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

What the return of Tannehill means for 2018

Discussion in 'Miami Dolphins Forum' started by Pauly, Feb 26, 2018.

  1. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    This thread builds on some threads I've started over the last few off seasons. This one is probably the most relevant: https://thephins.com/threads/building-a-winning-team.91138/

    Anyway, updating my data base to include the 2017 season. All PRs have been adjusted to a 2017 base number to account for the trend of increasing passer rating since the 2002 rule changes.

    2017 had an average league passer rating of 85.1 down from 87.1, 88.4, and 87.6 from 2014 to 2016. 2014 and 2015 in particular were steps above the long term trend and 2017 was a step down. The long term trend for passer rating from 2002 to 2017 is that 87.2 was the trend. 2017 seasons featured a number of significant injuries to quality starting QBs (including RT17), plus the situation of Jimmy Garropolo who lost 10 starts because there were teams out there that thought a second round pick was too high a price to pay for a starting QB. So I expect the league passer rating for 2018 to be closer to 90 than 85.

    The correlation of team passer rating to win%: 0.67
    Standard deviation of offensive PR:11.9
    Line of best fit: 68.52 = 0 wins 109.16 = 16 wins. An increasing your season passer rating by Increasing your team passer rating by 2.54 points increases your expected wins per season by 1.

    Correlation of team passer rating allowed to win%: -0.53
    Standard deviation of defensive PR: 9.2
    Line of best fit: 101.94 = 0 wins, 75.96 = 16 wins
    Decreasing your team passer rating by 1.57 points increases your expected wins per season by 1.

    Correlation of the difference between passer rating made and passer rating allowed: 0.80
    Standard deviation of PR differential: 16.1
    Line of best fit: -33.56 = 0 wins, 32.43 = 16 wins.
    Increasing the PR difference by 4.1 increases your expected wins per season by 1.

    There are several methods to look at whether a teams W-L record outperforms or under performs their expected wins based on statistical methods.
    According to our PR differential of -16.1 we should have been a 3.9-12.1 team.
    According to the points differential (http://www.footballoutsiders.com/stat-analysis/2018/2017-adjusted-pythagorean-wins) we should have been a 4.9 -11.1 team.
    We were 5-2 in 0-7 point games, and statistical experts expect a teams record to regress to 50% over time. if we went 3.5-3.5 in the 0-7 point games then our record would have been 4.5 - 11.5
    So based on a number of methods our expected win total with the team performance last year is in the 4-5 win range.
    NB I think that out few total melt down games bias our statistical expected wins downwards but that's a subject for another thread.

    The question is whether a defense or offense helps the other side of the ball. In the previous thread I started it showed that a really bad offense (more than 1 standard deviation worse than average) can hurt the defense by about 5 PR points, but a great unit doesn't help the other side if the ball, as measured by passer rating, and a bad defense doesn't hurt the offense.

    More generally from 2002 to 2017 there is a slight negative correlation between Offense and Defense of -0.16. So as your offense gets better your defense gets worse.

    With a +1 increase in offensive passer rating leading to 0.6 decrease in passer allowed. It should be remembered that the standard deviation for the difference is 16.1, so the variance is huge. With the scatter plot I come up with, I cannot judge by eyeball where a line of best fit would go, and I would have guessed a near 0 correlation.

    There is a general expectation that as team PR rating goes up team PR allowed will go down as the salary cap etc. means you will have more assets devoted to the defense. The numbers suggest that having a good offense gives a slight boost to the defense.

    What level of performance can we expect from Ryan Tannehill.
    In his last 3 seasons his simple average of passer rating is 91.7
    Last year Dolphins QBs had a combined PR of 78.7

    Assumption 1:
    He is the same as he was 2014-2016
    The simple expectation is that with a +13.0 increase in PR will lead to +4 wins, so 10-6 here we come baby! But we have to factor in the fact that we over-performed in expected wins last season. A more realistic assumption is that 8.5 - 7.5 is the record if he comes back and performs at his 2014-2016 levels.

    The effect of the defense is harder to gauge. Was it hindered by horrible offensive production? Well if we take the slight overall trend, our PR offense was 6.4 points worse than average which translates as roughly a 3.8 point penalty to the defensive PR allowed.
    If nothing else changes our PR differential would go from -16.1 to +13.0 for RT17 and +3.8 for removing the negative effect of poor QBing in 2017. leaving a passer rating differential of 0.7 or an 8-8 season.

    However Adam Gase went 7-2 and 5-2 in 0-7 point games over the last 2 seasons. If some of this is due to skilled coaching not random chance we might expect a +1 or +2 over expectations

    Bottom line. If RT17 is the same as he was in 2014-2016 and nothing else changes we're in 8-8 hoping to get to 10-6 territory.

    Optimistic Assumption
    RT17 comes back, and everything clicks with Gase offense and he posts a 95-100 PR season.
    I won't bore you with the math and my assumptions.
    Bottom line 10-6 or 11-5 and maybe we'll finally catch those pesky Patriots if we get lucky in some other areas.

    Pessimistic assumption.
    He comes back, but the knee doesn't hold up well and his performance drops to the 85ish PR.

    Puts us in 6-10 territory maybe getting up to 8-8 if Gase's bag of fairy dust keeps getting us more close wins than statisticians expect.
     
    mbsinmisc, Puka-head, Dorfdad and 3 others like this.
  2. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Your best-fitting lines are wrong.

    Here are the graphs of the data and best-fitting lines for 2017-adjusted passer rating, 2017-adjusted defensive passer rating, and 2017-adjusted passer rating differential, from 2002-2017:

    [​IMG]


    [​IMG]


    [​IMG]

    The slopes tell you how many extra wins you'd expect if you increase passer rating by one point (0.1785 extra wins), defensive passer rating by one point (0.1892 extra wins) or passer rating differential by one point (0.1551 extra wins).

    Flipping it around so that we're talking about how many extra passer rating points you'd need for one win, it's just 1 divided by those numbers, so 1/0.1785 = 5.6 extra passer rating points for 1 win (as opposed to 2.54), 1/0.1892 = 5.29 extra defensive passer rating points for 1 extra win (as opposed to 1.57), and 1/0.1551 = 6.45 extra passer rating differential points for 1 extra win (as opposed to 4.1).

    Looking at your numbers and the graph I think you're trying to visually fit a line through the data instead of doing an actual linear regression. Linear regression minimizes the sum of the squares of the differences between predicted and actual, and that's not easy to get right by eyeballing things (people often underweight large errors).

    In any case, this does show that while the correlation between wins and defensive passer rating differential is by far the highest, you're better off improving pass defense than pass offense if you go by expected wins.

    Oh.. one final thing. I think there's a small error somewhere in your passer rating allowed data. The correlation between 2017-adjusted defensive passer rating from 2002-2017 to wins is -0.5607. The correlation to win% (where ties are included as half a win) is -0.5618. Either way that's a big enough difference from -0.53 that I think you might have copied or adjusted some part of your data wrong.
     
    Irishman and Pauly like this.
  3. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    The best way to predict win/loss record from point differential is not that "pythagorean" stuff (not a criticism of you Pauly.. just football outsiders). You want to look at standard deviations above and below the mean for each year for points scored and points allowed and find the best-fitting plane against win%.

    That relationship is one of the most stable I've seen across era in the NFL and is one of the most important stats. From 1970-2017 this is how points scored vs. points allowed relates to win%:
    [​IMG]

    The slopes there are highly informative. That graph says that if you increase your z-score on offense (standard deviation above the mean) by 1, then you can expect an increment in win% of 11.29%, while increasing your z-score on defense by 1 increments win% on average by 10.76%.

    The ratio 11.29/10.76 = 1.0493 says that across NFL history the offense is about 5% more important for increasing win% than the defense.

    In any case, that equation can be used to estimate what our expected win% should have been and whether we outperformed or underperformed. In 2017 the Dolphins scored 281 points and gave up 393 points compared to a league average of 347.5. The league-wide standard deviation for points scored in 2017 was 65.15 and for points allowed was 44.71, meaning the Dolphins were almost exactly 1 standard deviation below the mean for both offense and defense.

    Using that equation that means our win% should be expected to be approximately 28% = 4.5 wins, which in this case is similar to the 4.9 predicted by football outsiders, but in general won't be.
     
    Last edited: Feb 26, 2018
  4. Serpico Jones

    Serpico Jones Well-Known Member

    4,697
    1,667
    113
    Feb 1, 2012
    I feel like I’m in geometry class. Get a grip, guys.
     
    Sceeto and shamegame13 like this.
  5. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,648
    67,540
    113
    Dec 20, 2007
    football porn for Cbrad and Pauly
     
  6. TheHighExhaulted

    TheHighExhaulted Well-Known Member

    5,819
    4,665
    113
    Jan 15, 2008
    Nobody actually read all of this, right?
     
    Sceeto and shamegame13 like this.
  7. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    I am unhappy with the “pythagorean” method used. Mainly because I feel that a few blowouts can bias a 16 game sample too much. Also as they imply in their article it’s the leading team that tends to pull away because they can be more conservative and avoid turnovers while the trailing team has a significantly increased risk of turning the ball over as They aggressively try to score.
     
    Irishman likes this.
  8. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    I’ll recheck my data on the trend lines, I had win% on the X axis and passer rating on the Y axis, but my graphs look pretty much the same as yours.

    I got a slight negative correlation between PR made and PR allowed, I was wonderin how that holds up for points scored and points allowed.
     
    Irishman likes this.
  9. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    They also implicitly assume offense and defense are equally important, which as you can see in post #3 isn't true.

    Ah! That explains it. You have the axes switched!! Linear regression is dependent on what the y-axis is because the errors are computed only along the y-axis. And since you're trying to predict wins or win%, that should be on the y-axis.

    And yes if you switch the axes I get similar numbers to yours except for defensive passer rating where I'm getting a slope of -1.66 instead of -1.57 so I still think you might have copied defensive passer rating incorrectly somewhere there (my data is directly imported so I doubt there's an issue). Oh, and correlation between offensive and defensive passer rating is -0.182 for me instead of -0.16, probably for the same reason.

    [​IMG]
    The average correlation for points scored vs. points allowed across NFL history is -0.334 but since 1978 (a key demarcation point that's seen in so many stats) it's a bit less at -0.289.
     
    Irishman likes this.
  10. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    Arigato sensei, wakirimashta.
    I didn’t realise that flipping the axes would make the difference in the linear regression.

    So the slight negative correlation on offensive production to defensive production implies that being good on one side of the ball is a slight benefit to the other side of tne ball, but the explained variance is small, less than 5% (also sucking on ine side of the ball hurts the other).
    So the ‘you need the offense to hold onto the ball to help the D’ line is looking at the fringes of whatever the real problem may be.
     
    Irishman likes this.
  11. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Dozo Pauly-san. lol
    Yeah, less than 10% of the variance if you go by points scored/allowed and less than 5% if you look just at pass offense/defense measured by passer rating.
     
    Pauly and Irishman like this.
  12. Phin McCool

    Phin McCool Well-Known Member

    713
    735
    93
    Jan 29, 2017
    United Kingdom
  13. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,648
    67,540
    113
    Dec 20, 2007
    I feel like Tommy boy when he goes into the factory and acts like he knows what going on.

    “ I need you to check out the speck on the gurter”
     
    Mafioso and Irishman like this.
  14. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    Don’t worry it’s just a fancy mathematical way of saying that even if Tannehill comes back fully healthy we will still be mediocre if nothing else changes.
     
    danmarino, Puka-head, Fin-O and 2 others like this.
  15. The_Dark_Knight

    The_Dark_Knight Defender of the Truth

    11,817
    10,319
    113
    Nov 24, 2007
    Rockledge, FL
    You know, this word medicore is being tossed around way too much and in my opinion, inaccurately.
     
  16. shamegame13

    shamegame13 Madison & Surtain

    3,451
    903
    113
    Dec 15, 2014
    Talks about drafting a QB at #11 are becoming very real. Reading reports everywhere right now. Looking forward to see who we land to groom behind Tannehill in 2018. Hopefully Baker.

    Mayock believes we may go with Lamar Jackson or Luke Falk in the 2nd round.
     
  17. Dorfdad

    Dorfdad Well-Known Member

    4,052
    2,347
    113
    Dec 9, 2007
    My honest opinion is that he comes back says all the right things tries to hard early on to prove he’s back and tosses a bunch of INT’s in the first few weeks. I believe he plays a bit Gunshy as well. I don’t believe we see progression from 2 years ago. I like Ryan but I rank him as a 10-20th quarterback who can win one or two games but not enough to win a Super Bowl. 9-7 with him at the helm. To win 11+ games he needs major talent at key positions to carry him.
     
    shamegame13 likes this.
  18. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Well.. let's say we had the offense from 2016 and the defense from 2017 and we also assume all the correlation between offense and defense is due to offense affecting defense. In other words, most rosy scenario.

    Dolphins scored 363 points in 2016 when the league average was 364.4 (no other team was closer to league average) so basically 0 standard deviations above the mean, and as pointed out before we were almost exactly 1 standard deviation below the mean on defense in 2017.

    The best-fitting line between points scored and points allowed in 2017 says you decease points allowed by -0.422 for each extra point scored. So if we're projected to be at 0 standard deviations above the mean on offense in 2017 (scoring 347.5 points), that's 67 points more than we actually scored (281), and -0.422*67 = -28.3 which puts us at -0.367 standard deviations below the mean on defense.

    Now use the equation in post #3 with OF = 0 and DF = -0.367 and you get an expected win% of 46%, which out of 16 games is 7.36 games. In other words, even in the most rosy scenario assuming ALL the difference on offense was Tannehill and ALL the correlation between offense and defense is due to offense affecting defense, we'd still most likely have won only 7 games. Not sure about you but I'd call that mediocre.
     
    Last edited: Feb 28, 2018
    Irishman, Pauly, Fin-O and 1 other person like this.
  19. jdang307

    jdang307 Season Ticket Holder Club Member

    39,159
    21,798
    113
    Nov 29, 2007
    San Diego
  20. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    It's exactly how some of the clinicians I work with feel when we talk about how to analyze the data they're collecting. However.. the older clinicians seem to be immune to it now after so many years, so you'll get there too jdang lol!
     
    Irishman and jdang307 like this.
  21. jdang307

    jdang307 Season Ticket Holder Club Member

    39,159
    21,798
    113
    Nov 29, 2007
    San Diego
    What is funny is, if I stuck with my strengths I'd probably be right next to you. It's cliche (since I'm asian) but my strongest skill is/was math (and hard sciences like chem and bio). Ridiculously easy for me through high school, taking a college calculus course in high school (didn't count as a high school course but counted towards my UC requirements etc.). My tiger dad forced me to learn algebra after 2nd grade or I couldn't play outside during the summer. Since I didn't study or do any homework the last two years of high school, I got A's in all the classes that didn't require you actually read something (math, chem, bio) and not that great in things that you had to read (american lit, etc.).

    But for some reason I ran away from it (all siblings did pre-med etc.) and decided law was my career path. To satisfy math breadth in college I took stats and philosophy (it counted towards my math requirements). After 100% in one year of stats without any effort I realized I made a mistake and should have taken more math.

    The "Asians are good at math" is not all that accurate either. Only two out of four of us were good in my family (myself and my oldest sister) and my wife is terrible at it.
     
    Irishman and cbrad like this.
  22. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Yeah.. it's mostly that "tiger mom/dad" culture (at least in East Asian culture) and the massive emphasis on education that gives (east) Asians that reputation.

    I'm half Japanese too so I know that first hand. However, the emphasis on rote learning and the "teacher knows best" attitude over there does NOT help when it comes to research mathematics, which is totally different from math you learn in college unless you do such research in grad school. Rote learning actually hurts and has to be "unlearned" to some degree.

    Still.. getting IN top schools can be done with that tiger mom/dad approach and that's half the battle so I guess it can't be criticized too much.
     
  23. Puka-head

    Puka-head My2nd Fav team:___vs Jets Club Member

    8,605
    6,743
    113
    Nov 25, 2007
    Slightly left of center
    1
















    It's the loneliest number.:smartass:
     
    eltos_lightfoot likes this.
  24. mooseguts

    mooseguts Well-Known Member

    362
    368
    63
    Jan 12, 2018
    The sum of the square roots of any two sides of an isosceles triangle is equal to the square root of the remaining side!
     
    eltos_lightfoot likes this.
  25. inFINSible

    inFINSible Bad ministrator

    1,989
    918
    113
    Nov 26, 2007
    I picked this thread to read first.
    I'm leaving again.
     
    Ohio Fanatic likes this.
  26. danmarino

    danmarino Tua is H1M! Club Member

    15,356
    20,976
    113
    Sep 4, 2014

    Hi



    Bye
     
    cbrad and eltos_lightfoot like this.
  27. 2socks

    2socks Rebuilding Since 1973

    8,141
    2,103
    113
    Nov 27, 2008
    Atlanta
    What happens when in the very near future they trade Jarvis Landry?
    How many years do the Phins double down on the same crap?
    How and when will the Phins dethrown the Patriots?
    Tannenbaum has repeatedly demonstrated he will never lead this team anywhere

    Anyway to factor that into the equations
     
  28. miamiron

    miamiron There's always next year

    2,354
    1,402
    113
    Jan 4, 2008
    [​IMG]
     
  29. 2socks

    2socks Rebuilding Since 1973

    8,141
    2,103
    113
    Nov 27, 2008
    Atlanta
    When will the Gase experiment end. Need to bring in a Wade Phillips type to learn Gase the ropes. Divide the responibility of a coach clearly in over his head, Greer begin the 3 yr process of fixing the roster and then shake up the league.
     
  30. Wilkimania

    Wilkimania Well-Known Member

    1,033
    649
    113
    Sep 11, 2016
    I might be showing my naivety here but how has Gase shown he's in over his head? So far we've had a season where we made the playoffs and a season where our QB was injured throughout the season. From varying sources it sounds like Gase has established that we had a problem with personalities in the locker room and has started making moves to address that.
     
  31. 2socks

    2socks Rebuilding Since 1973

    8,141
    2,103
    113
    Nov 27, 2008
    Atlanta
    There are very few coaches that by themselves could turn this team around in a time acceptable to those watching. The average Head coach gets 3 yrs. Gase is in yr 3 and dismantling the team - the writing is in the wall. Its rebuild time. The smart thing is to bring someone in Defensively who can fix that side of the ball why Gase concentrates on the other. Its not a sign of weakness but a sign of strength and maturity.
     
  32. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Here's some potentially useful information in deciding whether to build around Tannehill or gamble on a QB with a higher ceiling. Assuming that the goal is to win a SB and not just make the playoffs, it's worth looking at the probability a team with a QB playing at a certain level (measured by passer rating in this case) wins the SB.

    Top graph shows a histogram of how passer rating, in standard deviation units above or below the mean, is distributed from 1970-2017. The second graph does the same for the passer rating of the SB winner. The third graph just divides the 2nd graph by the 1st and fits the best-fitting exponential function to the data, giving us an estimate of the percent probability a team with a QB that plays at a certain level will win the SB.[​IMG]

    Everything is in standard deviation units, so we need to look at Tannehill's passer rating in those units.

    2012: -0.83
    2013: -0.36
    2014: 0.41
    2015: -0.18
    2016: 0.37

    So first of all we see that Tannehill's best year was actually 2014, not 2016. Once again, using ranks is misleading because ranks aren't measures (i.e. the difference between rank 1 and 2 isn't a priori the same as between ranks 6 and 7 for example) but standard deviation units (z-scores) are measures.

    Second, we see that for Tannehill's best z-score = 0.41, the percent probability of winning a SB using that equation is P = 3.84% or about 1 in 26 years. Going up to a z-score of 1 still isn't that good with a probability of 6.3% or 1 in 16 years. If you solve that equation for winning a SB once a decade you get a z-score of 1.55.

    For reference, in 2017 a z-score of 1 would be a passer rating of 96.8 and a z-score of 1.55 is a passer rating of 102.2. I think that's a good estimate for the minimum average passer rating you want in a starting QB. It also shows you just how unlikely it is to take a QB that averages less and build a SB winning team around him.


    * One technical note: those passer ratings above are "team" passer ratings that combine both starting QB and any backup QB stats, which in some sense means it's underestimating (only by a little) how good your starting QB should be because backups tend to be worse.
     
  33. hazed819

    hazed819 Well-Known Member

    1,034
    298
    83
    Apr 7, 2010
    Tannehill is at home right now being fed BonBon's by Mr's Tannehill watching this team burn to the ground via all these moves lol.
     

Share This Page