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

Thoughts on the Dolphins/Cowboys Game

Discussion in 'Miami Dolphins Forum' started by KeyFin, Sep 22, 2019.

  1. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    No. It's like you're saying they it's hot outside because your thermometer reads 90. But it's hot because the sun is heating the Earth. The number on the thermometer is a result of the sun heating.

    Basically, if passer rating was what determined wins, then you'd see every team just throwing short easy to compete passes, so they could hit they magic passer rating to get a win. The rating is simply a result of successful passing plays...it's not the cause. I mean, it's common sense that winning teams usually have better passer ratings, but the passer rating simply tells you that plays were successful...it doesn't actually tell you anything about the QB. It does not tell you if the QB was making good throws, just that whatever he was throwing was being caught.
     
  2. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    Obviously it's true that passer rating is the effect not the cause. However, I think you're underestimating how difficult it is to deliberately produce a high passer rating. For example.. let's say the QB threw the ball 50 times for 50 completions, 3 yards each time (150 yards) but no TD's or INT's. What do you think that passer rating is?

    It's 79.167. .. (league average in 2018 was 92.9)

    What if it was 4 yards each time? 83.33.

    You see how difficult it is to deliberate produce a high passer rating with lots of short high percentage passes? You really need high Y/A as well as a bunch of TD passes. So the link between QB ability and high passer rating is much tighter than you think.

    This analogy is more interesting than you guys are making it out to be. Saying something is "hot" or "cold" is referring to a sensation. It's not a physical property of a system. Heat and temperature are physical properties but "hot" and "cold" are not. So "hot" and "cold" are more analogous to a person's subjective rating of a QB, while passer rating is the temperature and QB ability is heat.

    There's another issue with the heat vs. temperature analogy: you can add tons of heat without increasing temperature. For example, the temperature of a cup of water might be the same as that of a swimming pool but the swimming pool has tons more heat. Temperature is a measure of the average energy of molecular motion while heat is the total energy. So the analogy works only in a closed system.
     
    Surfs Up 99, Pauly and The Guy like this.
  3. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    Not sure why you chose to do your rating with no tds. Plus, current past rating includes yards gained after the catch as part of the pass, so it's really an unfair comparison to average passer rating. Now if your were to find that l the average air yards for a pass on the league, and figure passer rating based on ONLY that (no tds, no ints, no tax), then we'd have a reasonable comparison.
     
  4. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    What I'm getting at here is the construct validity of passer rating, along the lines of the following:
    In other words, passer rating is a more valid construct than whatever set of events in the passing game some people are ascribing to winning.

    https://psychclassics.yorku.ca/Cronbach/construct.htm
     
  5. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    You can't just deliberately create passing TD's by having a strategy of mostly throwing short high probability passes. You CAN deliberate throw a lot of high percentage short passes, but they aren't going that far. You'd have to get to an average of just over 6 Y/A before you even get league average passer rating today.

    Point is.. a coach can't just say let's design a game plan to get a high passer rating and do it. It's not as easy as you think because the link between QB ability and passer rating is fairly strong (precisely because you can't just create high passer rating at will).
     
    Pauly likes this.
  6. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    None of that is really necessary when 1) career passer ratings correlate strongly with consensus perceptions of quarterbacks' individual ability, and 2) passer rating correlates strongly with winning.

    In other words, passer rating has construct validity and predictive validity, and so there is really no need to formulate another measure.
     
  7. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    The stuff they're talking about applies when you're trying to measure something defined through a latent variable – something that can't be directly observed. Speed, time, distance.. even passer rating are all "manifest variables" or things that can be observed so questions about construct validity don't really apply to these things (no need to ask the question).

    Stuff like construct validity applies to things like health status questionnaires where you are trying to infer a measure (e.g., of a person's ability) based on their responses to questions like "how difficult is it for you to do X". The responses are on a rating scale and mathematics like Rasch analysis is used to take those ordinal ratings and construct an actual measure from it. Where stuff like "construct validity" comes in is when a researcher develops or analyzes a new questionnaire and wants to "validate" it. That is, they want to show that the intrinsic psychometric properties of the questionnaire conform to basic assumptions about measurement like unidimensionality, etc...

    Also, that reference is somewhat outdated if they are suggested using Cronbach's alpha. That statistic has been panned like crazy because it's sensitive to sample size. For fun, you should read this article titled "On the Use, the Misuse, and the Very Limited Usefuless of Cronbach's Alpha" lol. Very insightful and funny at the same time:
    https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2792363/
     
    The Guy likes this.
  8. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    And that's precisely why increasing passer rating is best accomplished by doing what every team in the league is trying to do: acquire a quarterback with great individual ability.

    The Cardinals just jettisoned a QB drafted in the top 10 overall after 13 games started, simply because they had the opportunity to get another one they believed is much better. That's how important it is to have a quarterback of this nature.
     
    Pauly and cbrad like this.
  9. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    What I meant was passer rating as a measure of the latent variable of quarterbacks’ individual ability. Do you think it applies in that case?
     
  10. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    You'd need specific evidence of that which can't be obtained by testing "validity" because "validity" only tests consistency with assumptions about fundamental properties of measurement, which obviously any physical measure like passer rating will have.

    Really, the way to show to what degree passer rating is an actual measure of the latent trait we call QB ability is through that structural equation modeling approach described earlier. First, find a model of the game that as far as we can tell most accurately represents the relations between observable stats. Then, look at the correlation between whatever the implied "QB measure" is from that chosen model and passer rating. The r^2 should be a direct enough test of how close passer rating is tied to QB ability.

    Maybe there's another approach but I can't think of one right now.
     
    The Guy likes this.
  11. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    1. QBs with higher passer ratings, tend to be on teams that win.
    2. Most people attribute wins to QBs (but losses are usually put on the defense or special teams, not individual players).
    3. Since people attribute wins to QBs, they're always going to favor QBs on winning teams.

    I don't think it's anything special that career passer ratings correlate to consensus opinions on QBs. Add in that most people aren't watching all the games, they're just seeing highlights and stats. They see passer rating and think it means something that it doesn't. Then the little list above kicks in.
     
  12. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    There would be 3 years passes that a receiver breaks a tackle and goes for and. There would be touchdowns inside the red zone. It's ridiculous to compare a passer rating consisting of nothing but a bunch of 3 yard passes with the league average rating that also includes TDS and yardage after the catch. Surely you can understand that. The comparison you made is completely ridiculous.
     
  13. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    I think it's ridiculous to suggest you can manufacture a high passer rating.

    I was just pointing out that lots of short passes don't mean much in the formula for passer rating. That remains true even with many TD's or INT's. So your suggestion of using lots of short passes isn't going to do the trick.

    What you need to do to "manufacture" high passer rating is to get high Y/A and/or high TD%. So how do you do that? Right.. get a better QB and/or better surrounding cast. There's no short-cut here.
     
    The Guy likes this.
  14. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    What's the average air yards last season?
     
  15. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    According to Next Gen Stats it looks like average CAY = completed air yards (this is average air yards ONLY for completions) is around 5.86 while average IAY = intended air yards (average air yards for ALL passes) is around 8.1.
    https://nextgenstats.nfl.com/stats/passing/2018/all

    I say "around" because we're looking at individual QB's there and not passing stats for entire teams which is what we really want. The true CAY and IAY should therefore be a bit less than those numbers.
     
  16. The_Dark_Knight

    The_Dark_Knight Defender of the Truth

    8,678
    7,080
    113
    Nov 24, 2007
    Melbourne, FL
    Oh really? Difficult to do on my phone but be patient my young apprentice, I’ll find plenty of examples that nullify your virtual world of football and bring you back into reality.
     
  17. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    Why do you think that is?
     
  18. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    Because better teams have better overall talent around the QB. I think once you get your QB is easier to build around them. Teams that win more attract better coaches and better players. It's not rocket science. I can't make it any simpler than this: once the ball leaves the QBs hand, he is no longer responsible for the outcome. People give too much credit to the QB.
     
  19. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    Your original point was that QBs with high passer ratings tend to be on teams that win.

    Here's some information.

    Since 2004, there are the following correlations in the NFL.

    Between passer rating and win percentage: 0.83
    Between passer rating and points scored: 0.93
    Between points scored and win percentage: 0.92

    When we control for points scored, the partial correlation between passer rating and win percentage plummets to -0.15.

    In other words, passer rating doesn't equal winning when passer rating doesn't translate to points scored. A team's passer rating could be high, and if it doesn't translate to points, the team may or may not win.

    So QBs with high passer ratings don't necessarily win when those high passer ratings don't translate to points scored. They tend to win, you're right, but they don't always win.

    So, since 2004 we have the following examples of relatively high season passer ratings (greater than or equal to 100), ordered by points scored in those seasons:

    https://www.pro-football-reference....1comp=gte&c1val=100&c5val=1.0&order_by=points

    The range of season points scored in that table is 367 to 606.

    The question is, in the instances in which high season passer ratings translated to comparatively fewer points, and therefore the absence of winning in many cases, what were those high passer ratings measuring?
     
  20. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    Keep in mind that "my virtual world" as it relates to the conversation with you encompasses nothing more than the point that passer rating differential is associated with 87% of the variance in win percentage in the NFL since 2004.

    Given that, it was only too easy to determine that the 45-point passer rating differential your original point consisted of would translate to an overwhelming winning record (28-1 since 2004) for the teams on the positive end of that differential.

    The problem you're having is that you're speculating about how the league functions without really knowing how it functions.
     
  21. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    That's all well and good. Passer rating is incredibly flawed, favors QBs on good teams, and is a result of successful plays, not a cause. When you constantly correlate to win%, you're applying the rating erroneously, imo.
     
  22. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    The point of the post you responded to is that passer rating doesn't favor QBs on good teams, in some cases.

    And in those cases, what is it measuring?
     
  23. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    What I mean by that, is it favors them as regarding popular opinion on QBs. You keep referencing that passer rating correlates with who is thought to be the good QBs. As if that is proof of something. I'm simply saying most QBs on good teams that generally have the better passer rating. That is expected, as better teams win more, and better players make more good plays happen for QBs. You're looking at the result ( passer rating and wins) and using those two things to prove your point. It's kind of like using a word to define itself. It's not how it works.

    I'm saying passer rating favors winning teams, and the QBs on that team. Is it harder to make one good play or 30 good plays? I mean, clearly the fewer number of pays, the less chance there is for bad plays. There's a reason why you guys like to use 150 attempts as a minimum threshold. So that QB on a good team that is winning, is facing probably a worse defense, or at least a defense that is usually playing worse than his defense. While the opposing QB, depending on how much they're losing by, may no longer be able to take safe plays, and now has to to try to force things down field, into a defense that knows they're passing. It's ask symbiotic. You're trying to separate out something can't be separated.
     
  24. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    Wait.. this isn't correct. Those values are way too high even intuitively.

    For example, if the correlation between points scored (PS) and win percentage (WP) is 0.92 then 85% of the variation in WP is explained by the offense? It should be closer to 50% (should be a bit over 50% because some of what the defense does is incorporated in points scored).

    Since 2004, the average correlation between PR and WP is 0.669, between PR and PS is 0.814, and between PS and WP is 0.762.
    Since 1966, the average correlation between PR and WP is 0.633, between PR and PS is 0.778, and between PS and WP is 0.752.

    Using the 1966-2018 values, 56.5% of the variation in win% is explained by points scored, 40% of win% is explained by passer rating, and 60.5% of points scored is explained by passer rating.
     
    Irishman likes this.
  25. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    Again the question is, when a quarterback achieves a high passer rating in an offense that isn’t good, what is that passer rating measuring?

    I don’t see anything in your post above that answers that question.
     
  26. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    What that proves is that the discrepancy between how humans view QB's and what you get from the stats isn't that great with large sample size. And while it's true humans don't ignore the stats, it's also presumptuous to assume that all those HoF QB's were voted in because people were looking at stats and ignoring what they think they saw/remembered about the QB's.

    Of course the advantage of stats in that case is that they're objective and you can quantify differences even if information is thrown away.

    Passer rating favors winning teams, but the ONE constant in passer rating over time is the QB. The QB isn't facing the same defense all the time nor throwing to the same WR, etc.. So just intuitively the MOST important factor in passer rating is the QB.

    And humans also need to look at a lot of plays to evaluate QB's. Show a human only one play from a QB they've never seen and see how close their opinion of that QB aligns to after 150+ passing plays. Sample size may be less important for humans but it's not irrelevant.
     
    The Guy and Irishman like this.
  27. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    Correlate the passer rating column with the win percentage column in the following table:

    https://www.pro-football-reference....mp=gte&c1val=0&c5val=1.0&order_by=points_diff
     
  28. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    Show me specifically an example. If a passer rating is high, then they were successful on passing plays, regardless of how good that offense is usually.
     
  29. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    The ONLY constant is the QB? He's throwing to the same WR corp. If a QB ONLY threw to one receiver, it would be easily defendable, and his passer rating would almost certainly be poor.
     
  30. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    "WR corp" is not what the QB is throwing to on each passing play. The QB is throwing to a single WR on each passing play (e.g., if you took sample size = 1 with passer rating). So yes the ONLY constant is the QB.
     
  31. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    Again, this isn't something strange. Winning teams, have better ****ing offenses, and greet functioning defenses, so I expect that passer rating of a QB on a winning team will be higher than a QB on a poor team. However, that doesn't mean automatically that the QB on the worse team is worse, and the QB on the better team is better. It means that they were more successful, and there's a plethora of reasons, outside of QB, as to why that might be the case. Passer rating is a result, not a cause.
     
  32. resnor

    resnor Derp Sherpa

    13,399
    7,626
    113
    Nov 25, 2007
    New Hampshire
    Dude.

    Stop. On every given play, every receiver is an option. So yes, the WR corp is a constant. If you have an offense with one great receiver, and 3 terrible receivers, the QB is going to struggle because the great receiver will get doubled, and he won't be effective, and the offense will be not good. But put that great receiver on a team with some decent/good receivers, and that offense will be more successful, AND the great receiver will look even better.

    Cbrad, did you play any organized sports? Not looking to make fun or anything, but it seems that you don't understand some basic things about sports that people who played organized sports kind of get intuitively.
     
  33. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    Yes! Now.. WHY is the correlation so high when you just combine across years like that? It's because you're not taking into account passer rating inflation. That is.. you're assuming that it was equally likely to get a 100 rating in 2004 as in 2018.

    But that's not true of course. So what you want to do is to calculate the correlation to win% separately for each year and then average. That gives you a correlation that more closely aligns with what to expect in any given year. Otherwise, the "variance explained" makes no sense as I pointed out before.

    Oh, and note how much lower the year-by-year correlations are. In some cases (e.g., 2015) the correlation between passer rating and win% is below 0.5 while in other years it's over 0.8.
     
    The Guy likes this.
  34. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    I only played organized sports when I was a kid. Doesn't matter for this discussion though because you can't say throwing to WR1 is the exact same thing as throwing to WR2 even if the group of WR's is the same, and the group of WR's out there often isn't the same from play to play. In fact people calculate passer ratings when throwing to different WR's to get a better sense of how good/bad the WR is.

    No the ONLY constant in passer rating is the QB. That's provably true btw because passer rating sums up stats on a play-by-play basis and you aren't throwing to the same WR on every single play.

    btw.. if you REALLY believe what you're saying (which I doubt), then from now on you shouldn't blame Preston Williams for those drops. You should blame the "WR corps" for his drops.
     
    Last edited: Sep 27, 2019
  35. AGuyNamedAlex

    AGuyNamedAlex Well-Known Member

    1,888
    1,664
    113
    Sep 12, 2015
    The line is a constant too in that case. Barring injury you always have the same 5 guys protecting you.

    Also now you're disregarding a lot of information.

    How do you quantify a route combo that WR#2 runs to perfection to free WR#1?

    How can you quantify the TE going down the seam and opening the underneath for WR#2?

    There are moving parts on every play and any attempt using a passer rating to completely analyze a QB will fail.

    It's a piece of team data that gives you a look into how well the offense runs. It may be mainly due to the QB or it could be a highly efficient offense and a slightly above average QB.

    The argument shouldnt be whether or not these numbers are important, it's how much weight on their own they carry without context.
     
  36. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    The OL isn't a constant because of injury and new players over time obviously. And we don't have to know how to quantify every possible factor if we're keeping one thing constant. Point is, in the actual formula for passer rating the only constant is the QB. That's a fact.

    And I added an edit to the previous post before you quoted it: if you really think it's the "WR corps" that we should be talking about, then let's no longer blame Preston Williams for HIS drops. Let's say it was the "WR corps" that dropped the passes.
     
    Irishman and The Guy like this.
  37. AGuyNamedAlex

    AGuyNamedAlex Well-Known Member

    1,888
    1,664
    113
    Sep 12, 2015
    I mean, I'm not even arguing the validity of the end number on winning. I just think it's silly to look at QB rating as a measure of the QB himself.

    Even if you want to say the QB is a constant, it doesnt change anything at all. There are still 10 others players who either did or didnt do their job on a play that will impact the QB every one of his reps.

    You're also basing everything off the assumption that QB rating is a good way to analyze a QB individually when just about every scout will tell you otherwise. It only tells you if the QB is being successful within his offense, not his ability level overall.

    The QB rating is an offensive grade, not a QB grade and its a shame it was named what it was or we wouldnt be having this debate.
     
  38. cbrad

    cbrad . Club Member

    7,602
    8,999
    113
    Dec 21, 2014
    Passer rating is a passing offense stat obviously.

    However, because the only thing the stat itself keeps constant is the QB it becomes MORE of a QB-only rating the more you can average out the effects of the QB's surroundings, which can only be seen with increase in sample size. And as I showed you here:
    https://www.thephins.com/threads/thoughts-on-the-dolphins-cowboys-game.94643/page-3#post-3207211

    you can visibly see the effect of sample size averaging out things even over the course of a single season. As I pointed out though, the standard deviation in 5-year win probability is 0.15 which is half-way between truly random (0.29) and no random variation (0). So you can't average out everything.

    So it is what it is. What's important to remember is that it's easy to criticize but hard to show how to do things better.

    First, can you come up with a better system that not only is objective ("expert" opinions disagree way too often) but predicts outcomes better? Really hard to do. So it's fine to understand the limitations of passer rating but it's one of the best tools out there for measuring QB ability over longer periods of time.
     
    Irishman and The Guy like this.
  39. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    What you said originally was that quarterbacks with higher passer ratings have tended to be on teams that win.

    What you’re implying there is that passer rating is really a measure of the overall functioning of the offense, which in turn contributes to winning, and which would occur via points scored on offense.

    However, there are plenty of examples of quarterbacks with high passer ratings who have been in not so great offenses, in terms of season points scored, and I provided those examples above.

    In those instances, those quarterbacks are neither part of stalwart offenses nor part of great teams in terms of winning, yet they have high passer ratings nonetheless.

    The question is, in those instances, what is passer rating measuring?

    You certainly can’t say it’s measuring offensive point production or winning in those instances, because it’s doing neither.
     
  40. The Guy

    The Guy Well-Known Member

    1,880
    868
    113
    Oct 1, 2018
    cbrad, if you are willing and able, please post the top 200 quarterbacks in NFL history with a minimum number of passing attempts you feel is sufficient, ordered by era-adjusted career passer rating.
     

Share This Page