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

Building a winning team

Discussion in 'Miami Dolphins Forum' started by Pauly, May 5, 2017.

  1. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    Now the dust is starting to settle on the draft, I want to look back at the last 2 drafts with a view to what is statistically relevant to building a winning team.

    Firstly I will mainly be looking at passing for the following reasons.
    1) Team passing offence, team passing defense, and the differential between passing offence and defense are correlated to winning. I have collected team data from 2006 to 2016. Because passer rating has steadily been increasing in that span I adjusted the passer rating to the long term trend to be able to properly compare older teams to more recent teams.

    Long term trend average PR = 88.0

    Correlation of team PR made to win%: 0.67
    Standard deviation: 12.04
    Line of best fit: 68 = 0 wins; 108 = 16 wins; +2.5 = +1 win

    Correlation of team PR allowed to win%: -0.53
    Standard deviation: 9.01
    Line of best fit: 102 = 0 wins, 78 = 16 wins; -1.5 = +1 win

    Correlation of PR differential to win%: 0.80
    Standard deviation: 16.07
    Line of best fit: -32 = 0 wins; +32 = 16 wins; +4 = +1 win

    So the most important aspect of improving your statistical chances of winning games is to improve your PR differential, with improving your passing PR being more important than improving your PR defense.

    2) Rushing and rushing defense is not statistically tied to w/l%. A long discussion is here https://thephins.com/threads/rt17-and-the-run-game.88889/ . The TL;DR version is that teams pass to gain yards and score points; teams rush to get opposing defenses to commit more defenders to the run and open up the pass, control the clock, control field position, set up misdirection plays as well as gain yards and score points. So while rushing is important to the game of football the data gets messy because teams use rushing for a wider variety of reasons than simply scoring points.

    3) What effect does a 'great' unit have? I will define great as being more than 1 standard deviation or more than average. My expectation was that if you built a great unit you would strengthen the other side of the ball. PR efficiency is strongly affected by team win probability, as shown on this page http://www.pro-football-reference.com/years/2016/splits.htm.
    0-19% win probability: 64.8 PR
    20-39% win probability: 73.2
    40-59% win probability: 89.5
    60-79% win probability: 99.8
    80-99% win probability: 117.9
    So if you have a great unit the other side of the ball will benefit because they are playing a higher % of their games with a good win probability ad the opponents are playing a higher percentage of their game time in the low win%

    Offenses +1 SD
    Average PR made: 106.9
    Average PR allowed: 87.7
    Average win% 68.7%

    Defenses +1 SD
    Average PR made: 88.8
    Average PR allowed: 74.9
    Average win%: 63.7%

    Interesting the other side of the ball for the great units statistically ended up being average. Maybe this is because of the effects of salary cap resource allocation cancelling out the win probability effects.

    4) So I looked at 'sucky' units as well, with "sucky" defined as being one standard deviation or worse then average

    Defenses -1 SD
    Average PR made: 87.4
    Average PR allowed: 101.9
    Average win %: 35.5%

    Offenses - SD
    Average PR made: 70.0
    Average PR allowed: 92.3
    Average win% 28.7%

    So of the 4 units being great or sucky on one side of the ball didn't have a benefit to the other side of the ball in 3 of the 4 situations, which was contrary to my expectations.. But being sucky on PR offense hurts your PR defense.

    So in conclusion is that the best way to build a winning team is to concentrate on passing offense first, as being bad there hurts you more and being good gives you more benefits, and then to look at building your defense.
     
  2. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Nice data Pauly!

    I think it's real interesting that there's relatively little effect of improving one side of the ball on the other, at least when it comes to passer rating made vs. allowed.

    I do agree with your conclusion about it probably being more beneficial to first improve passing offense, but I think that can be gleaned (as you point out) from the data at the top of your post: PR for offense has higher correlation to win% AND PR for offense has a higher standard deviation, meaning you can improve passer rating differential more easily by improving PR for offense.

    I don't think that 92.3 for PR allowed under "sucky" offenses is statistically significant however, so I don't think the reasoning at the end of your post has the necessary evidence for it. You'd have to use the Central Limit Theorem to calculate what the standard error is for samples of the size you took before you could prove one way or another whether that 92.3 is statistically significant (I'd bet it isn't).
     
  3. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    I wonder if there is a sweet spot for rush attempts that correlates to PR made? Whether its rushes per pass or per game, etc.

    I think it would have to be rushing attempts and not rushing efficiency, yards, etc. I say that because it is important to keep the defense off balance a bit by including rushing into the offense.
     
  4. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,651
    67,546
    113
    Dec 20, 2007
    I feel stupid..
     
    Georgia Fin and Agua like this.
  5. djphinfan

    djphinfan Season Ticket Holder Club Member

    111,651
    67,546
    113
    Dec 20, 2007
    does this mean you would of taken David Njoku in the draft?
     
  6. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    The sweet spot seems to vary team by team, which indicates offense design plays a part in where the sweet spot is. However less than 40% rushing attempts seems to be about where the drop off occurs.
     
    Fin D likes this.
  7. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Personally, I was happy with a draft that focused on defense in the early rounds. From a stats perspective, Dolphins ended up 8th in overall passer rating as a team (Tannehill + Moore) and in general it's harder to improve when you're near the top than when you're near the bottom. We were much lower ranked in other stats (passer rating allowed was tied for #16, and our run defense rankings were pathetic!!).

    And while it's true that the correlation between rushing efficiency and winning is real low, all that really means in this case is there isn't an overall trend of improving rushing Y/A offense or defense and winning. Still.. it's relatively rare for a team with a bad rush Y/A defense to go deep in the playoffs. Yes, it's happened, but normally rush defense for top teams is nicely distributed around the mean, and we were LAST in this category last year.

    Yeah.. happy about the focus on defense in the draft and especially about picking Raekwon!
     
    Surfs Up 99 and Bpk like this.
  8. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    Going back to the OP.
    What we saw last year was a commitment to building the offense and they waited a year to build the defense.
    By statistical analysis this is the right way to build a team. Also by selecting Harris they prioritized the pass defense over the run defense, which again is statistically the sound thing to do.

    The problem with statistics and he rush game is that rushing, both in accumulated totals and efficiency, is used to influence many things. So the stats don't really tell you the full story. For example Lamar Miller average about 5 ypc in Miami, but he got a boost to his efficiency stats because Lazor underused the run so badly that opposing defenses weren't really trying to stop the run. When he goes to Houston and his ypc drops to about 4. Lamar Miller didn't change, what changed was that opposing Ds knew Houston was going to rush and committed more resources to stopping him because Brock Osweiler.
    What appears important though is devoting sufficient resources to rushing/stopping the rush so that passing offense/defense can operate at a sweet spot. Its hard to tell exactly, but if you can rush around 45% of the time and gain 4.5 ypc you're forcing the opposing D to commit defenders to stay home against the run. On run defense if you can get ypc down to under 4 without affecting you passing D you're in a sweet zone.
    There doesn't seem to be any particular benefit to being great at rushing or stopping the run unless you are also,getting gains in the passing game.

    What the phins are attempting to do in the draft and FA over the last 2 years makes a lot of sense. What needs to be seen is whether the talent evaluation and the ability of coaches to develop the acquired players is up to a high enough standard.
     
    Last edited: May 7, 2017
    Fin D likes this.
  9. Dolphin North

    Dolphin North Well-Known Member

    366
    387
    63
    Apr 30, 2017
    I would hesitate to use pass rating as a basis to build my team, even theoretically as we are doing here. I think passer rating itself is flawed because it favors short safe passes. Short safe passes do not always equate to scoring and winning, or even first downs in some cases. This could be because of how teams defend a dink and dunk team. Ultimately, they can sit at the first down markers, or wherever your team's passing identity or "sweet spot" is and let you pass underneath, constantly making tackles or "safely" contesting those passes enough to end your 12 play drive at some point, even if it is in the red zone where teams can more aggressively defend your short pass offense without the risk of losing a receiver for a deep pass behind them. That is if the offense managed not to make a mistake and stall the drive earlier, which is usually the case. Defending like this also pushes the offense to make more third downs than you want to have to make, while again ensuring your big armed QB does not hit one over your head very often. In other words, you could be managed into a decent passer rating, but yet another 24-21 type loss. Nobody is consistently winning with 15 play drives. At some point they break some big plays and do it with enough consistency to win. So while you don't aim to be the old Al Davis Raiders, constantly throwing it up for grabs and expecting your HOF receivers to go get it, you need the ability to hit some big plays each game. The expectation is there with some teams and those are probably your most successful teams. I say all of this being well aware that the Patriots won the SB a few years ago while not attempting anything over 5 yards, but as cheaters often do, they were the exception to the rule.

    Not to criticize the original post, it was really well done. I probably just have a lesser view of the importance of PR.
     
    Last edited: May 6, 2017
    Fin D likes this.
  10. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    I understand the limitations of the PR. The reason i use it is because
    1) It has a strong correlation to win%.
    2) it uses a small number of moving parts attempts, completions, yards, TDs and Ints.

    I don't like models that use too many moving parts, because I have found simpler models are more robust and the more complex the model the more it tells you about the assumptions of the model maker than it does about the reality.

    The only other method of rating passing efficiency I respect is yards per attempt and/or adjusted ypa, but I found when doing some analysis last year that ypa or a/ypa doesn't have increased correlation to winning over passer rating, so I have stayed with PR simply because it is the more well known measure.
     
    Last edited: May 6, 2017
    Dolphin North and Fin D like this.
  11. Dolphin North

    Dolphin North Well-Known Member

    366
    387
    63
    Apr 30, 2017
    Yes, PR certainly is popular. In any event, I certainly enjoyed reading your post. You put a lot of good work into it!
     
    Pauly likes this.
  12. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    I think there is a human preference for 'hero-ball' in any sport. It's more dramatic and it makes for better stories. But the numbers show that being efficient is more condusive to winning.

    Having a QB stand deep and throw a 50 yard bomb is great entertainment and we want it to be the important event. What we overlook is the strings of 3 and outs that happen in between the heroic plays. To put it another way you have to be on the field to have a chance of making the big play. So a QB who has a high completion% making 5-10 yard plays and makes a few 20 yard plays might well be more likely to win than the QB who takes heroic shots downfield but gets more 3 and outs and interceptions. The hero ball QB will have games where he 'carries the team on his back' by making a cluster of low percentage plays in one game, but in statistics random distribution does not mean even distribution. I.E. for every game the hero ball QB wins there maybe Another 1 or 2 or 3 he loses because the high risks didn't pay off.

    Ideally you want someone who is efficient and heroic.
     
    Fin D likes this.
  13. cdz12250

    cdz12250 Season Ticket Holder Club Member

    10,265
    7,907
    113
    Nov 28, 2007
    Coconut Grove
    Last season, the Bills were atop the league with an average gain per rush of 5.3 yards. The Browns were second with 4.9.

    In yards per pass attempt, the Falcons were first with 8.9, and the Pats were tied for second with 7.8.

    Not hard to see what wins games.


    Sent from my iPad using Tapatalk Pro
     
    Dolphin North likes this.
  14. Finster

    Finster Finsterious Finologist

    3,087
    2,038
    113
    Jul 27, 2013
    This exercise should be done with playoff teams only, because if we're talking about having a winning team, the only wins that really matter are playoff wins, so getting an idea of what it takes to get to and win in the playoffs is the real key to a winning team.
     
  15. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    I posted these before, but they're ever-useful (red circle is SB winner, green circles are playoff teams and blue diamond is league average):

    http://img.pixady.com/2017/04/980661_pass_playoff.png
    http://img.pixady.com/2017/04/964903_opp_pass_playoff.png
    http://img.pixady.com/2017/04/336674_rush_playoff.png
    http://img.pixady.com/2017/04/881322_opp_rush_playoff.png

    You can see that what correlates with getting into the playoffs as well as winning the SB is efficient passing offense and efficient passing defense (first 2 links). Those links only look at Y/A but the situation is similar with passer rating. Efficiency in rush offense and defense aren't as important (last 2 links).

    As far as what it takes to improve most, you want to look at all teams not just playoff teams.
     
  16. Finster

    Finster Finsterious Finologist

    3,087
    2,038
    113
    Jul 27, 2013
    I disagree on that last sentence, it doesn't help to add how bad teams are lucking into wins, as there's no continuity, you look at what works and go with that.
     
  17. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    "Improvement" refers to a difference between two values. You can't tell what helps you improve most if you don't look at differences. So if you're a bad team, you really really should include bad teams as well as good teams in your analysis or you won't have a good idea what helps bad teams improve.

    For example.. if you only knew the stats of the SB winning team but you had NO idea what any other team's stats were, how could you tell what helps you improve most? You wouldn't even know the range over which improvement is generally seen, nor would you know how much different variables help.

    I think the only justification for looking at only playoff teams to see what helps you improve most is if you're already a playoff team and you want to get better.

    btw.. correlations change when you look at subsets of data (e.g. playoff teams only) vs. all teams. So in general you really should use all the available data to see what the overall trend is.
     
    Pauly likes this.
  18. Finster

    Finster Finsterious Finologist

    3,087
    2,038
    113
    Jul 27, 2013
    Or you could just learn from the teams that do it right, bad teams give you inconsistent data, which kind of makes it counter productive.

    We can agree to disagree.
     
  19. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    For playoff teams:
    The average PR (adjusted to long term trend 2006-2016)
    Offense: 98..2
    Defense: 86.2

    So playoff teams have significantly stronger than average PR offense and slightly better than average PR defense.
     
  20. Pauly

    Pauly Season Ticket Holder

    3,696
    3,743
    113
    Nov 29, 2007
    I\m not sure how to calculate the statistical significance cbrad,but here are some numbers that may help.
    Total teams: 352 (11 seasons each with 32 teams)

    Great Offenses: 69
    Playoff teams: 55
    Playoff % with great O: 79.7%

    Sucky Offenses: 39
    Playoff teams: 2
    Playoff % with sucky O: 5.1%

    Great Defenses: 34
    playoff teams: 23
    playoff% with great D: 67.6%

    Sucky Defenses: 84
    playoff teams:12
    playoff% with sucky D: 14.3%

    To go to the next level, I'll look at teams +/- 2 standard deviations from overage. I"ll call +2 SDs "Elites" and -2 SDs "Putrids", although I was tempted to call them Dons and Cams
    Elite Offenses
    Teams: 16
    Play off teams: 16
    Playoff team%: 100.0%
    Adjusted PR Made: 116.6
    Adjusted PRM Allowed: 88.5
    Win %: 78.1%

    Elite Defenses
    Teams: 7
    Play off teams: 5
    Playoff team%: 71.4%
    Adjusted PR Made: 87.8
    Adjusted PRM Allowed: 68.6
    Win %: 69.8%

    Putrid Offenses
    Teams: 8
    Play off teams: 0
    Playoff team%: 0.00%
    Adjusted PR Made: 62.0
    Adjusted PRM Allowed: 92.5
    Win %: 22.7%

    Putrid Defenses
    Teams: 14
    Play off teams: 1
    Playoff team%: 7.1%
    Adjusted PR Made: 116.6
    Adjusted PRM Allowed: 88.5
    Win %: 30.4%

    This was a little surprising because I expected that with a smaller number of teams that random sampling wold cause a bit more variation in the average on the other side of the ball with the +/-2 SD teams compared to the +/- 1 SD teams. However the pattern of being bad or good on one side of the ball not affecting the outcomes on the other side of the ball apart from bad offenses holds at the same level.

    At this point I would like to make a special comment about Rex Ryan and the 2009 Jets.
    This team had a defense that was 2.3 Standard deviations below the average but Rex Ryan hobbled it by coupling it to a passing offence that was 1.8 standard deviations below the average. So if my reasoning is correct then that Jets Defense was penalized by 4 to 5 PR points by this handcuffing. Ryan sabotaged what should have been an all time great defense by trying to make it carry Mark Sanchez. With common garden variety bad QBing the Jets should have gotten to about 13-3, and had homefield advantage throughout the playoffs, and with average QBing people would be putting them up beside the 1985 Bears or the 2000 Ravens.

    The 2 teams with Elite Defenses that failed to make the playoffs were the 2016 Denver Broncos and the 2009 Buffalo Bills
    The Team with a Putrid Defense that made the playoffs was the 2016 Detroit Lions
     
    Last edited: May 8, 2017
    cbrad likes this.
  21. cbrad

    cbrad .

    10,659
    12,657
    113
    Dec 21, 2014
    Regarding how to calculate statistical significance in cases like your original post where you are calculating the mean of some condition, you have to ask whether something is at least 2 times the "standard error" away from the population mean. The formula for standard error = (population standard deviation) / (square root of the number of samples).

    So, for example.. your overall population standard deviation for PR allowed from your original post is 9.01. Divide that by the square root of 39 and you get 1.4428, which is your standard error.

    The comment I made in my first post in this thread where I said I'd bet that 92.3 average passer rating allowed for sucky offenses is probably not significant can now be tested. Is 92.3 at least 2*(1.4428) away from the overall mean of 88? Yes, so the 92.3 is actually statistically significant. All has to do with sample size. Didn't know you had that many samples!

    Remember.. this approach is to only be taken if you're interested in whether the MEAN of a sample from a population is statistically significant.
     
    Pauly likes this.

Share This Page