Writers of Pro Football Prospectus 2008

02 Aug 2012

FO VIDEO: Playmaker Score 2012

Our latest video for the Sabermetrics Video Network is a look at updates we've made to our Playmaker Score system to project wide receivers into the NFL, along with the highest and lowest rookies from the 2012 draft class. This year, Playmaker Score is spelled J-E-T-S Jets Jets Jets.

Posted by: Vincent Verhei on 02 Aug 2012

13 comments, Last at 02 Apr 2013, 11:10am by Cathedralplumbing


by Danish Denver-Fan :: Thu, 08/02/2012 - 6:19pm

Great Vince - very interesting subject, and, taking the format into account, more importantly easy to follow en well communicated.

Allthough if you plan on doing this again, may i suggest you talking a bit slower - i struggled to keep up and digest all the information at times. (Full disclosure: I consider myself fluent in english allthough it is not my first language)

by Vincent Verhei :: Thu, 08/02/2012 - 10:05pm

Thanks. I've been doing podcasting for like five years, and I still have to struggle with speaking too quickly. Plenty of people who have spoken English all their lives (and American English at that) have trouble understanding me.

by akn :: Thu, 08/02/2012 - 8:36pm

0.49 is at best a moderate correlation (which as you noted is about as well as you can do when it comes to these kind of things). I wonder what the corresponding significance of that correlation was, however.

Also, I know you guys are on a budget, but how old was that intro music? It sounded like you waited for that jingle to makes its way into the public domain.

by Bigg Johnson :: Thu, 08/02/2012 - 9:35pm

.49 is a pretty strong correlation when it comes to this football stuff. im happy that stats can translate that high because it shows that more breakthrough statistics are coming through every day. great work vince!

by Danny Tuccitto :: Thu, 08/02/2012 - 10:16pm

for bivariate normal r = 0.49 and n = 149, p <.0001, so pretty dog-gone significant.

by Vincent Verhei :: Thu, 08/02/2012 - 10:17pm

Thank God Danny is here to handle all our maths.

by Danny Tuccitto :: Fri, 08/03/2012 - 3:22am

Technically, we can thank Aaron (who may be the same thing depending on one's level of idol worship).

by akn :: Fri, 08/03/2012 - 12:18am

So, the model is based on 149 sample points. If I understood the video correctly, your model is based on at least 4 categories (2 combine #'s, some measure of the type of offense the players were in, conference) and evaluated on yards/season in the pros. In addition, you implied that the model was refined after pruning several other categories (the other combine #'s, etc), presumably because they either a) produced undesirable results, or b) there wasn't enough power with so many categories (how many players coming from the WAC with a vertical between x1 and y1 and a 40 time between x2 and y2 and a Wonderlick between x3 and y3, etc.).

So what made you choose only these categories? What single number contributes most to that 0.49 correlation? In other words, what are your principle components? I'm curious because teasing out info like that may go a long way into determining what exactly makes a good prospect (in the yards/season sense), other than an abstract Playmaker Score. It's the next step when trying to tease out what a complex model actually means.

Perhaps you've done this in your almanac, and I simply have to buy a copy. But it would be nice to know if you went past the data/dimension reduction step and into the analysis/synthesis step of statistical modeling.

by Vincent Verhei :: Fri, 08/03/2012 - 1:48am

There is much more detail in the almanac, though honestly not all of your questions are answered. The quick answer(s) is, we checked correlation for all Combine data (including height, weight, and BMI) with NFL success, and found that the 10-yard split and vertical jump were the only useful predictors. We also checked correlations with single-season highs in receptions, yards, and touchdowns, adjusted for team pass attempts. We combined those five categories into one number, weighting those with the highest correlations most heavily, and called that product Playmaker.

by Ryan Wanger (not verified) :: Fri, 08/03/2012 - 3:24am

I guess it's correct to say that Jordy Nelson "hasn't done much" if you look average his career totals across four seasons. But as soon as he became the #2 receiver, he put up the 2nd best DVOA ever for a receiver.

by KK Probs (not verified) :: Fri, 08/03/2012 - 12:51pm

Along those lines, a question that I would pose is, at what point doesn't college pedigree matter? Or, at what point does professional coaching and physical development as a pro become the overriding factor? In a guy's rookie year, he hasn't had hardly any pro coaching, so his college stats and experience would be a big factor - not only all we have to go on, but a current reflection of where he is physically. When a guy's been in the league 4-5 years, I would think that what he's learned in the pros and how he's developed once he got there more than trump his situation when he first got out of college. This is the case, I think, with Jordy Nelson: it seems silly to expect less out of him because he went to a smaller college at this stage of his career.

The other factor along the same lines as coaching is learning the speed of the game. Every rookie says that's the biggest difference between college and the NFL. I suspect this is a much smaller rookie adjustment for an SEC player than it is for a small-conference player, hence they're more ready in their rookie year. Again, after a guy's survived the league for 4 years, no matter what his situation was in college, he's learned how to play at NFL speed, or he wouldn't still be playing.

So has the Playmaker Score been looked at for correlations in Year 1, Year 2, Year 3, etc of a player's NFL career?

by Vincent Verhei :: Fri, 08/03/2012 - 1:15pm

I think I said "until last season" somewhere in there. If not, I definitely should have.

by Cathedralplumbing (not verified) :: Tue, 04/02/2013 - 11:10am

i am impressed and i like to come again in future i found that the way of writing to clarifying that exactly want to say was very good so chatrandom