Predictive Criteria for QBs from 2011 Draft

[zonabb]

The draft behavior of teams is fascinating because it defies logic in a lot of cases and it troublesome because it's intended to be predictive, that is making some assessment on future results based on past results. Its like stocks, except you also have to predict the change in talent-level and ascertain whether the player who dominated at the NCAA has the ability to do so at the NFL level. So the process is fraught with problems. But the QB position is one I've been even more fascinated with over the years, mainly as I've watched the Bills bungle one QB pick after another.

 

I think there is one way that could be used (and I hope teams are) to try to better inform the argument over what qualities are important in a college QB to predict performance at the NFL level. That's using multiple regression analysis, which means you try to predict the change in a dependent variable, in this case NFL regular season wins by a QB, based on a handful of indepenent college and personal variables thought to be predictive, like college wins, Wonderlic score, completion percentage, etc. Since regression is a predicative model, it'll explain how much impact each variable has and also, more importantly determine which ones have significant impact and which don't. What it won't do, and this is important, is tell you with 100% certain all the qualities that matter, which are probably the system of offense the're in in the NFL, the quality of the offense/players on their NFL, coaching, mental make-up, combine measurables, etc. But it's the best option to narrow down a QB based on his college stats and personal attributes, like "toughness" for example which is hard to quantify.

 

But what I find the most interesting and often controversial is the Wonderlic score. Those that are "manly" (in terms of QBs) think that the score is meaningless because it's a game of athelticism, toughness and brawn. I don't fall into that category, I think its very important for QBs and less so for say WR and RBs. Here's why I think its important, because after looking for all the Wonderlics I could and the scores for Supoer Bowl winning QBs, I found that all SB winners with reported or leaked scores, with the exception or one or two (Bradshaw for sure but I think he was dyslexic and that might explain the score?), scored at least a 22 or 23 (don't have that info accesible). What does that mean? Well, I bet if you ran a resgression with SB wins as the dependent variable, Wonderlic would be signifcantly important. It doesn't mean a low scoreing QB won't have a lot of NFL wins, because there are lots that have, but it does mean they don't win a lot of Super Bowls. So for me, given the probability of a guy with a 15 on a Wonderlic winning a Super Bowl so low, I'd avoid drafting him, unless I needed a backup.

 

As I said, the regression will have some errors and contradictions due to the unknown variables but it'll still be important. A perfect example would be Losman and Edwards. Losman was highly regarded as an "athlete" with an arm and the protypical NFL skillset, except between the ears. He scored a 14 on his first Wonderlic but took it again and scored a 31 (the biggest joke, since the players then have seen a test and are highly coached). Edwards was not as physically gifted but scored a 34 on his Wonderlic. They both sucked. I think we all agree that Losman was bad a decision-making and even with his skills, he didn't have the "smarts." Edwards had the smarts but then got skittish and flighty after taking a beating and didn't display what we might call "toughness."

 

But what these two show is that drafting a QB is tough due to the myriad of variables that impact a QBs skills and abilities. Ultimately, regression would still only be a tool in analyzing QBs, but it would allow for a ranking of variables that are high correlated to NFL regular season and Super Bowl wins.

 

Why don't I do it? The collection and aggragation of the data would take way too long but I would gladly accept it if anyone for some crazy reason! I could run the regression with the right data and have some results in 10 minutes but since I'm supposed to be working on my dissertation right now and not wasting time here, I can't!