabilities, between 10% and 30%. Meanwhile, the three previous classes averaged just eight such players per year.

A clear tier one, and a clear tier two

Another statistical study loves N’Keal Harry. Surprise, surprise. Much more striking is JJ Arcega-Whiteside’s finish with a 59% probability. Look at his college production, though, and there’s no surprise: He managed three breakout seasons despite playing in less than three seasons’ worth of games, and had a reasonably high peak season. Unfortunately, we won’t figure him out very quickly in Philadelphia, but eventually, he’ll be an interesting case to look back at.

Hello DK Metcalf

Metcalf had long been a favored punching back for dynasty stats guys throughout the draft process. But, one thing I like about this model is that it does well in accounting for missed playing time. Here, Metcalf stays afloat, despite a single breakout season, because the model takes his missed time into account. (Perhaps you think that this would give him the benefit of the doubt with durability concerns. That might be true, but this model is more focused on finding success provided the player stays on the field.)

Four very different prospects with bad draft capital, yet decent breakout probabilities

Dillon Mitchell was a sleeper of mine in the pre-draft stage, a strong route runner who developed into a go-to-guy for Justin Herbert. But a draft with fewer spots than capable receivers saw him go in the seventh round and plummet in potential dynasty value. Scotty Miller put up some huge market share numbers at Bowling Green and remains on the Buccaneers roster, from what I can tell.

You might have heard of the other two? Hakeem Butler’s drop to the fourth round of the NFL draft was immensely disappointing both for film guys and, potentially, stats guys, after he put up a couple pretty strong seasons at Iowa State. Preston Williams, meanwhile, went undrafted entirely, due to off-field concerns.

Those probabilities certainly look too high for Mitchell and Miller, but look at this foursome together: The model, collectively, expects one of them to break out. Given Williams’s promising start, it may actually be right.

What happened to AJ Brown?

The low grades given to Mecole Hardman and Parris Campbell are understandable, given their low college usage numbers, but Brown put up some strong seasons at Ole Miss. The problems for him are subtler. First, his 11 games played in his freshman year, without a breakout, hurt him. It seems nitpicky to downgrade a guy for not blowing up immediately, but this is just one part. His 32% max market share also isn’t the best — it’s good, but it’s not great. Of course, that number was driven down by the likes of Metcalf and DaMarkus Lodge, but target competition has been an issue for people using market share for any purpose.

Finally, the weirdest predictor that I hadn’t talked about until now: He was not drafted to a returning offensive coordinator. You might ask, “So what?” This is what I wrote in the introductory article:

“My best guess is that a returning OC generally has an established, successfully functioning offense, and would know the best way to put a new receiver to use. It doesn’t seem that meaningful, but it has (coincidentally or not) been a huge indicator the past 15 years: 29% of receivers drafted to a returning OC had breakout seasons, while just 21% of those without a returning OC had breakout seasons; 25 of 30 receivers taken in the first round to returning OCs broke out eventually, while just 9 of 23 receivers taken in the first to new OCs broke out. Perhaps that association weakens or even disappears over time, but in a sample of 321 receivers, it seems significant for now.”

Without any more evidence to contract that much statistical power, offensive coordinator returns appear to be very important. As for the direct impact it plays here, I messed with the numbers, and in the hypothetical scenario where Matt LaFleur returned to Tennessee as OC, Brown’s chances would’ve increased by about 10%. Another small hit.

Simply put, Brown takes a couple of dinks that collectively have a significant negative impact on what seems to be a strong profile. I’m intrigued to see how his situation progresses in the future.

Gross third-rounders

Miles Boykin leads the four lackluster third-round picks with a 17% chance to break out. In his case, a single decent breakout season and not-too-many college games played are enough to keep him from ruining his third-round draft capital. Meanwhile, Diontae Johnson put up two breakout seasons, but took 48 college games to get there.

Still, Jalen Hurd and Terry McLaurin are the two interesting cases here. Hurd’s numbers are artificially low as he played as a miscast 6’4” running back at Tennessee and had little receiving production to show for that time; Julian Edelman broke out despite a 1.5% probability of breaking out after playing quarterback at Kent State, and Hurd could follow that lead. McLaurin, on the other hand, doesn’t have nearly as great of an excuse for his lack of production. However, he’s blown up to start his career in the NFL. It would be virtually unprecedented for a player without any collegiate breakout seasons and just a 13% maximum market share to break out in the NFL — especially if he were to do it so quickly.

What next?

These are interesting results (to me at least), but they answer just one question of many. I’ve been looking into some other queries, such as: How likely is a player to sustain his productivity, after breaking out for the first time? How likely is a player who hasn’t broken out yet, and has played at least one season, to break out in the future? Given what we know about a receiver’s past, how likely are they to have a breakout season next year? Do these projections change significantly once we’ve seen players’ starts to the new season? I’m hoping to cover each of these topics in the future using similar processes. For now, let’s count the days until N’Keal Harry is healthy.

stephen gill