Thursday, May 2, 2013

A statistical approach to drafting wide receivers

I thought I would post up a rather simplified example of what would happen if you asked a spreadsheet to analyze wide receiver prospects in the NFL Draft.  This isn't exactly how I go about doing things, but is close enough to give some idea as to how it can improve a team's odds of making a decent pick.  For this brief look, I will show what the computer thought of the draft classes from 2004 to 2012.

Player's will be given a score based in equal parts on their combine data, and their production in college.  Their Stat Score makes a simple adjustment to their raw college stats to normalize things for the purpose of comparing guys who played in different offenses.  The Stat Score is fairly similar to Shawn Siegele's Dominator Rating,  though I use their college team's total offense rather than just their passing offense.  I'm not claiming that this is an improvement over the Dominator Rating, it probably isn't, it is just the way I have historically done things. The other half of their score will be the Athletic Score, which will come from their combine results.

Since there is obviously going to be a difference between how a guy like DeSean Jackson succeeds, compared to a player like Calvin Johnson, players will be divided into two groups.  The "Big" group will consist of player's over 200 pounds, and their Athletic Score will put more of an emphasis on their Kangaroo Score, though the other combine drills will still be a factor.  These player's are expected to succeed by physically overpowering their opponent.  The "Small" group will consist of player's under 210 pounds, and their Athletic Score will put more of an emphasis their agility score (based on the short shuttle and 3-cone drill), as well as raw speed.  Basically, if you can't overpower your opponent, you want to be able to evade them.  Player's who are between 200 and 210 pounds will be graded on both scales to see where they fit best.

One issue that occurs, is you sometimes have a player who so thoroughly dominates either his Athletic Score or his Stat Score, that it can bury a failure somewhere else.  Take Troy Edwards, the 13th pick of the 1999 draft as an example.  His Stat Score was 2.160 standard deviations above average, which is shockingly good.  Unfortunately his Athletic Score was -0.630 standard deviations below average.  Since I am looking for people who have a physical advantage over their opponent, and a history of meeting this potential, the computer will cut any prospect from consideration who wasn't at least average in both areas.

There is an additional issue.  I'm still not sure whether to adjust the Stat Score for players who competed at the Division II or III level.  Making such an adjustment would be easy, but I just haven't decided yet how much of a deduction to make.

So, here are the top 5 results from each year, including their final score, where they were selected, and whether they were graded as a Big or Small receiver:

Derek Carrier Undrafted 1.606 Big
Michael Floyd  pick #13   0.717 Big
Marvin McNutt pick #194 0.660 Big
Justin Blackmon  pick #5    0.557  Big
Alshon Jeffrey pick #45   0.483 Big                              

Derek Carrier is a bit of an oddball, coming from tiny Beloit College.  It will be interesting to see if he becomes anything , or if I will have to start creating a penalty for players from such low levels of competition.

Jonathan Baldwin  pick #26   1.098 Big   Next stop, Bustville!  
Julio Jones pick #6   0.814  Big
Torrey Smith pick #58   0.643  Small           
Stephen Burton  pick #236   0.513 Big 
Cecil Shorts pick #114   0.484 Small

Dez Bryant pick #24   0.998 Big
Mike Williams pick #101   0.690  Big
Victor Cruz  Undrafted    0.685  Big or 0.590 Small(listing both because it’s interesting to me)
Andre Roberts pick #88    0.640 Small
Emmanuel Sanders pick #82   0.588 Small           

There was no combine data for Eric Decker, Demaryius Thomas, or Danario Alexander.  If there had been I suspect they would have made the list.

Kenny Britt pick #30  0.807 Big
Hakeem Nicks pick #29   0.720  Big
Ramses Barden  pick #85   0.705  Big
Mike Thomas pick #107    0.503 Small
Mike Wallace  pick #84    0.315  Small

No combine for Michael Crabtree

James Hardy pick #41   0.413  Big  Ooops! That didn't turn out well. 
Jordy Nelson  pick #36   0.411  Big
Pierre Garcon  pick #205   0.388  Big
Donnie Avery pick #33   0.363  Small
Earl Bennett pick #70   0.169   Small             

This was just a terrible year for receivers.

Calvin Johnson  pick #2   1.881   Big
Robert Meachem  pick #27   0.661   Big  One of the more disappointing players.  
Mike Sims-Walker  pick #79   0.490  Big
Dwayne Bowe  pick #23   0.435   Big
Laurent Robinson   pick #75   0.381   Big

Miles Austin Undrafted   0.851  Big
Greg Jennings pick #52   0.750  Small
Brandon Marshall  pick #119   0.640  Big
Marques Colston  pick #252    0.608  Big
Derek Hagan  pick #82   0.599  Big                           

A banner year for the computer.  Fortunately the computer is a humble guy, and doesn’t  make much fuss about it.

Vincent Jackson  pick #61    2.019  Big
Dante Ridgeway pick #192   0.990  Big  Not one of the computer’s finer moments
Mike Williams pick #10   0.988 (now known as “the fat Mike Williams”) Big
Roddy White pick #27   0.815  Small
Braylon Edwards pick #3   0.658  Big

That’s not a typo.  Vincent Jackson is indeed 2 standard deviations above average. 

Larry Fitzgerald pick #3   1.014  Big
Lee Evans pick #13  0.855  Small
Reggie Williams pick #9   0.797  Big
Rashaun Woods pick #30  0.605  Big  Honestly, I'm not sure what happened with this guy.
Jerricho Cotchery pick #108   0.588  Big 

As I said before, this is just a very basic way of doing things, and not what I would really recommend.  Trying to boil things down to one overall score just doesn't work as well as looking over a broader set of smaller scores.  Using a broader set of data lets you get more of a sense as to how balanced a player is in a wide range of areas.  

Still, while the computer does make some mistakes here (Jon Baldwin, Mike "The fat one" Williams, James Hardy, etc.) many of these mistakes are no worse than what actual NFL teams did.  The spreadsheet also scored major coups in selecting players like Victor Cruz, Marques Colston, Brandon Marshall, Miles Austin, and Mike "Not the fat one" Williams, as well as others, all in the later rounds or undrafted.  Overall, out of the 40 listed prospects (not counting the 2012 draft class, since it is too early for that), 67.5% arguably became successes according to my odd definition of the term.  That is in comparison to a overall average success rate of 24.39% for all drafted receivers, or a 22.5% median league-wide success rate for individual NFL teams.  Some players like Emmanuel Sanders also appear poised to enter the "success" list as their playing time increases, but I'll leave that alone for now.

I'm not saying that players should be graded in such a simple manner.  I'm just saying that even a method this ridiculously simple should outperform most NFL GMs, and that a more sophisticated version, along with some limited film study, should produce excellent results.  I'll get into exploring some of the ways to refine things even further here.

No comments:

Post a Comment