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

Ryan Tannehill

Discussion in 'Other NFL' started by bbqpitlover, Oct 16, 2019.

Ryan Tannehill is...

  1. A terrible QB

    0 vote(s)
    0.0%
  2. A below average QB

    4 vote(s)
    5.7%
  3. An average QB

    7 vote(s)
    10.0%
  4. An above average QB

    39 vote(s)
    55.7%
  5. An elite QB

    16 vote(s)
    22.9%
  6. The GOAT.

    4 vote(s)
    5.7%
  1. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    Here are Tannehill's z-scores season to season, in terms of passer rating adjusted to 2019:

    2012: -1.07
    2013: -0.57
    2014: 0.19
    2015: -0.30
    2016: 0.22
    2018: -0.18
    2019: 2.29

    So you can see there that prior to 2019, he had never been more than 0.22 standard deviations above the league average.

    If we toss aside his rookie season, his average z-score from 2012 to 2018 was -0.13, with a standard deviation of 0.30.

    So I'll say he'll finish within two standard deviations of that figure (-0.13), which would place the upper limit 0.47 standard deviations above the league average in 2020, and well below the 2.29 he posted in 2019.

    In terms of recent performers, that would put him in the neighborhood of Deshaun Watson in 2019 (0.52), Kirk Cousins in 2017 (0.49), Ben Roethlisberger in 2017 (0.45), and Kirk Cousins in 2018 (0.43). And that's if he's at the top of that range.
     
  2. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Well I don't know.. I still don't trust ESPN's proprietary "clutch" weighting, but looking at the list:
    https://www.espn.com/nfl/qbr/_/sort/cwepaTotal/dir/desc

    One thing seems to stand out: EPA doesn't seem to be EPA per game. Instead it seems to be total EPA (in fact they explicitly define it that way when you scroll over the "EPA" column label). That would totally explain why Tannehill is low. You can be highly efficient but if you don't play many games your total EPA is low. It also explains Prescott: a bit above average but played all games. Also explains a lot at the bottom of the list: either they didn't play many games OR they played many but were terrible.

    So that's my first guess in which case you really shouldn't use "total EPA" to compare QB's within a season because it's mixing efficiency with volume. Where you'd want to do that is for a career because wins added does depend on volume.

    Regardless.. I don't trust the measure as much as you're willing to. I've seen too many stupid things done with proprietary stuff that later came out, and I'm certainly not giving ESPN the benefit of the doubt. But that's my guess right now.
     
    The Guy likes this.
  3. Etrius24

    Etrius24 Well-Known Member

    682
    685
    93
    Mar 4, 2020
    Tannehill just posted the 4th highest QB in the history of the NFL.... I am not expecting him to do that **** again...
     
    mooseguts and resnor like this.
  4. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    No I'm using EPA per play here. That's where Tannehill is 3+ standard deviations from the league norm from 2017 to 2019 in terms of the difference between his EPA per play z-score and passer rating z-score, for both 2018 (when his passer rating was low) and 2019 (when his passer rating was high). And these are measures (EPA per play and adjusted passer rating) that correlate at 0.79 across the league.
     
  5. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Remember when we compared z-scores and they were off by a bit and I said it's probably due to you using Excel vs. me using MATLAB or R even though R agrees with MATLAB? Well your z-scores are a bit too far off for comfort so I went ahead and tried Excel out (which just reinforced why I don't use it lol).

    You're doing the calculations wrong in Excel, though not by much so I don't know precisely what the issue is. However, one thing you say in your posts might be a clue: you said you adjusted to a common year and then calculated z-scores. Not that I think that should have changed the calculation but you don't need to adjust ratings when calculating z-scores (that's the whole point of z-scores actually) because z-scores are already on the same scale.

    Let's go through this step by step so there's no miscommunication. First, download the Excel file from here:
    https://www.pro-football-reference.com/years/2019/index.htm#all_passing

    You want team passer ratings, not individual ratings, even for comparing individual ratings because otherwise you'll have biases due to passing attempts. In that Excel file, all the team passer ratings are in column Q. So, click on some empty cell way to the right (e.g., AA2) and type in this formula (the 117.5 in the formula is Tannehill's 2019 rating):

    = (117.5 - AVERAGE(Q2:Q33))/STDEV(Q2:Q33)

    Type that in, hit enter and you'll get 2.552824, not your 2.29. MATLAB shows 2.552823943068400. In other words identical.

    And for 2018 I'm getting -0.00235 from Excel and -0.002345967554519 from MATLAB. Also identical.

    So your calculations are wrong somewhere.
     
    The Guy likes this.
  6. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    If that's true (haven't verified it), that's a bigger mystery. Most obvious way I can think of to get a discrepancy like that is through TD% because EPA per play won't give you anywhere near as much credit for that as passer rating, especially for short TD passes.

    Superficially at least, it might fit with Tannehill since the one stat he was really high on in 2018 was TD% (came in 6th) and of course he was 2nd in 2019. Not sure.. but of the 4 components of passer rating that's the most obvious source of potential discrepancy.
     
    The Guy likes this.
  7. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    Please post the data you are using. Where are you getting EPA per play from?
     
  8. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    Why can't you just answer a simple question?

    If he drops to 100, is that "significantly worse"?
     
    resnor likes this.
  9. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    We actually don’t know that until we see what the league does in 2020.
     
  10. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    On this page:

    https://www.espn.com/nfl/qbr

    And you simply divide EPA by PLAYS. And then of course previous years are available there as well.
     
  11. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    I think you touched on it where I was adjusting to 2019 and then calculating the z-score.
     
  12. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    OK.. the only reason it seemed weird is because 2019 is wrong. No reason to adjust 2019 to 2019. Anyway, just fix it and it's fine.
     
  13. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    OK I think the issue is that the z-scores are derived from adjusting to 2019 and then determining the deviation from the average of my sample of 93 QBs from 2017 to 2019, all of whom had passer ratings adjusted to 2019.
     
    cbrad likes this.
  14. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    Sacks is another source of difference between EPA and passer rating. I also agree that ESPN’s subjective tweaking shouldn’t be trusted.

    there is definitely something weird with their Pass EPA for Tannehill. 18.1? For a QB that led the league in passer rating, CPOE, and YPA? I call bull****.
     
    Last edited: Apr 2, 2020
    resnor likes this.
  15. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    There is an average amount of discrepancy between EPA per play and passer rating across the league. The issue is that Tannehill is three standard deviations above that average in two different seasons of his career. There is nobody else who has come close to deviating from the league average to that degree in the past three seasons.

    What it suggests to me is that Tannehill tends to accumulate yards that aren’t meaningful in terms of expected points, thus his passer rating is artificially inflated in a way that’s less predictive of scoring and therefore of winning. Of course the correlations between passer rating and points, and between passer rating and winning, aren’t perfect.
     
    Last edited: Apr 2, 2020
  16. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    That's a bold claim. It's a claim that can be tested and I think you should at least present evidence for this before claiming it.

    Specifically, show what expected win% is using ESPN's clutch-weighted EPA per play vs. passer rating to see which predicts Tannehill's win% better. I can do the predictions for passer rating so that methodology can be copied. If you let's say adjust all 2002-2019 regular season (team) passer ratings to 2019, the best fitting line between win% and passer rating is:

    Win% = 1.08*(passer rating) - 47.6

    (and you can plug in the 2019 league average of 90.4 in there and you get 50% win%)

    So.. with a 117.5 rating you'd expect Tannehill to win 79.3% of his games. He won 70% (these are regular season stats).

    If you do the comparable analysis with ESPN's EPA per play and show it does better than 9.3% away from actual win% I'll be impressed. Since it's some work I'd be willing to accept going 5-10 years back, but you'll need to get sufficient sample size to make the best-fitting line reliable. Oh, and no need to adjust to a common year with EPA because it's already adjusted. Also, note that the best-fitting line is to QB EPA per play (meaning you have to use win% for that QB) as there is no comparable team stat to do the regression with.

    Anyway, for Tannehill I'm predicting passer rating does better than ESPN's EPA per play.
     
  17. Etrius24

    Etrius24 Well-Known Member

    682
    685
    93
    Mar 4, 2020
    How can you say that Tannehill's yards are not meaningful when the titans went from one of the worst offenses in the NFL in terms of scoring to the #2 scoring offense in football.

    PLease come up with some other crazy theory... This was awesome... in it's own messed up way
     
    Hiruma78, Cashvillesent and resnor like this.
  18. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    He was also one of the best, if not the best, QB in the red zone.

    I found another site that uses EPA (and not the manipulated version that ESPN uses). They also break the plays down by type so the number of plays equals the pass attempts.

    This is their definition under passing stats:
    EPA: Expected Points Added; the total change in the offense’s Expected Points that came on passes thrown by the player

    It is a pay site but they have a free trial.

    http://wiki.sisdatahub.com/

    In their data, Tannehill is 6th in the league in EPA pre play and 2nd in the league in percentage of pass plays that produced a positive EPA.

    upload_2020-4-3_8-4-54.png
     
    Last edited: Apr 3, 2020
    The Guy and resnor like this.
  19. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    I also notice that the ESPN list restricts the list to the top 30 QBs. I'd be willing to be that the list he is using for passer rating includes more QBs. So he is likely comparing the z scores for passer rating and EPA using different populations.
     
  20. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    If we do that, how will we determine whether the discrepancy between passer rating and EPA per play, in terms of predicting win percentage, reflects that one of them is measuring team functioning (as opposed to individual functioning) to a greater degree than the other?
     
  21. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    The sample I've been using consists of the players listed on the ESPN site (https://www.espn.com/nfl/qbr) for the past three regular seasons.
     
  22. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    ESPN mysteriously leaves off a number of QBs.

    Ryan Tannehill - 286 attempts, included
    Matthew Stafford - 291 attempts, included
    Mason Rudolph - 283 attempts included
    Joe Flacco - 262 attempts, not included
    Case Keenum - 247 attempts, not included
    Dwayne Haskins - 203 attempts, not included
     
  23. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    You can't. Both are team stats, and one of them is proprietary. My point is you made a claim without any evidence.

    Furthermore, note that the expected win% for Tannehill is higher than actual so you can't even superficially say he was performing worse than expected from a team with 70% win%. Besides, the standard error (standard deviation along the y-axis relative to the estimate) is about 10 passer rating points (a bit less for higher ratings since things asymptote) so the predicted win% (95% confidence interval) goes approximately from low 60's% to high 90% and 70% lies well within that.

    So there's nothing you can claim here with passer rating not being a good predictor, and I would bet that if you actually did the comparable analysis to what I did that you'd find that ESPN's clutch-weighted EPA per play is a worse predictor for Tannehill.
     
    resnor and The Guy like this.
  24. resnor

    resnor Derp Sherpa

    16,329
    9,874
    113
    Nov 25, 2007
    New Hampshire
    Let's also not forget, when Tannehill eventually retires, he will be in the record books as 4th all time...unless someone surpasses him by then...and they won't adjust his stats.
     
  25. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    Here's the thing, however -- EPA per play predicts win percentage better than passer rating. For the past five regular seasons those figures are 0.617 for EPA per play (95% confidence interval 0.510 to 0.706) and 0.572 for adjusted (to 2019) passer rating (95% confidence interval 0.456 to 0.669). The confidence interval for EPA per play is smaller as well.
     
  26. resnor

    resnor Derp Sherpa

    16,329
    9,874
    113
    Nov 25, 2007
    New Hampshire
    The Guy is basically Tannehill's Carol Baskin.
     
    Irishman likes this.
  27. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    I wonder if that's related to the Charlie Batch story. When ESPN's QBR came out (which is based on their clutch-weighted EPA) people pointed out that Charlie Batch's 17 attempt 186 yard 2 INT game was the highest ever QBR ever recorded for any game at 99.9. ESPN instead of modifying their formula decided to artificially remove Batch from their list of highest QBR by setting a minimum threshold for attempts. Batch's 99.9 is still there for all to see:
    https://www.espn.com/nfl/player/gamelog/_/id/1490/type/nfl/year/2010

    But he's not on any list of highest QBR (at least not last I checked). It's possible they decided to remove QB's with too few total attempts to prevent sample size related issues like that. Now.. Batch had 17 attempts in that game and 17*16 games = 272 attempts, and all the ones you listed that are not included are below 272 attempts. Coincidence?
     
    resnor likes this.
  28. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Means nothing because you don't know what extra parameters ESPN is including in their proprietary "clutch-weighted" formula. Every mathematician will tell you that by including more parameters you can predict things better, so it's trivial to say you can predict something better if you use more parameters. How many does ESPN use?

    Remember, we know QBR uses 10,000 lines of code when you only need a few lines for a look-up table, and you yourself showed that ESPN's clutch-weighted EPA is nearly perfectly correlated with QBR, so that means that most of those 10k lines of code are for "clutch". Add more parameters => trivial increase in predictive power. Doesn't mean they're actually capturing something relevant in the game. That's just a mathematical trick you can play with things that have no physical meaning.

    Best to stay away from clutch-weighted EPA. Use EPA if you want. That's a solid concept. But stay away from anything proprietary.
     
    FinFaninBuffalo likes this.
  29. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    OK so let's use the above data, which apparently removes ESPN's clutch weighting.

    For the top 33 QBs in the league in terms of pass attempts in 2019 (ending with Dwayne Haskins at 203 attempts), here are the z-scores for the discrepancy between EPA per play z-score and passer rating z-score:

    Ryan Tannehill 2.38
    Russell Wilson 1.61
    Joe Flacco 1.56
    Dwayne Haskins 1.39
    Drew Brees 1.12
    Daniel Jones 1.08
    Kirk Cousins 1.07
    Case Keenum 0.84
    Mason Rudolph 0.71
    Gardner Minshew 0.66
    Carson Wentz 0.51
    Jimmy Garoppolo 0.36
    Sam Darnold 0.31
    Aaron Rodgers 0.26
    Mitchell Trubisky 0.12
    Kyler Murray 0.07
    Derek Carr -0.01
    Tom Brady -0.34
    Josh Allen -0.39
    Jacoby Brissett -0.41
    Matthew Stafford -0.49
    Kyle Allen -0.62
    Matt Ryan -0.63
    Andy Dalton -0.65
    Ryan Fitzpatrick -0.79
    Deshaun Watson -0.89
    Lamar Jackson -0.95
    Baker Mayfield -1.09
    Jameis Winston -1.25
    Dak Prescott -1.26
    Patrick Mahomes -1.30
    Philip Rivers -1.35
    Jared Goff -1.61

    Now consider that when we use ESPN's EPA per play for the past five regular seasons, here are the top 20 QBs in the league during that period of time, again in terms of the z-score of the discrepancy between EPA per play z-score and adjusted (to 2019) passer rating z-score:

    2018 Ryan TannehillMIA 2.76
    2019 Ryan TannehillTEN 2.70

    2017 Jared GoffLAR 2.61
    2019 Kirk CousinsMIN 1.83
    2015 Derek CarrOAK 1.82
    2015 Russell WilsonSEA 1.79
    2017 Drew BreesNO 1.76
    2017 Mitchell TrubiskyCHI 1.70
    2015 Blaine GabbertSF 1.68
    2018 Russell WilsonSEA 1.39
    2019 Gardner Minshew IIJAX 1.37
    2017 Kirk CousinsWSH 1.32
    2015 Ryan TannehillMIA 1.30
    2016 Sam BradfordMIN 1.28
    2019 Drew BreesNO 1.26
    2017 Andy DaltonCIN 1.23
    2016 Ryan TannehillMIA 1.19
    2018 Derek CarrOAK 1.10
    2015 Sam BradfordPHI 1.10

    So, this isn't a one-season thing for him, and it occurs at various levels of passer rating. Of course he didn't play in 2017, or that figure might be in the top 20 above as well.
     
    Last edited: Apr 3, 2020
  30. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    For some reason you don't seem to understand that when a player has a single season that's extremely discrepant from six previous seasons in the league, it makes some sense to take a look at what sorts of situational advantages he might have been benefiting from.

    To shut down that process without exploring it fully actually shows a bias for Tannehill, and (perceived) bias is the thing you're apparently railing against here.
     
  31. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    Yeah, well, I care as much about this as you do about YPA, CPOE, red zone performance, and the many other stats that indicate a high level of QB play. You can have your single unexplained quirk.
     
    resnor likes this.
  32. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    LOL... bias..... laughable after you spent 6 years denying the situation disadvantages that hindered him.......
     
    Hiruma78, Irishman and resnor like this.
  33. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    Since it involves at least four seasons of his career and appears to be highly specific to him and not to other QBs, why not try to explain it?
     
  34. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    Maybe because is a fudged number? Haven't we already established that?
     
    resnor likes this.
  35. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    Well hold on now, ESPN was supposed to have been providing the "fudge," and when we remove ESPN's proprietary component, we get a consistent result. Tannehill is still atop the league.
     
  36. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    LOL at you looking for an alternate explanation for a QB that was:

    1st in passer rating
    1st in YPA
    1st in NY/A
    1st in ANY/A
    1st in CPOE
    1st in red zone completion %
    1st in inside the 10 completion %
    1st in red zone on target %
    2nd in TD %
    2nd in red zone TD %
    4th in average completed air yards
    3rd in average intended air yards
    3rd in air yards to the sticks
     
    Hiruma78, Irishman and resnor like this.
  37. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    And what happens to all the other results? Everyone else stay pretty much the same?
     
  38. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    Sure, because what we're examining here is whether he enjoyed a situational advantage, and EPA incorporates the situation in its formulation.

    Take a look at the following article and realize that the two variables fundamentally utilized in this very comprehensive approach are EPA and WPA:

    https://www.degruyter.com/view/journals/jqas/15/3/article-p163.xml?tab_body=pdf-69320
     
  39. FinFaninBuffalo

    FinFaninBuffalo Well-Known Member

    2,474
    2,954
    113
    Dec 13, 2007
    LOL.... sure. Did a situation "advantage" explain his 2018 season? Also, you are not using the EPA described in the article for the most part. You are using the fudged numbers.

    BTW, CPOE takes into account the situation (distance to the receiver, separation from defender) also. He was first in that metric, by a lot.
     
    resnor likes this.
  40. The Guy

    The Guy Well-Known Member

    6,598
    3,323
    113
    Oct 1, 2018
    A situational advantage could explain each of his four seasons noted above, in that if his passer rating would've been consistent with his EPA and not so discrepant from it, his passer rating would've been lower every one of those years.

    You seem to be implying that the situation in 2018 wasn't advantageous, and I understand that concept, but there are numerous situations in every game, regardless of the team a player is on, that offer the opportunity to increase passer rating without correspondingly increasing EPA.

    That's what I mean by "situational advantage" in this instance. His performance, combined with the situation, did wonders for his passer rating, but far less for his EPA. Or conversely, there were instances in which his EPA was lowered, and his passer rating was unaffected.

    Not surprising, given the following:

    "That chart shows that Tannehill has been the least accurate quarterback in the league on deep throws, but Titans receivers have caught EVERY catchable pass he’s thrown. As you can see with all the red on the chart … that’s not normal. That chart was made before the Raiders game, but Tennessee’s receivers continued to be perfect on deep targets, hauling in all three of Tannehill’s catchable deep targets."

    https://ftw.usatoday.com/2019/12/ryan-tannehill-titans-franchise-qb-new-contract
     

Share This Page