Looking at Wide Receivers Analytically

The Model

The next position I’ll take a look at are the wide receivers. I’ve done a lot of research into wide receivers in the past, and the data has shown to be pretty good for analysis. I decided the best course of action would be to create a multiple linear regression using different kinds of production, measurable, and age related metrics. I collected data from the 2006 to 2012 draft classes, and will use data from the 2013 and 2014 draft classes to test the model. Like what I used to evaluate Quarterbacks in my previous post, I will use Adj CarAV as the dependent variable. This is a holistic measure of performance that can be compared across positions. To break down what variables are important, I’ll post the output that R gives me:

Coefficients:
              Estimate Std. Error t value Pr(>|t|)   
(Intercept)    39.000680 15.101783 2.583   0.01127 *
YDPTA        0.541357   0.429871    1.259   0.21087   
TDPTA         83.359293   36.143256    2.306   0.02318 *
Rush.Attempts -0.039080    0.034835   -1.122   0.26463   
Rush.Yards     0.005702    0.004703    1.212   0.22826   
Rush.TD         0.338539    0.180853    1.872   0.06417 .
P.Yds           0.001526    0.000912    1.673   0.09740 .
BO.Age         -0.336226    0.214842   -1.5 65 0.12078   
Draft.Pick     -0.009767    0.003627   -2.693   0.00833 **
Weight          0.026155    0.019290    1.356   0.17822   
X40.time       -4.339576    2.418242   -1.795   0.07578 .
X3.cone        -2.506422    1.359026   -1.844 0.06813 .

You can ignore the first line for the most part, that is just the y intercept for the model. The second and third lines are Yards Per Team Attempt (YDPTA) and Touchdowns Per Team Attempt (TDPTA). These stats are similar to other metrics such as market share, which adjust wide receiver stats compared to the offense they are in. This helps put era and scheme into perspective and give a more predictive metric. Both of these are positive correlations, with touchdowns being more significant.

The next three lines are all based on a prospect’s career rushing production. Yes, rushing production matters for a wide receiver. Some of the more prolific “running” wide receivers that were included in the model are Randall Cobb, Jeremy Maclin, and TY Hilton (Percy Harvin would also be up there, but he did not perform agility drills predraft). Raw yards and touchdowns are positive values, and are negatively adjusted based on rushing attempts.

The next line is a prospect’s career punt return yards. Special teams production seems to be one of the more under the radar traits a prospect can have, and found a place in the model. Many later round guys that went on to have great NFL production such as Antonio Brown, Emmanuel Sanders, and TY Hilton had great special teams production.

The next line is breakout age, which is the age a prospect is at when they first record a market share of yards (prospect’s receiving yards divided by the whole team’s receiving yards) that is over 35%. Typically, college players that breakout early end up playing better in the NFL.

The remaining metrics are draft position, weight, 40 time, and 3 cone drills. Each of these are as you would expect, higher draft players perform better, heavier prospects perform better, and faster and more agile players perform better.

This is a plot of the model’s projected Adj CarAV against the actual results. The r^2 is 0.44, meaning that the model is able to explain about 44% of the variation in AV. To put this in comparison, draft position alone is able to explain 23% of the variation in AV. This means that the model is almost twice as good at explaining the variation in performance than just the draft is. However, this is just using data that the model is already aware of. In order to really test the model, we will now use new data to see if the model is still predictive of player performance.

As you can see, the r^2 of this model holds to a number of 0.4381, almost the same as the training data.

The 2017 Draft Class

When I did my QB model, I noticed that a lot of you had questions about what these analyses meant for the 2017 class. While it’s early, we don’t know official weights, 40’s, 3 cones, and draft positions for these players, I’ll give a ranking of the players using estimates of this data. Weights and 40 times will come from nfldraftscout.com, and the projected draft position will come from CBS’ prospect rankings. If you have a question on some of the guys that I didn’t mention, just post and I’ll run them through.

1. Corey Davis, Western Michigan

Maybe somewhat surprisingly, Corey Davis is the number 1 projected WR in the 2017 draft. Corey has a history of high production at an extremely young age, which is a good sign that a player will continue that success in the NFL. Corey is also a big bodied guy who projects to have a decent combine.

2. Dede Westbrook, Oklahoma

Westbrook clocked in at the most yards per team attempt in the class, just edging out Corey Davis with 4.00. What brought Westbrook back is that fact that he broke out at age 23 (almost 5 years older than Davis’ breakout age) and that he is a much smaller guy. Still, Westbrook projected to be a very nice pick in his projected high 2nd round area.

3. John Ross, Washington

Maybe more than any other prospect, the combine will matter for John Ross. NFLdraftscout.com projects him to have the fastest 40 time of the class, at 4.35. His receiving production was solid, not spectacular, but he did show some ability in the run game. I expect him to be a fairly polarizing prospect, but should be a justifiable selection at the end of the first round.

4. Taywan Taylor, Western Kentucky

Taywan is the first big sleeper that the model has identified in this year’s class. Projected to go at the top of the 4th, Taylor has the production you would see out of a potential first rounder. Additionally, he is projected to a be at least an above average athlete at the combine. He should be a guy to keep an eye out this week at the Senior Bowl.

5. JuJu Smith-Schuster, USC

Once thought to be the top WR in this class, Smith-Schuster had a disappointing season that has caused some to question his game. It also doesn’t help that JuJu’s former teammate Nelson Agholor is having struggles transitioning to the next level. Still, JuJu showed at a young age in college that he has the ability to dominate when firing on all cylinders. He only managed a below average YDPTA (1.99) this season, so a team selecting him will be banking on him returning to the player he once was. While he isn’t a perfect prospect, a 2nd round selection wouldn’t be the worst thing in the world for a team that believes in him.

8. Mike Williams, Clemson

I expect to get some backlash for this. I have the consensus number 1 WR, potential top 5 selection, Mike Williams projected to be worth about a 3rd rounder. Let me explain myself. The thing you need to keep in perspective about Williams is that he played in an abnormally high volume passing attack. For some perspective, Clemson had a total of 626 passing attempts as a team this year. Out of all the draftable FBS WR prospects I have listed, that is by far the most of any player in the draft. The next highest, Zay Jones of ECU, saw 553 passes, about a 12% decrease from 626. In that offense, Williams managed 2.17 YDPTA and 0.0176 TDPYA. The draft class averages for those numbers are 2.49 and 0.0233 respectively. Williams also did not get any production through rushing or special teams throughout his college career. From all angles, Williams was not a good college producer. Since 2006, the only receiver to go in the first round and have production of that level was Darrius Heyward-Bey and Jonathan Baldwin. If you expand that to the first two rounds, you can add Arrelious Benn, Mohamed Massaqoui, Derrick Williams, and Greg Little to that list.

When it’s all said and done, I think that Mike Williams ends up as a decent big bodied possession guy, and not much more than that. I think that a better Jonathan Baldwin is a fairly decent comp for what to expect from Williams in the NFL. To me that is worth a mid round pick. But certainly not a selection in the first round, let alone in the top 10.

The Wildcards

There are a few more players from the 2017 class that I have decided to exclude for the time being due to positional questions. If the NFL deems them to be able to play WR at the next level, they could be great players.

Curtis Samuel, Ohio State

If there is ever a player that is in the Percy Harvin mold, it is Curtis Samuel. Playing with Urban Meyer as a gadget player, Samuel has had heavy production on both the ground and the air. If evaluated as a receiver, since 2006 only Randall Cobb has ever produced more on the ground than Samuel. If I were to include Samuel, he would surpass Corey Davis as the number 1 WR in the class.

Billy Brown, Shepherd

Billy Brown is a huge man. Measuring in at 6’3” 255 lbs, Billy Brown has the size to be an NFL WR… maybe too much size. There has been a lot of talk about converting him to TE because of his sheer size, and his inability to separate at the division II level. His production is off the charts. His touchdown rate is right up there as the best since 2006, a spot owned by Dez Bryant. However, keep in mind that this was all done at a much lower level of football. The model doesn’t currently have any FCS or lower players in it. So keep all of this with a grain of salt. However, also don’t be surprised that if he comes in and impresses at the combine, that he ends up carving out a potentially great career in the NFL. To me, Brown seems like the classic late round, boom or bust lottery ticket. If I were to include him, he would be 3rd on the list, in between Westbrook and Ross, despite his projected 7th round draft slot.