The Combine Drills That Really Matter For Wide Receivers

We’re going to expand the regression tree analysis applied to NFL combine measurables/drills for running backs, and now take a look at the physical profiles for successful NFL wide receivers.

Our results give us a few big picture takeaways, some intuitive, some not. The exercise isn’t meant to definitively answer the question of whether or not a particular wide receiver will be successful based on his physical profile. Instead, we’re going to use NFL combine data from 2000-2013 and find physical profiles that tend to perform better than others in the NFL.

This analysis does not incorporate production — which I’ve already explored somewhat — or draft position, but is meant to only look at measurables to see how focused we should be when the combine results get picked apart across the interwebs.

I’m defining early success for a wide receiver prospect as having at least one top-24, or WR2 season (PPR scoring), in his first three years.

Here’s how to read the regression tree nodes. The “yval”, or predicted value, in this case is the likelihood of success (from 0 to 1). The darker the node, the higher the yval.

tree

Unlike our running back analysis, the initial results for wide receivers included some counterintuitive decisions and divided into so many nodes that overfitting, or creating an overly complex model with poor predictive value, became a major concern.

Here is the original, complex decision tree that is generated when using all the combine measurements.

 

original_combineWRTree

Many props to anyone who can read and make sense of this tree. The first and most significant split is weight at 208 pounds, which makes sense and should hearten #TeamBigReceiver. Then things start to get a little weird. A faster short shuttle time (“twentyss”) comes out as worse for lighter wide receivers. The splits for heavier wide receivers make sense to start (longer broad jump being better), but then vertical jump has an ideal split between between 32 and 33 inches, and again slower shuttle times and shorter broad jumps help results.

When you inspect the regression tree results further, you’ll see that adding splits is creating more errors for wide receivers, whereas it was improving the running back results. The model was having trouble determining which measurables make a successful wide receiver, bolstering the notion that athleticism is less predictive for wide receivers generally, and that more of our wide receiver prospect analysis should be focused on production.

Here is a simplified version of the regression tree that minimizes error, but still provides some meaningful results to digest.

combine_wrTree_noSSBroad

Again, weight is the first and most important split, with heavier wide receivers’ average success rates more than double that of their lighter brethren. The regression tree isn’t able to filter wide receivers weighing less than 208 pounds in any meaningful way, so we should remember to be especially production focused on those guys.

Bigger wide receivers with quicker three cone times (less than 7.0 seconds) have double the success rate (0.3 to 0.15) of those with slower times, and we can find even higher levels of success if we again split by weight at a new threshold of 218 pounds.

Here is a list of those very heavy, quick wide receivers, listed by draft year.

Name College Draft Year Draft Pos Weight Three Cone Top 24
Tyrone Calico Middle Tennessee State 2003 60 223 6.73 N
Larry Fitzgerald Pittsburgh 2004 3 225 6.94 Y
Mike Williams USC 2005 10 229 6.98 N
Vincent Jackson Northern Colorado 2005 61 241 6.84 N
Marques Colston Hofstra 2006 252 224 6.96 Y
Brandon Marshall UCF 2006 119 229 6.96 Y
Dwayne Bowe LSU 2007 23 221 6.81 Y
Arrelious Benn Illinois 2010 39 219 6.78 N
Mike Williams Syracuse 2010 101 221 6.9 Y
Julio Jones Alabama 2011 6 220 6.66 Y
Greg Little North Carolina 2011 59 231 6.8 N
Junior Hemingway Michigan 2012 238 225 6.59 N

Draft position is only included for reference. It was not part of the analysis.

You can see that the cohort has a fairly high success rate (50 percent) as predicted by the regression tree. It also includes Vincent Jackson, who didn’t have early NFL success, but went on to have multiple WR2-or-better seasons.

I’ll emphasize, again, that most NFL combine measurements could be overvalued for wide receivers. I’m sure that it helps for wide receivers to have faster 40-yard dash times, and higher vertical jumps. But better measurables alone shouldn’t drastically alter your prospect assessments. Preliminary research I’ve done modeling wide receiver prospects post-draft shows that 40-yard dash times are likely overvalued, and Shawn Siegele found the same when he looked at hits and misses of previous drafts.

Remember to cut through the NFL combine noise and have a healthy dose of skepticism when you see wide receiver prospects rocket up draft boards without the strong collegiate production to back it up. We should be much less concerned that Laquon Treadwell has chosen to not run at the combine, than the suggestion that his collegiate production says he’s basically Donte Moncrief with more hype.