Tight ends typically take time to develop. Taking the wrong TE in a rookie draft is not only a waste of a pick, but can clog a roster spot for years as you wait for production that never materializes. Missing on a TE can be a costly mistake. Fortunately, we have a secret weapon here at RotoViz, one that was able to predict busts with a 98 percent accuracy.
Phil Watkins (@Advantalytics) published his TE model here last year, and it was a smash success. The players predicted for the highest probability of success are guys that you’ll see at the top of the dynasty TE rankings, while players who failed to meet the 15 percent success threshold are a collection of busts and one-year-wonders. While it’s important to remember these are all just probabilities – even the best players have a chance to bust and some uninspiring guys can break out – knowing the odds before you commit draft capital to a player is crucial to making winning decisions.
This year, Phil has opted to pass the torch. He’s too busy taking on new, exciting challenges in sports analytics, and other mundane stuff, like helping find the cure for cancer.1 I’ll be your guide through the 2018 model results, and will follow up next week with a look at the bargains and busts we can find based on rookie ADP.
Before diving in to this year’s class though, here are some refreshers and background information on how Phil built the model and how to interpret the results:
- Tight End Prospecting: Beware of Fools Gold – Phil explains the making of the model
- Tight End Prospecting: 2017 Draft Class – Phil applies the model to the 2017 daft class
- Who to Target in the Best Tight End Class in Years – I use Phil’s model to compare success odds with ADP to find 2017 rookie values
- The data the model is based on comes courtesy of Jim Cobern (@Jimetrics), a metrics guru with a deep TE database and a very informative Twitter follow.
Applying The Model
For the sake of comparison, I’m replicating Phil’s excellent visualizations to break down the model results for this year’s class.
The first viz breaks out the four inputs to the model into separate charts to show how each prospect fared in the respective categories. Note that we don’t have 40 times for Dallas Goedert, Jordan Akins or Troy Fumagalli as all were recovering from injury during the combine and pro day period. For purposes of this exercise, I’ve estimated their 40 times based on this report on Goedert, the lower estimated time for Fumagalli, and Akins’ pro day short shuttle matching that of Hayden Hurst and Mark Andrews, who both ran 4.67 in the 40. Note that this is giving these players the benefit of the doubt here, and is probably overly optimistic.2
Players are sorted in order of their draft position to help visualize how they fare relative to their draft stock.
A few things immediately jump out when looking at this visualization. Hurst and Akins are both incredibly old relative to their positional peers. Goedert has an elite market share of receiving yards, even if it did come against lesser competition.3 Ian Thomas did not perform the bench press, however for the purposes of the model, bench press is categorical in its contribution to NFL TE success (21 or more, less than 21, did not participate), so no estimate is required.
This next viz shows the log-odds of NFL success for each player. This helps visualize the contribution of each category in determining whether a player passes the 15 percent probability threshold.
For some of the more borderline prospects, making it to 21 reps on the bench press could have pushed them into starter territory, however none of them missed that threshold by fewer then 3 reps, so it wasn’t even particularly close. Despite being the final two players drafted, Jaylen Samuels4 and Jordan Thomas both make it just past the threshold even without the benefit of the bench press. Meanwhile, Akins still fails to project as a starter due to his advanced age.
Finally, here’s the breakdown of the starter probability sorted by the ranks in the class.
Are you tired of hearing about TEs who were former basketball players? Well you’re in luck, because now you get to hear about TEs who were former baseball players instead. Both Hurst and Akins first pursued careers in baseball prior to returning to play college football, and both are extremely old prospects because of it. It’s impossible to excuse their old age as a product of circumstance without also casting a skeptical eye towards their market share numbers, which undoubtedly benefit from frequently being covered by players several years younger than them.
Hurst not only looks like one of the worst first-round TEs ever, he’s also almost the worst in his own class. Obviously his age hurts him the most, but he performed poorly in the bench press, and was merely good — not elite — in his Speed Score and market share of yards. You can believe the model is underestimating him because of his unusual path to the NFL, but there’s no denying that the rest of his profile outside of age isn’t first-round worthy. Much of the same can be said about Akins, however he went much later in the NFL draft and actually looks more intriguing than Hurst in a lot of ways.
None of the other players failing to meet the 15 percent threshold are much of a surprise. They were drafted in the later rounds, and it’s not shocking to see them with low odds of success.
Ever since his ridiculous combine, there was little doubt that Mike Gesicki would compete for the top spot in this year’s model. His 60 percent score compares favorably to the 61 percent that O.J. Howard scored last year. Gesicki has been generating considerably less hype than Howard though, and is going later in early rookie drafts thus far, making him look like a potential bargain.
The lack of an accurate 40 time for Goedert makes it hard to accurately predict him. Using the optimistic estimate of 4.67, he projects as the second best in the class with a 50 percent success rate. If his speed is actually closer to 4.75 though, he may actually only be the second best in the class with a success rate closer to 35 percent. Either way, his other metrics make him an excellent option and the only downside is his landing spot behind the entrenched Zach Ertz.
Perhaps the biggest surprise here is just how well Chris Herndon performed. While he didn’t excel in any one category, Herndon was above average across the board. He would have just barely failed the Rule of 15 without eking by with 21 reps on the bench press, but his well-rounded profile makes him look like an intriguing sleeper.5
The log-odds chart for NFL success shows how Mark Andrews would actually be the second best prospect in the class were it not for his mediocre showing on the bench press. The model likes him as the better TE in Baltimore, despite being drafted well after Hurst. The Ravens run a lot of two-TE sets, and Andrews has always been known as more of a receiver than a blocker. Hurst’s draft capital has him going earlier in rookie drafts, however the model indicates Andrews is the smarter play here.
Check back in the coming week as I take a look at how this class compares to classes from recent years, and look to find bargains based on how their rookie ADPs compares to similar players.
- At least I assume that’s what he’s doing as a biostatistician. He could also be developing new treatments for athlete’s foot. Also noble in it’s own way, but “cancer cure” is sexier, so I’m going with that. (back)
- In no case does the estimate make a difference between the player being above or below the 15 percent threshold though, and all seem to be in a reasonable range based on the players’ tape. (back)
- Note that Phil points out in his initial explanation of the model that, “college strength of schedule was insignificant to the model (p=0.365) in the test years (1999-2006), which may bode well for some of the smaller-school prospects on this list.” (back)
- He may not actually be a TE in the NFL or in fantasy, but we’ll include him for now. (back)
- Note that Herndon also may not have been full speed at his pro day after recovering from a torn MCL suffered at the end of the college season. (back)