1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Is Ryan Tannehill Going to Become a Franchise QB?

Discussion in 'Miami Dolphins Forum' started by shouright, Jan 5, 2013.

  1. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    The fact that it has a correlation of 0.75 with career passer rtg does indicate that is predictive. Certainly it can't predict at 100% accuracy, but it does have strong predictive value.
     
    shouright likes this.
  2. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    But some variables covary. They vary together. That's what covariance is, and covariance is the basis of correlation.

    For example, height and weight covary. It just so happens that rookie WPA and career QB rating covary as well, and when two variables covary, you can use one to predict the other within a certain range and with a certain degree of confidence. That confidence is not 100%, but it's certainly far greater than zero.

    Again, if you tell me someone's height, I'll do a much better job of telling you his weight, again within a certain range and with a certain degree of confidence, than if you told me nothing about him at all.

    Likewise, I can do the same thing with regard to career QB rating if you tell me a quarterback's rookie WPA.

    The point of the thread is that Ryan Tannehill's rookie WPA puts him comfortably in the group of "franchise QBs" and well out of the group of "non-franchise QBs," based on his predicted career QB rating.
     
  3. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    Ok that makes sense to me. However, i've tried to collect my thoughts on this, because I realized I got lost in the argument. I'm going to try and explain my issues with this stat.

    75% is a strong correlation, but why isn't it stronger? I think partly because the stat doesn't account for the very basic concept of surrounding talent. A dropped pass is going to cost the QB the same whether the drop was the QB or WR's fault. That does not paint an accurate picture of the issue with that play. Things like that happen enough times, and we end up with a skewed stat. Considering this whole thread was born of another thread, that was born of another thread that was born of another thread that was born from a discussion about Tannehill's surrounding talent holding him back, I think that flaw is relevant.

    There has been no sufficient proof at this point, that Tannehill's ratings whether QB rating or WPA or Completions or TDs or Yards or ETC. are not impacted significantly by his surrounding talent and their lack of dynamic play making abilities. Since we haven't determined that, one way or the other, these predictive correlations aren't relevant.
     
  4. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    All this is very true. There is a reason why its not a 1.00 correlation. Part of that reason is exactly what you've mentioned (surrounding talent).

    That doesn't need to be determined for the correlations to be relevant. The correlation being 0.75 inherently acknowledges the shortcomings.

    But its also worth keeping in mind that with a correlation of 0.75, there is only so much that can impact it significantly. The stronger the correlation, the less we can assume its being impacted by other factors.
     
    shouright likes this.
  5. maynard

    maynard Who, whom?

    18,425
    6,346
    113
    Dec 5, 2007
    clearwater, fl
    Ok. A string of farts and an upcoming sh*t have a correlation of .75

    A shart has a correlation of 1.0

    Why are you making this so difficult?
     
    Perfectville_USA, Fin D and shouright like this.
  6. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    And sometimes the correlation is so strong that, as the saying goes, you "can't fart with confidence." ;)
     
    maynard likes this.
  7. Onehondo

    Onehondo Senior Member Club Member

    2,671
    879
    113
    Dec 20, 2007
    Chesapeake, Virginia
    You left out one, not with that running game, he's not.
     
  8. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    Just because you don't want to know what's causing the sharts (high mositure & low uniformity) or foops (low moisture & high chunkage) doesn't mean I'm making it difficult. It just means your life is so boring the only excitement you get is when there's a surprise coming out your ***. You nasty, nasty bastard.
     
    maynard likes this.
  9. cdz12250

    cdz12250 Season Ticket Holder Club Member

    10,265
    7,907
    113
    Nov 28, 2007
    Coconut Grove
    True. But our backs can run if we fix the offensive line so it opens holes.
     
  10. JMHPhin

    JMHPhin Season Ticket Holder Club Member

    7,684
    3,323
    113
    Nov 27, 2007
    Ohio

    But see any stat boils down to one thing, it doesnt think, it just computes. It deals in constants and variables, standard deviations, mean avgs etc. it is numbers. a person future ceiling can MAYBE be predicted with enough variables or even covariables, but it is still just that, a prediction because future varaiables or covariables are not known. In other words what is your input going to be for 2014?

    Your tolerable error for instance is subjective. it may be based on objective evidence but it is subjective, you may accept 5% whereas I accept on 3, someone else may accept 7. all based on the same objective data.
     
  11. Bpk

    Bpk Premium Member Luxury Box

    I was funny once. lol.

    But then I moved on to something that paid the bills. :)

    I'll post them in the lounge sometime for people who wanna **** the dog.
     
  12. Bpk

    Bpk Premium Member Luxury Box

    Is there a way to cross reference the two (WPA and rookie QB rating) and find an overlap that gives you a slightly higher than 0.75 correlation (which I consider the same as confidence) in predicting QBR?

    Intuitively it feels like that should be manageable, but I have't studied stats.
     
  13. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Correct. Nothing is perfect. Everything is limited.

    However, speculating about whether Tannehill will be successful based on your "eyesight" alone is a whole lot more limited IMO.
     
  14. maynard

    maynard Who, whom?

    18,425
    6,346
    113
    Dec 5, 2007
    clearwater, fl
    If I expand on this analogy, I would have to give myself points and possible time off.
     
    Fin D likes this.
  15. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Actually no, because like I said in the OP, when you control for WPA, the correlation between rookie QB rating and career QB rating drops from 0.58 to 0.19.

    You would think those variables could have an "additive" sort of relationship in predicting career QB rating like you're saying, but the truth is that when you introduce WPA "into the equation," it all but nullifies the correlation between rookie QB rating and career QB rating.

    Interesting, eh?

    It'd be about the same thing as looking at the correlation between height and weight, and then introducing the variable of number of calories eaten daily on average. That third variable is going to predict weight so much better than height that when you control for it, the correlation between height and weight will plummet.

    Make sense? This is what partial correlations can do. They can control for, or partial out, the impact of a third variable.

    I once did a study like this right after we got Daunte Culpepper and found that, when you controlled for the strength of the defenses of the teams he'd played on previously, the correlation between his performance and his teams' winning percentages was no different from that of any other QB. It wasn't like Culpepper was somehow "incapable" of winning (prior to his major injury, of course), which some were arguing at the time.
     
  16. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Here's a recent Palm Beach Post article that tries to make a case for the lack of a difference Matt Ryan would've made for the Dolphins:

    Matt Ryan's rookie QB rating may have been two points lower than Tannehill's, but Matt Ryan's rookie WPA, which as the original post shows is much more predictive than rookie QB rating of future QB rating, was 3.75, compared to Tannehill's 1.36. Matt Ryan's WPA was a whole 1.8 standard deviations above the mean (96th percentile) in the sample of rookie QBs since 2004, whereas Ryan Tannehill's was a mere 0.58 standard deviations above the mean (72nd percentile).

    And sure enough, the numbers bear that out: Matt Ryan's QB rating this year was a very good 99.1, good for 5th in the league overall.

    So it may be a way to "feel better about all of this," but it's using the wrong measurement for such mood enhancement, and therefore a ruse.

    http://www.palmbeachpost.com/news/s...dont-worry-too-much-about-what-matt-ry/nTwxr/
     
  17. CRAZYFACE

    CRAZYFACE Active Member

    230
    70
    28
    Sep 13, 2011
    England
    If you want to see how a QB should perform, watch that Colin K fella over at the 49ers. Always alert for any gaps that appear and is man enough to make a break for it when the opportunity arises to devastating effect. And fired up? The guy was sooooo pumped !!!! That is what I want to see from our QB. Not the dead eyed jog off the pitch when his last pathetic attempt fails. Are you listeniing Tannhill??????????
     
  18. Patssuck

    Patssuck Well-Known Member

    1,160
    432
    83
    Dec 2, 2012
    M.I.A
    What are you talking about?
     
  19. PhinishLine

    PhinishLine Season Ticket Holder Club Member

    4,276
    2,893
    113
    Dec 3, 2007
    Maryland
    All good until you have to go see Dr. Andrews.

    Sent from my Galaxy Nexus using Tapatalk 2
     
  20. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    Well, in all fairness, his face is crazy.
     
  21. Perfectville_USA

    Perfectville_USA Mr Perfect

    607
    302
    0
    Dec 27, 2012
    Syracuse NY

    I was so impressed watching the GB, to do that Vs a banged up defense. That has as many holes to fill, as our offense. What a great job! SF is loaded with weapons & has maybe the best O-line in football today. He out there in my Dynasty football league. But i'm not complety sold on the guy yet.
     
  22. byroan

    byroan Giggity Staff Member Administrator Luxury Box

    27,269
    44,475
    113
    Nov 25, 2007
    NC
    Tannehill ran just fine before the knee injury. Even ran fine once it healed. Did you see his 31 yard run against Buffalo?
     
  23. VManis

    VManis Season Ticket Holder Club Member

    5,753
    9,844
    113
    Nov 10, 2010
    If I understand this correctly WPA is not a measue of how well a player is playing but rather how much they contributed to the team's win. So the Dolphins success was tied more heavily to Tannehill's play than average and the Falcons was even more tied to Ryan's play. What that says to me is that rookies who are asked to go out an win games are more likely to end up being franchise QBs than those that are asked not to loose them.
     
  24. rafael

    rafael Well-Known Member

    27,364
    31,261
    113
    Apr 6, 2008
    If you look at RT's and CKap's rushing stats from after RT got his brace off and after CKap became a full time starter, they're pretty close. RT ran 28 times for 181 over those final six games while CKap ran 44 times for 304 yards over those final 8 games. RT's average was 4.67 carries for 30 yards (6.42 avg.). CKap's averages were 5.5 carries for 38 yards (6.91 avg.). I think that if Miami were to add a little zone read the numbers would be just about identical. CKap is faster, but otherwise they're very similar prospects.
     
  25. Serpico Jones

    Serpico Jones Well-Known Member

    4,697
    1,667
    113
    Feb 1, 2012
    Kaepernick is much faster and has a better arm.
     
  26. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Is there a difference? :headscratch:

    If that were true, then numbers of pass attempts per start, and percentage of pass attempts downfield would be predictive of (i.e., strongly correlated with) franchise QB status, but as the original post points out, they are not.
     
  27. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    Well, yes, a player can play amazing and the team can lose. A player can play horribly and the team can still win. This has been my problem with so much emphasis put on this stat.
     
  28. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    If the player plays amazing, he's still going to have a high WPA. Calvin Johnson and Vincent Jackson both were top-4 in WPA for WRs, but their teams weren't winning a lot.
     
    shouright likes this.
  29. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    Yeah, but the question at hand was about this statement:
    A player couldn't have contributed to a win if they didn't win.
     
  30. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Thats really just a matter of semantics. The player is contributing either way.
     
    shouright likes this.
  31. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    i.e., the player is contributing to his team's likelihood of winning, as measured by WPA (i.e., "Win Probability Added").
     
    Stringer Bell likes this.
  32. rafael

    rafael Well-Known Member

    27,364
    31,261
    113
    Apr 6, 2008
    I was very high on CKap coming out. In fact, I was one of the few here who was, but the difference isn't as great as you imply. Their arms strength is pretty close. CKap is faster, but RT is also faster than most QBs and more than fast enough to be a running threat. And despite RT's lesser college starting experience, he was more polished than CKap coming out and better at reading defenses. You're looking at CKap after having a year to adjust to the NFL and comparing it to RT's rookie season. As prospects they're pretty close.
     
  33. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,651
    67,546
    113
    Dec 20, 2007
    lololol
     
    Fin D likes this.
  34. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,651
    67,546
    113
    Dec 20, 2007
    maynard likes this.
  35. Saucy Puppet Show

    Saucy Puppet Show New Member

    6
    22
    0
    Apr 23, 2009
    Hi. I never really post here because I generally like to listen more than I like to talk. This topic is near and dear to my heart, however, and I had some free time this morning so I decided to perform a little analysis of my own based on the data from the original post. I built a logistic regression model using the other variables to predict the likelihood of each of the 5 new QBs becoming “franchise” (as defined in the post).

    There are a couple of caveats.

    First, there are not very many observations. I would love to have 100s and we have roughly 20. Secondly, the definition of franchise is based on QB rating which in turn uses several other variables. While the predictor variables are all from the first year while the definition of franchise is based on years 2+, there is still a correlation between first year performance and future performance. Hence, the predictive variables are not completely independent from the response variable making it a less than perfect scenario, nevertheless, it should be close enough for our purposes.

    Observation 1: Of the 10+ variables there were only 3 that had predictive power: WPA, Rookie QB Rating, and Pass Attempts per TD. Each of these had a p-value of less than .05, with WPA being the strongest by a nose.

    Code:
    WPA Model Coefficients:
                Estimate Std. Error z value Pr(>|z|)  
    (Intercept)   0.5253     0.5890   0.892   0.3724  
    wpa           1.1321     0.5321   2.128   0.0334 *
    
    Rookie QBR Model Coefficients:
                     Estimate Std. Error z value Pr(>|z|)  
    (Intercept)      -8.16123    3.94006  -2.071   0.0383 *
    rookie_qb_rating  0.12060    0.05761   2.093   0.0363 *
    
    Pass Attempts per TD Model Coefficients:
                    Estimate Std. Error z value Pr(>|z|)  
    (Intercept)      17.4022     8.3294   2.089   0.0367 *
    attempts_per_td  -0.5446     0.2634  -2.068   0.0386 *
    
    
    Observation 2: Due to the small sample size, attempts to build a multivariate model were unsuccessful. Modeling on more than one independent variable gave you no significant variables. Again, this is likely to due to the relatively small sample size.

    Observation 3: Ryan Tannehill is either very likely to succeed or very unlikely to succeed based on which model you choose. All the numbers in the table below are the likelihood the QB turns out to be a "franchise" QB. (To be more technically correct, it actually says that if you had 100 QBs with the same WPA as Ryan Tannehill, roughly 89 of those 100 would be franchise, but I digress...) Notice that Luck, Wilson, and RGIII have relatively stable scores under the models while RT and Weeden have drastically reduced chances of success under the PAPTD (Pass attempts per TD) Model.

    Code:
    QB	WPA	RQBR 	PAPTD
    Thill	88.75%	73.43%	[B]1.29%[/B]
    Luck	99.64%	74.36%	92.80%
    Wilson	98.12%	98.01%	99.99%
    RGIII	98.26%	98.51%	99.88%
    Weeden	49.83%	64.44%	6.24%
    
    So why this discrepancy? I think it's an interesting question and that the answer will vary based on your own personal bias. Empirically, I would suggest that this is the one category that having "playmaking" receiver(s) can make a big difference. Throwing 1 80 yard TD pass is "worth more" (according to this model) than 16 consecutive complete passes that result in a 80 yard drive for a TD. Having a receiver dive across the goal line is worth a lot more than having him stopped at the 1, especially if the ball is run in on the next play. Additionally, the more you have to throw the ball, the worse in general this metric will be, so a strong running game could have an influence. Conversely, having a QB who is good going down the field but can't convert will show up here as well.

    My personal conclusion from this exercise is that RT has an above average chance of success but I wouldn't buy tickets to Canton just yet. It also suggests to me (revealing my personal bias) that we need to get some support around Ryan before we can truly know his floor and ceiling.

    Hope you enjoy!
     
  36. VManis

    VManis Season Ticket Holder Club Member

    5,753
    9,844
    113
    Nov 10, 2010
    According to the guys that came up with WPA it is intended to be a moneyball type stat so yeah there is a difference. A good example would be B Marshall's 2011 season, he put up over 1,200 yards ranking him in the top 10 for wrs but his WPA was a more pedestrian 0.98. So Marshall played well but did not impact the outcome of the games.

     
    Fin D likes this.
  37. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,651
    67,546
    113
    Dec 20, 2007
    I wonder how more well rounded Tannehills game would of looked like if he had an offensive coordinator like Greg Roman/ Harbaugh if they utilized his running and zone read skills...There is no reason imo to not take advantage of getting Tannehills speed on the perimeter of a defense and let him read, run, and have an option to pitch..

    Unless we know he's not good at it or has the correct agility/coordination for it.

    Do You remember seeing anything in college that made you feel he has talent in the zone read game?
     
  38. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    If what you're saying is correct, then it also indicates that Tannehill's WPA & RQBR percentages could have been much higher and more in line with Luck, Wilson & RGIII.
     
  39. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    They've more or less admitted they limited Tannehill on purpose. Let's not forget, CKap is vet with more starting experience, then 17.
     
  40. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    I think the issue with what you did was that presumably the independent variables you used potentially are measuring the same thing. A model that is using TDs/att. as well as passer rating is going to suffer from over-fitting.
     
    shouright likes this.

Share This Page