QB Wonderlic Scores 2018

[Figster]

2 hours ago, iinii said:

So you would take Fitzpatrick over Kelly?

No

[mramefa]

I've been hearing people discuss pros and cons of wonderlic forever,  usually giving data that can only be interpreted anecdotally. I was curious, so I took a few minutes, and compiled a scatter chart of wonderlic scores vs career passer rating (for QBs with over 1500 career passing attempts from Source: https://www.pro-football-reference.com/leaders/pass_rating_career.htm).

The results are pretty straightforward. Wonderlic has a correlation of 0.31 and an r-squared of 0.10 with passer rating. That means 10% of the variation in career passing rating can be explained by the variation in wonderlic scores. 10% of success coming from a single metric makes it very important, regardless of the exceptions you might hear about. The chart also shows pretty clearly that wonderlic is not just a good indicator of success below a certain range (i.e. quarterbacks below a score of 16 will not succeed). A higher wonderlic score is associated with a better passer rating regardless of how high the score is (with Fitzpatrick being a slight outlier on the lower side, otherwise the relationship would be even stronger).
 

Here is the same chart with Fitzpatrick removed. The r-squared jumps to 0.15.

 

 

It's pretty clear that teams should be looking at wonderlic when they evaluate quarterbacks, and it's a very good sign that Josh Allen has a strong score.

 

Edited March 31, 2019 by nrenegar

[YoloinOhio]

11 minutes ago, nrenegar said:

I've been hearing people discuss pros and cons of wonderlic forever,  usually giving data that can only be interpreted anecdotally. I was curious, so I took a few minutes, and compiled a scatter chart of wonderlic scores vs career passer rating (for QBs with over 1500 career passing attempts from Source: https://www.pro-football-reference.com/leaders/pass_rating_career.htm).

The results are pretty straightforward. Wonderlic has a correlation of 0.31 and an r-squared of 0.10 with passer rating. That means 10% of the variation in career passing rating can be explained by the variation in wonderlic scores. 10% of success coming from a single metric makes it very important, regardless of the exceptions you might hear about. The chart also shows pretty clearly that wonderlic is not just a good indicator of success below a certain range (i.e. quarterbacks below a score of 16 will not succeed). A higher wonderlic score is associated with a better passer rating regardless of how high the score is (with Fitzpatrick being a slight outlier on the lower side, otherwise the relationship would be even stronger).
 

Here is the same chart with Fitzpatrick removed. The r-squared jumps to 0.15.

 

 

It's pretty clear that teams should be looking at wonderlic when they evaluate quarterbacks, and it's a very good sign that Josh Allen has a strong score.

 

Who invited the smart guy??

 

thanks for this 

[Gugny]

1 hour ago, YoloinOhio said:

Who invited the smart guy??

 

thanks for this 

 

This guy makes @DC Tom look like @Cripple Creek.

[NoSaint]

1 hour ago, nrenegar said:

I've been hearing people discuss pros and cons of wonderlic forever,  usually giving data that can only be interpreted anecdotally. I was curious, so I took a few minutes, and compiled a scatter chart of wonderlic scores vs career passer rating (for QBs with over 1500 career passing attempts from Source: https://www.pro-football-reference.com/leaders/pass_rating_career.htm).

The results are pretty straightforward. Wonderlic has a correlation of 0.31 and an r-squared of 0.10 with passer rating. That means 10% of the variation in career passing rating can be explained by the variation in wonderlic scores. 10% of success coming from a single metric makes it very important, regardless of the exceptions you might hear about. The chart also shows pretty clearly that wonderlic is not just a good indicator of success below a certain range (i.e. quarterbacks below a score of 16 will not succeed). A higher wonderlic score is associated with a better passer rating regardless of how high the score is (with Fitzpatrick being a slight outlier on the lower side, otherwise the relationship would be even stronger).
 

Here is the same chart with Fitzpatrick removed. The r-squared jumps to 0.15.

 

 

It's pretty clear that teams should be looking at wonderlic when they evaluate quarterbacks, and it's a very good sign that Josh Allen has a strong score.

 

 

Thanks for doing this! 

 

If if you cut the old guys what does it look like for the modern Nfl?

[DC Tom]

36 minutes ago, Gugny said:

 

This guy makes @DC Tom look like @Cripple Creek.

 

Except that I know a correlation of 0.31 sucks.  Sometimes linear regression is not the proper analytical tool.  

 

What's telling is that no one with a Wonderlic lower than 27 has a passer rating above 95.  You want to draft a great QB, don't look at anyone with a Wonderlic below 30, and your chances of getting one increase significantly.

[KD in CA]

Huh....surprised no one discussed Jackson's 13 before the draft last year.

[mramefa]

5 hours ago, DC Tom said:

 

Except that I know a correlation of 0.31 sucks.  Sometimes linear regression is not the proper analytical tool.  

 

What's telling is that no one with a Wonderlic lower than 27 has a passer rating above 95.  You want to draft a great QB, don't look at anyone with a Wonderlic below 30, and your chances of getting one increase significantly.

My take is different. I think 10-15% of variation in passer rating being explained by a single combine measurable is actually a lot. It's not the correlation for a complicated model, just a single measurable. Keep in mind that to some extent passer rating is also a function of things outside of a quarterback's control (offensive scheme, receiver and oline talent). I think teams would be foolish to ignore that measurement as they try and predict prospect success, and I'd be surprised if you could find a single combine measurable that's a better predictor of QB success (I'd guess velocity is also important, although less so).

 

I also looked at nonlinear relationships (e.g. exponential), but the best fit curves were still approximately linear. You can confirm that a linear fit is reasonable from the chart visual.

 

[Pine Barrens Mafia]

6 hours ago, KD in CA said:

Huh....surprised no one discussed Jackson's 13 before the draft last year.

 

Welcome back, Rip Van Winkle.

 

[DC Tom]

7 hours ago, nrenegar said:

I also looked at nonlinear relationships (e.g. exponential), but the best fit curves were still approximately linear. You can confirm that a linear fit is reasonable from the chart visual.

 

 

Actually, you can't confirm that, which is pretty much the whole point.  It's widely scattered date, the variance is much higher than the mean.  You also assume the Wonderlic score is an independent variable - which is a necessary assumption for a linear regression, but is nonetheless invalid: your chosen analysis requires it to be, so you treat it as such.  But based on what?  Do you see any reliable evidence of a normal distribution in that data?  Did you analyse it for the best-fit distribution, or just eyeball it?  

 

And that's all beside the fact that a 0.31 correlation is a weak correlation.  Pretending that's meaningful is the sort of piss-poor analysis that leads to space shuttles exploding - literally, you just duplicated the basic flaws in NASA's analysis of O-ring blow-by, that there was some sort of linear dependence on temperature.  Their incorrect statistical model masked the simple observation that all blow-by occurred below a certain critical temperature.  Much like your analysis: trying to pretend a weak linear correlation on non-linear data of uncertain statistical distribution masks the very simple observation that no QB with a score under 27 has a rating of 95 or higher.

[SouthNYfan]

6 minutes ago, DC Tom said:

 

Actually, you can't confirm that, which is pretty much the whole point.  It's widely scattered date, the variance is much higher than the mean.  You also assume the Wonderlic score is an independent variable - which is a necessary assumption for a linear regression, but is nonetheless invalid: your chosen analysis requires it to be, so you treat it as such.  But based on what?  Do you see any reliable evidence of a normal distribution in that data?  Did you analyse it for the best-fit distribution, or just eyeball it?  

 

And that's all beside the fact that a 0.31 correlation is a weak correlation.  Pretending that's meaningful is the sort of piss-poor analysis that leads to space shuttles exploding - literally, you just duplicated the basic flaws in NASA's analysis of O-ring blow-by, that there was some sort of linear dependence on temperature.  Their incorrect statistical model masked the simple observation that all blow-by occurred below a certain critical temperature.  Much like your analysis: trying to pretend a weak linear correlation on non-linear data of uncertain statistical distribution masks the very simple observation that no QB with a score under 27 has a rating of 95 or higher.

 

So what you're saying is that you are smarter than NASA rocket scientists??

 

[DC Tom]

5 minutes ago, SouthNYfan said:

 

So what you're saying is that you are smarter than NASA rocket scientists??

 

 

No.  I'm smarter than NASA lawyers.

[SouthNYfan]

4 minutes ago, DC Tom said:

 

No.  I'm smarter than NASA lawyers.

 

Riiiiiiight.

[DC Tom]

1 minute ago, SouthNYfan said:

 

Riiiiiiight.

 

I'm guessing...you're a NASA lawyer, then?

 

Simply based on the fact that I'm clearly smarter than you, too.

[SouthNYfan]

5 minutes ago, DC Tom said:

 

I'm guessing...you're a NASA lawyer, then?

 

Simply based on the fact that I'm clearly smarter than you, too.

 

Right then.

I should have known better to respond to your pompous a**.

[K-9]

15 minutes ago, DC Tom said:

 

No.  I'm smarter than NASA lawyers.

Not exactly a high bar, that. ?

[DC Tom]

Just now, K-9 said:

Not exactly a high bar, that. ?

 

It truly isn't.  People don't remember that the rocket scientists told the lawyers "No, this is high-risk," and the lawyers said "No, it's okay, just watch."

 

And it's worse today, with the "I'm a statistician, I have Excel!" crowd.  The "statisticians" at work stopped sending me emails after I shredded their work one too many times.  I consider that a sufficient result.

[Tcali]

unless your wonderlic score is under 10 the wonderlic score means absolutely nothing regarding football smarts.--Based on the list of names above.

we all knew that Jim Kelly and Dan Marino werent taking AP chemistry and math in high school...But they were very smart football players.

[K-9]

8 minutes ago, DC Tom said:

 

It truly isn't.  People don't remember that the rocket scientists told the lawyers "No, this is high-risk," and the lawyers said "No, it's okay, just watch."

 

And it's worse today, with the "I'm a statistician, I have Excel!" crowd.  The "statisticians" at work stopped sending me emails after I shredded their work one too many times.  I consider that a sufficient result.

Those NASA lawyers forgot who their customers were.

[Tcali]

14 hours ago, DC Tom said:

 

Except that I know a correlation of 0.31 sucks.  Sometimes linear regression is not the proper analytical tool.  

 

What's telling is that no one with a Wonderlic lower than 27 has a passer rating above 95.  You want to draft a great QB, don't look at anyone with a Wonderlic below 30, and your chances of getting one increase significantly.

Even better...dont look at the wonderlic at all.And look at the game film or scout them live...And see if they make smart decisions on the field.