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

Stringer's Power Rankings and STL @ MIA Preview

Discussion in 'Miami Dolphins Forum' started by Stringer Bell, Oct 10, 2012.

  1. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    This is something that I've been working on for a while now. Its a model based on various efficiency statistics. I think I've reached the point where I've refined it enough to feel comfortable sharing it. Ideally, the model will essentially pick out the "signal" and eliminate the "noise", to determine the actual ability of teams. By no means do I believe this to be the end-all of rankings, or even more accurate than any other rankings. Its just what I've developed based on a lot of research. I will happy to discuss why certain teams are ranked where they are, but unfortunately at this point I won't say exactly what the model is made of.

    The ranking is adjusted for opponent strength. I've included the strength of schedule in the table for your reference. Keep in mind that these rankings are 100% based on the statistical inputs. If a team's ranking doesn't match my perceptions, I don't adjust anything. Having said that...


    [table="class: grid"]
    [tr][td]TEAM[/td][td]SOS[/td][td]RANK[/td][/tr] [tr][td]San Francisco 49ers[/td][td]-0.02[/td][td]1.805[/td][/tr] [tr][td]Houston Texans[/td][td]-0.08[/td][td]1.765[/td][/tr] [tr][td]Minnesota Vikings[/td][td]-0.01[/td][td]1.184[/td][/tr] [tr][td]Denver Broncos[/td][td]0.20[/td][td]1.106[/td][/tr] [tr][td]Chicago Bears[/td][td]-0.14[/td][td]0.802[/td][/tr] [tr][td]Atlanta Falcons[/td][td]-0.05[/td][td]0.767[/td][/tr] [tr][td]Miami Dolphins[/td][td]-0.01[/td][td]0.585[/td][/tr] [tr][td]New England Patriots[/td][td]-0.06[/td][td]0.577[/td][/tr] [tr][td]Detroit Lions[/td][td]0.22[/td][td]0.493[/td][/tr] [tr][td]Baltimore Ravens[/td][td]-0.12[/td][td]0.363[/td][/tr] [tr][td]Green Bay Packers[/td][td]0.09[/td][td]0.130[/td][/tr] [tr][td]Seattle Seahawks[/td][td]-0.07[/td][td]0.122[/td][/tr] [tr][td]St. Louis Rams[/td][td]0.06[/td][td]0.087[/td][/tr] [tr][td]Philadelphia Eagles[/td][td]-0.12[/td][td]0.018[/td][/tr] [tr][td]Cincinnati Bengals[/td][td]-0.12[/td][td]-0.040[/td][/tr] [tr][td]San Diego Chargers[/td][td]-0.19[/td][td]-0.064[/td][/tr] [tr][td]Pittsburgh Steelers[/td][td]-0.02[/td][td]-0.107[/td][/tr] [tr][td]Carolina Panthers[/td][td]-0.14[/td][td]-0.201[/td][/tr] [tr][td]New York Giants[/td][td]-0.26[/td][td]-0.306[/td][/tr] [tr][td]Washington Redskins[/td][td]-0.13[/td][td]-0.307[/td][/tr] [tr][td]Indianapolis Colts[/td][td]0.09[/td][td]-0.337[/td][/tr] [tr][td]Oakland Raiders[/td][td]0.16[/td][td]-0.349[/td][/tr] [tr][td]Arizona Cardinals[/td][td]0.12[/td][td]-0.372[/td][/tr] [tr][td]Dallas Cowboys[/td][td]-0.08[/td][td]-0.441[/td][/tr] [tr][td]Tennessee Titans[/td][td]0.31[/td][td]-0.625[/td][/tr] [tr][td]New York Jets[/td][td]0.20[/td][td]-0.852[/td][/tr] [tr][td]New Orleans Saints[/td][td]-0.13[/td][td]-0.989[/td][/tr] [tr][td]Cleveland Browns[/td][td]-0.09[/td][td]-1.024[/td][/tr] [tr][td]Kansas City Chiefs[/td][td]-0.08[/td][td]-1.155[/td][/tr] [tr][td]Jacksonville Jaguars[/td][td]0.26[/td][td]-1.211[/td][/tr] [tr][td]Buffalo Bills[/td][td]-0.07[/td][td]-1.250[/td][/tr] [tr][td]Tampa Bay Buccaneers[/td][td]-0.13[/td][td]-1.602[/td][/tr][/table]

    Obviously the big thing that stands out is Miami being ranked 8th. They've essentially climbed up the rankings quickly, as they were near the bottom of the league the first couple of weeks. Why are they ranked so high? Miami is not below average in any of the major statistics the model considers. The biggest factor for the rankings is passing efficiency, which shouldn't be a surprise. Miami is slightly above the league average, both offensively and defensively in that area. But aside from that, they are near the top of the league in almost everything else. They are the best run defense in the league. They also are one of the least penalized teams in the league. Believe it or not, that is a somewhat predictive. Also keep in mind that the model observes interceptions to have a high-degree of randomness. As such they aren't heavily weighted. The bottom line is that Miami a very complete team. There are no glaring weaknesses, at least statistically.



    In regards to the game against St. Louis, I believe Miami has a slight edge. The biggest problem for St. Louis is that they just aren't very good passing the ball. In fact, they are near the bottom of the league in offensive passing efficiency. They aren't very good running the ball either. Defensively, they are very good against the pass however, which is what should keep the game close. Ultimately though, I think Miami wins because of its superior passing game (when was the last time you heard that?).

    Please feel free to share any thoughts you may have. I welcome any thoughtful discussion or criticism.
     
    muscle979, Finrunner, Mainge and 36 others like this.
  2. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    How do you determine SoS? Is it organic to account for how teams are currently playing?
     
  3. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Its based on the average rating of a team's opponents. I will take that rating then re-run the model using that average rating as an input. The SOS gets updated as often as the stats are inputted.
     
    Bpk, maynard and Fin D like this.
  4. MonstBlitz

    MonstBlitz Nobody's Fart Catcher

    21,176
    10,130
    113
    Jan 14, 2008
    Hornell, NY
    Always appreciate statistical stuff like this. Will you ever share the exact inputs or is that a secret "recipe?"

    My 2 main critiques -

    1. Depending on how highly you're rating run defense, it doesn't correlate very highly with winning and losing in today's NFL. Definitely not when compared to...

    2. turnovers. Specifically interceptions. TD/INT ratio is very indicative of success. I'm just not sure how you can base this hugely on passing efficiency but not give appropriate weight to interceptions. The better QBs in the league tend to have very good TD/INT ratios so I don't think it's as random as you suggest. When I look at your list, without looking up the stats and going off the top of my head, it looks like the teams that don't belong that high - Miami, Denver, and the Detroit Lions have all had their problems with interceptions this season.
     
    Boik14 and HardKoreXXX like this.
  5. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Let me guess:

    The model consists of statistics that are weighted based on their correlations with wins, and are included in the model only if their variation isn't more random than systematic. ;)
     
  6. xphinfanx

    xphinfanx Stay strong my friends.

    10,823
    2,214
    113
    Nov 1, 2009
    What am i missing you say they are ranked 8th but in the chart I count them as seventh?
     
  7. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    You're missing that a guy who can't count did a statistical analysis.:tongue2:
     
    Boik14, HULKFish, Pandarilla and 5 others like this.
  8. Fin D

    Fin D Sigh

    72,252
    43,684
    113
    Nov 27, 2007
    So then SoS isn't used in determining the ranking?
     
  9. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Secret for now. I'm not sure its really necessary to remain a secret, but its something I've spent a lot of time on, and obviously its not possible to make it a secret once I share the info.

    Correct. It doesn't correlate very highly and its weight reflects that.

    One thing to keep in mind, as I mentioned in the OP, is that I'm most interested in determining a team's ability, not necessarily their results. Turnovers are very highly correlated to winning. The problem is that turnovers can be very random. Past INTs specifically don't correlate very well with future INTs. Miami and Denver are two good examples here. All of Peyton Manning's INTs came in one game. In fact, they all came in one quarter. Is that indicative of his ability, or just bad luck that 3 of his bad passes ended up in opponents hands? As many here have discussed, were Ryan Tannehill's INTs a representation of his ability, or just a matter of bad luck?

    Having said that, INTs are part of the model. They are not entirely random. There is an element of ability that contributes to INTs. But their weight is representative of their randomness.
     
    Boik14 and MonstBlitz like this.
  10. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    It is. I come up with an initial ranking, which you don't see here. Then I factor in SOS, and get the ranking you see in the OP. Sorry for the confusion.
     
    Paul 13 likes this.
  11. xphinfanx

    xphinfanx Stay strong my friends.

    10,823
    2,214
    113
    Nov 1, 2009
    [​IMG]:sidelol:
     
  12. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    LOL this is true. Miami is ranked 7th.
     
    Fin D and xphinfanx like this.
  13. Paul 13

    Paul 13 Chaotic Neutral & Unstable Genius Staff Member

    85,620
    51,681
    113
    Dec 3, 2007
    clearly this is a random statistical anomaly...


    :shifty:

    [​IMG]
     
  14. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    No, there are stats that are more random than systematic included. Anything that shows any systemic relevance is included. If it includes a highly random element to it, it is weighted accordingly.
     
    shouright likes this.
  15. xphinfanx

    xphinfanx Stay strong my friends.

    10,823
    2,214
    113
    Nov 1, 2009
    Ones that stand out to me are the Broncos being as high as they are and the Falcons being unbeaten but ranked at 6th. The Falcons rank seems about right it's par for the competition they have played.. This is the first ranking I've seen with the 49ers on top it's interesting knowing they have lost a game but I agree with it. The Broncos on the other hand just don't seem to be getting enough done in game time so far, speaking strictly as viewing the game not stat wise.

    Thanks for sharing the chart.
     
  16. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Yeah, Denver looked too high to me as well. Keep in mind that they've faced an extremely tough schedule. Despite that schedule though, they are near the top of the league in pass efficiency, offensively and defensively, which is pretty rare.
     
  17. MonstBlitz

    MonstBlitz Nobody's Fart Catcher

    21,176
    10,130
    113
    Jan 14, 2008
    Hornell, NY
    Fair enough. I like it. Will you update it every week? I'll be interested to see how it shifts as the sample sizes grow.
     
    xphinfanx likes this.
  18. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Let me ask you this:

    How strong is the correlation between your statistic and wins, and other than that almost no correlation is perfect, how do you make sense of how far it is from 1.0?
     
  19. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Do you mean backtesting the model against past results? The problem is that I think the correlation would be somewhat misleading because it would be tested against games it was developed on.

    My guess is that no model could be more than 70-75% accurate. There is too much randomness that occurs. Also keep in mind that it doesn't consider extraneous circumstances such as traveling on a short week, injuries, etc.
     
  20. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    Appreciate the work that goes into building these kinds of models. I can speak from firsthand experience.

    But hard for me to buy the Vikings as the 3rd best team in the NFL, or the Dolphins as 7th best ahead of Baltimore and New England.
     
    Pandarilla and Stringer Bell like this.
  21. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    No I mean why is your stat not correlated very strongly with these teams' current records? What sense do you make of that? I know it's a small sample size, but I'm wondering what you think outside of that.
     
  22. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Thats a great question. I looked it up, and the correlation between their rank and win % is 0.700532. I'd attribute the difference that does exist to randomness.
     
    MonstBlitz and shouright like this.
  23. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    What are you trying to predict with this stat? Anything? I think if the correlation is only 0.7, you have to ask yourself if you're getting any incremental predictive validity over and above something like YPA differential, or if in fact the inclusion of other variables is actually decreasing that validity.

    Now, if this has some other magical capability, like predicting who will advance through the playoffs despite their record, then that may be another matter.
     
    Bpk and dolfan32323 like this.
  24. Tone_E

    Tone_E Season Ticket Holder Club Member

    13,777
    7,574
    113
    Dec 8, 2007
    The only way that will work is if you have pre-defined values for each starting position that you can put in a mathematical equation. But then again, an injured starting CB (Revis) differs so much from another injured starting CB (V.Davis). That may introduce human discretion into it, which then skews everything - so it is best to leave that out.

    I've been playing with this idea myself (for non-betting purposes!) but got lost in too many variables and human discretion.
     
  25. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    I think sho brings up a valid question.

    What was the mission statement in developing the model? Is it a predictive model? Is it supposed to be like PFF's ratings system where they claim to be giving a more accurate/detailed glance at how teams have done?
     
  26. MonstBlitz

    MonstBlitz Nobody's Fart Catcher

    21,176
    10,130
    113
    Jan 14, 2008
    Hornell, NY
    It will be interesting to see if the correlation increases as you add more games to the sample size. Is this going to be something that will be easy to put together every week now that you have the model in place?
     
  27. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    For Las Vegas purposes, I've got something I've been toying with that's about 70% predictive this year against the spread through 77 games...which is an insane number. I'm sure it'll go down by the time we get to game #256.
     
    Bpk likes this.
  28. HardKoreXXX

    HardKoreXXX Insensitive to the Touch

    20,459
    14,210
    113
    Apr 2, 2008
    Coral Springs, FL
    Just hard to believe that a team (Miami) -4 in turnover margin could be ranked so high.

    I know your reasoning for it, I just think it should be weighted more.

    Anyways, good work.
     
  29. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    I have a tough time, personally, coming up with these kinds of models. The reason is because I have a tough time deciding what I want this model to do. Is it just supposed to tell me, in a number, what teams have been the best? For what purpose? So that I know who will beat who? If it's not that, what do I need to know more than just the win-loss record, and maybe some basic points data? If I'm trying to get more specific than that, my instinct is to dig deeper and go down the rabbit hole taking a look (visually) at the teams playing football, the players, etc...rather than the opposite, trying to go higher up in the sky and take one unified stat.

    I get more obsessed with predictive models.
     
    Bpk likes this.
  30. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008

    Net YPA-differential correlates with winning percentage this season at 0.637.

    Passer rating differential correlates with winning percentage this season at 0.86.




    I had a whole drawn out response to this. But I'll keep it simple and say that yes, the mission of this model is to be predictive. It is predictive because it measures a team's ability, while eliminating whatever randomness and chance clouding the win-loss records. At least that is the goal.
     
    Bpk likes this.
  31. Stringer Bell

    Stringer Bell Post Hard, Post Often Club Member

    44,356
    22,480
    113
    Mar 22, 2008
    Yes, I will put this up every week as soon as the Monday night game ends.
     
    Mile High Fin, maynard, Bpk and 2 others like this.
  32. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    Correct 70% of the time against the spread? :huh1:
     
  33. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    So you could very well say at the end of the season that one team had more ability than another but had a worse record because it was victimized by more negative, random events, no?

    This sounds like something you could use to save a GM or a head coach's job. ;)
     
  34. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    Yes. I'm tired of betting on the horses. Time to start betting on the bettors.

    Through 77 games it's 69.7 percent weighted, 66.2 percent unweighted (51-26).

    The encouraging thing is that the greater the magnitude of the signal, the greater the chance the pick is right. If you were just to isolate the top 8 picks each week, those picks are 29-11 (72.5 percent) unweighted against the spread. If you isolate the rest of the picks excluding the top 8 signals, they're only 22-15 (59.5 percent) unweighted.
     
    dolfan22 and Bpk like this.
  35. dolfan32323

    dolfan32323 ty xphinfanx

    12,587
    1,574
    113
    Nov 27, 2007
    Washington DC
    Wow. Thats ridiculous. good stuff... let's hope it doesn't fizzle out.
     
  36. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    Apropos of nothing, here's some more numbers for numbers' sake.

    Tebow's playoff game last year: 316 yards passing, 31.6 per attempt, Ben threw a pick on 3rd and 16, and TV rating in last 15 minutes: 31.6.

    After the 666th MNF, Mark Sanchez has 6 TDs, 6 INTs, is averaging 6.6 yards per attempt, has a QB rating of 66.6, and wears No. 6.

    And as @Deadspin noted, Sanchez's longest completion this year is 66 yards
     
  37. shouright

    shouright Banned

    22,845
    8,861
    0
    Dec 13, 2007
    What the hell are you doing here? I'd be out in Vegas with Awsi Dooger! :yes:
     
    Bpk likes this.
  38. miamiron

    miamiron There's always next year

    2,354
    1,402
    113
    Jan 4, 2008
    I'd like to see that 70% plus picks every week BEFORE THE GAMES
    Post them here so we can see
     
  39. RickyNeverInhaled

    RickyNeverInhaled Well-Known Member

    6,771
    1,680
    113
    Mar 15, 2009
    Alabama
    It would be interesting to see if we can pick the winners and losers of next weeks match ups. If that works I might use it in vegas. I think if you double your money 30 times and start with a penny or dollar you will have a million dollars.
     
  40. ckparrothead

    ckparrothead Draft Forum Moderator Luxury Box

    79,599
    159,162
    113
    Dec 1, 2007
    Hah, even Awsi's not in Vegas. He's down local in FL at the moment, last I heard from him. He's kind of scarce lately because of some personal situations.
     

Share This Page