Can we define consistency?

Discussion in 'Pistons and NBA' started by TaS, Oct 23, 2007.

  1. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    After hearing Flip's quotes regarding Jason Maxiell's "inconsistency," I started thinking about how I can test to see whether Flip is right on or can't tell consistency when he sees it. My theory is that he hasn't done any sort of calculations and bases it off of his own perception... which could possibly be faulty.

    Here is my method:

    I decided to filter out those games where players get less than 10 minutes. A lot of this is garbage time and/or the player can't be expected to get into any flow.

    Then, I take the results from each individual game and create 40 minute stats for each game. This is to compensate for the variation in minutes from game to game... because we are trying to isolate inconsistency of the player, not the minutes.

    Then, I calculate the standard deviation (we can call it the "Beta"). I did this for 1) points and for 2) a blended stat line (2R+2A+2S+2B+P), which I will call "Production". This Production gives the player credit for other measurable contributions.

    The lower a player's Beta, the more consistent he is.

    So far, I've only done this for 3 players, Sheed, Webber, and Maxiell. Here are the results:

    Sheed- Each game, his average per 40 minutes = 15.1 points and 43.4 in Production. His Beta is 4.3 and 8.3 respectively.

    Maxiell- his averages = 15.1 points and 40.4 in Production. His Beta is 6.0 and 10.7 for the two stats (not as consistent as Sheed, but averages the same scoring efficiency. Provides almost as much as Sheed overall).

    Webber- his averages = 8.2 points and 44.6 in Production per 40 min. His Beta is 4.6 and 10.5. So, he scores less than Maxiell, but is more consistent in doing do. Overall, their total contributions are equally consistent from night to night (per minutes given).


    My conclusion is that Maxiell is not as consistent as Sheed, but is equally as consistent as Webber. All 3 of these players provided about the same total package when factoring in pts, rebounds, assists, steals, and blocks on a per minute basis.


    Since I have the method down for importing to a spread sheet, I'll run the rest of the players through it later. Don't know how telling it will be, but I'm sort of interested to see how the players rank.
     
  2. 16 Mile

    16 Mile Bench Warmer

    Joined:
    Sep 8, 2005
    Messages:
    1,301
    Likes Received:
    0
    Location:
    NY, NY
    Flip wants 6 sigma consistency. Also, coaches never let stats mess with their perception.

    My prediction, after you finish, Tay will turn out to be the most inconsistent Piston we have, yet he plays the most minutes.
     
  3. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    I was thinking that too. There are obviously things that these players do that I'm not taking into account here, like good defense, low turnovers, high fg%, etc. And some players may not have had suitable backups, so Flip would have been justified to play them more.

    I'm open to the possibility that Flip is able to digest all of this information and comes to an intuitive conclusion that almost optimizes our situation.

    It's just that this is a quantitative way that I could think of to measure consistency.


    I often think that coaches magnify mistakes by younger players in their minds compared to similar mistakes from vets.
     
  4. coynejeremy

    coynejeremy All-Star Administrator 1x Fantasy Champion Forum Donor

    Joined:
    Oct 9, 2005
    Messages:
    6,255
    Likes Received:
    1,158
    Location:
    Georgetown, KY
    Excellent job. I really need to take that college statistics class so that I can do these nifty analyses you guys are doing. Now the question is: does Flip not notice this because he doesn't play him enough? Or does he not notice because he's biased against him? Am I not giving him enough credit? What's the deal Flipper???? :mad2:
     
  5. Slippy

    Slippy All-Star Administrator Forum Donor

    Joined:
    Sep 22, 2005
    Messages:
    4,398
    Likes Received:
    830
    Location:
    Oakland, CA
    Tashawn,
    did i ever disagree with you? If so, I won't anymore.:pound:

    awesome take.
     
  6. BillLaimbeer

    BillLaimbeer All-Star 4x Fantasy Champion Forum Donor

    Joined:
    Sep 7, 2005
    Messages:
    15,598
    Likes Received:
    3,578
    Location:
    Michigan
    Tashawn,

    Your numbers have shown that JMax is more inconsistent than Sheed and Webber. This is the opposite of what you have been arguing. :nerd2:
     
  7. linwood

    linwood All-Star

    Joined:
    Sep 2, 2005
    Messages:
    4,425
    Likes Received:
    1,442
    Gender:
    Male
    Location:
    Seattle, WA
    True Dat.



    How did you decide to weight the Production variable?

    Since consistency is the issue at hand:

    Your results show that Maxiell is equally as consistent as Webber, who is no longer on the team, even at a bargain price.

    His consistency is equal to that of Rasheed Wallace. Sheed is a tremendous player, but one who seems to have issues with consistency.
    It will be interesting to see how it plays out during the season. If Maxiell's beta is lower than Sheed, and his production is the same, will Flip play him?
     
  8. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    I was arguing that Flip probably has nothing to base his statement on besides his perception, which could very well be faulty.

    In my conclusion, I say that Sheed is more consistent than JMax. And JMax is about as consistent as Webber.

    I really don't have an agenda here and didn't know how the calculations would turn out.

    I was just getting a little frustrated that there was so much definitive talk from Flip about a statistic that I figured he was guessing at.
     
  9. FreshPrince22

    FreshPrince22 Bench Warmer

    Joined:
    Sep 21, 2005
    Messages:
    1,375
    Likes Received:
    7
    The problem for Max is that he relies a lot on his teammates to get offensive production. If his teammates don't look to get him involved or if Flip makes him the last option on every play, he's not going to have consistent offensive production. I remember quite a few games where he never seemed to get any touches. This is a trend with young players on this team, we'll see how that changes now that virtually our entire bench is young. Sheed or Webber don't have that problem.

    The thing I need to see from Max is consistent rebounding. If he brings the effort on the defensive glass like he has the last few games, the rest will take care of itself.
     
  10. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    That is something that Ben learned when he was playing ball with his brothers. You don't have to depend on anybody else to rebound the ball. Definitely the easiest way to be consistent.

    On offense, if you miss a couple J's, you might get cut off or yanked.
     
  11. buddahfan

    buddahfan Retired from Forum

    Joined:
    Sep 1, 2005
    Messages:
    5,361
    Likes Received:
    1
    You have to depend on your opponents missing their shots. LOL

    :hoops:
     
  12. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    After researching further, I have refined my method. I need to calculate the coefficient of variation for each player on the team. This equals the standard deviation DIVIDED by the average (and then multiplied by 100 to get a percentage). So this stat will now let us compare consistency between a player who is really productive and one who isn't.

    The problem that I was running into was that guys like Will Blalock have a really small standard deviation BECAUSE they have a low production number. Guys like Sheed had a bigger standard deviation, but only on a gross basis. On a percentage basis, it is really smaller. Capeesh?

    I will post my findings in a separate post in a minute.
     
  13. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    So here is how I rank the 06-07 Pistons in terms of consistency (Coefficient of Variation "CoV"):

    1) Rip Hamilton- 22.9% CoV
    2) Rasheed Wallace- 23.0%
    3) Chauncey Billups- 25.6%
    4) Tashaun Prince- 27.9%
    5) Chris Webber- 29.8%
    6) Antonio Mcdyess- 31.5%
    7) Jason Maxiell- 33.3%
    8) Nazr Mohammed- 36.7%
    9) Dale Davis- 38.0%
    10) Carlos Delfino- 38.5%
    11) Flip Murray- 38.9%
    12) Lindsey Hunter- 50.3%
    13) Will Blalock- 51.9%


    Notice that adjusting to the CoV dropped JMax in the rankings a bit. There are many possible flaws in the analysis that I can think of.

    • correlation between playing time and consistency. Since I give each game equal weighting and adjust the minutes, small sample sizes are exaddurated. So, the lower a players average minutes, the more error will be built into the adjustment. Because of this, I think that it is only really useful to compare players generally. The closer their average mpg, then the more it will tell you.
    • Weighting system. Sprocket Points is a much more complex way to weight the statistical variables. My method is a crude approximation for the major categories, but doesn't look at missed shots, turnovers, fouls, etc. These are all major reasons for inconsistency I would admit.
    So, I guess I will have to refine it further for it to be more meaningful.
     
  14. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    By the way, here is sort of an interesting table. It shows how the Pistons ranked PER 40 minutes PER game in terms of the Production stat I defined (2xReb + 2xAst +2xStl + 2xBlck + Pts).

    This is part of the reason why I realized that my weightings might be whack... or at least might be leaving important information out (like defensive +/-).

    1) Antonio Mcdyess- 46.8
    2) Chris Webber- 44.6
    3) Nazr Mohammed- 44.1
    4) Chauncey Billups- 43.8
    5) Rasheed Wallace- 43.4
    6) Jason Maxiell- 40.4
    7) Rip Hamilton- 39.8
    8) Dale Davis- 38.4
    9) Tayshaun Prince- 35.7
    10) Carlos Delfino- 35.5
    11) Lindsey Hunter- 32.2
    12) Flip Murray- 30.5
    13) Will Blalock- 19.5


    The top 3 are the surpises... along with Rip at 7 and Tay at 9.

    This makes me think that I'm putting too much weight on rebounds.
     
  15. mikhail1973

    mikhail1973 All-Star Administrator 1x Fantasy Champion

    Joined:
    Mar 23, 2007
    Messages:
    9,654
    Likes Received:
    478
    Gender:
    Male
    Location:
    Kiev - Farmington Hills, MI - Santa Clara, CA
    I don't think it is that big of a surprise. Tay is not one of the primary scoring options, he also doesn't get many boards. Rip's strength is scoring, he's also not high on assists and rebounds on average.
     
  16. dba

    dba All-Star Moderator 1x Fantasy Champion Forum Donor

    Joined:
    Jan 24, 2006
    Messages:
    4,213
    Likes Received:
    362
    Location:
    Ann Arbor/Chicago/Walland, TN
    Pretty interesting. The COV ranks the players pretty well I think. Where did you find a source of game by game data that was easy to deal with? (Or maybe you didn't?)


    Probably nonsensical thought...

    I suspect when we think about consistency, what we really mean is something about a floor, a level of performance that the player has to meet or exceed or he isn't consistent. I think what you're doing is cutting out both bad (more than 1 SD below the mean) performances, but also really good ones (1+ SD above). I wonder what would happen if you looked at the share of games in which the player performed at least at 90% of his average level?
     
  17. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    Just copy and pasted from the NBA.com game by games into a spreadsheet and deleted all the unnessesary rows.

    Your 2nd point... I could probably figure that out. 90% might be too demanding. At least 75% would probably be more useful.

    One thing I already calculated was the MEDIAN production per 40 per game. This means that half the games were more productive and half were less productive than this number. So if you had to predict how productive a player would be in the next game, you would go with this number.

    In this ranking system, Nazr actually gets the top spot and Webber falls to 5th.

    This would seem to say that Naz has a few really bad games that tow down his average, and Webber has a few really good games that skew his average upward. I'm sort of baffled by how good Nazr is looking here. I'm looking forward to incorporating all the Sprocket Point data to see how much that lowers him in the rankings. Fouls and turnovers will hurt him, but FG% will help him.


    Just a thought here. Maybe Naz and Webber are both better than we are giving them credit for?
     
  18. dba

    dba All-Star Moderator 1x Fantasy Champion Forum Donor

    Joined:
    Jan 24, 2006
    Messages:
    4,213
    Likes Received:
    362
    Location:
    Ann Arbor/Chicago/Walland, TN
    Nazr comes out ranked #67 of all players using Sprocket Points per minute (i.e., making everybody equal in terms of time played). He is the 21st among centers, putting him in the same general area as Bynum, Bogut, Miller, Dampier, Thomas, Okur, and Haywood.

    Pistons ranked by Sprocket Points per minute come out...

    1 - Johnson
    2 - McDyess
    3 - Webber
    4 - Billups
    5 - Mohammed
    6 - Davis
    7 - Wallace
    8 - Prince
    9 - Delfino
    10- Maxiell
    11- Hamilton
    12- Hunter
    13- Murray
    14- Dupree
    15- Blalock


    Hmmm... Nobody really likes per 40 minute stats or per minute stats since you assume that performance goes up directly in proportion to minutes which is generally not the case. E.g., a 20 MPG guy who scores 12 isn't very likely to score 24 if he played 40. I wonder if a simply transform like the log or something would account for that and give a better basis of comparison for players who play differing numbers of minutes. Might have to think about that.
     
  19. TaS

    TaS All-Star Forum Donor

    Joined:
    Sep 1, 2005
    Messages:
    17,637
    Likes Received:
    2,970
    Gender:
    Male
    Location:
    Chicaaago
    So Webber, Dyess, and Nazr still look pretty good under the sprocket point system... as does Dale Davis.

    Also, Tay and Rip are down the list. It looks like my system spit out roughly the same ranking as the Sprocket system.


    Considering the log scale you mention. I'm wondering if I couldn't just make a scale based on the data set I have.

    For instance, I can sort my list based on minutes for each game and see how the Sprocket points change with the minutes distibution.
     
  20. buddahfan

    buddahfan Retired from Forum

    Joined:
    Sep 1, 2005
    Messages:
    5,361
    Likes Received:
    1
    A player whose minutes per game actually go up from 20 a year to approach 40 a year will more than double their performance.

    Kobe in his second year played 26 mpg and averaged 15 points a game. 5 years later his minutes went up to 41 a game (just under a 60% increase) yet his points went to 30 a game or about 100% increase

    Kobe Bryant Statistics - Basketball-Reference.com

    Captain Quirk (Arenas)

    In his rookie year he averaged 25 minutes and 10 points
    In his fourth year he average 41 minutes (60% increase) and 26 point or a 160% increase.

    Gilbert Arenas Statistics - Basketball-Reference.com

    KG

    In his rookie year he averaged 29 mpg and 10 ppg
    In his second year he averaged 39 and 17 ppg. (An increase of 33% in mpg and 70% in ppg)

    Nowitzki

    In his rookie year he averaged 20 and 8.
    Four years later he was at 39 and 25.
    A 45% increase in minutes and 200 percent increase in ppg.

    Dirk Nowitzki Statistics - Basketball-Reference.com

    I think you will find with young players, especially the better ones that as their minutes go up their points and other stats (areas of their game that they ares strong at) will go more than proportionately than their minutes, not the other way.

    With older players you may be correct going in the opposite direction.

    With mediocre players your analysis probably doesn't apply because mediocre players who are playing 20 mpg rarely if ever get their minutes raised to 40 mpg.

    :hoops:
     

Share This Page