Loading ...
Sorry, an error occurred while loading the content.

Re: USAToday giving statistics a bad name . . .

Expand Messages
  • Dean Oliver
    ... performance ... available ... Funny, I started with football back in the mid 70 s. When I saw James stuff on baseball in 84, I was impressed with how
    Message 1 of 20 , Oct 29, 2002
    • 0 Attachment
      --- In APBR_analysis@y..., "danthestatman2002" <danthestatman@h...>
      wrote:
      >
      > Thanks Mike....
      >
      > I'm not new to this - I've been attempting to find player
      performance
      > values since I was quite young (like 11 - I'm 31 now). Of course -
      > like many - I did this in baseball, thanks to Bill James. I've
      > turned to basketball and most recently football because, honestly,
      > there isn't that much decent statistical analysis out there
      available
      > to the public. There is a TON for baseball.

      Funny, I started with football back in the mid 70's. When I saw
      James' stuff on baseball in '84, I was impressed with how well it all
      linked together. I realized pretty quickly that basketball had a
      nice structure as well. Football was waaay too hard (though I have
      since mentioned a structure I like to both Pete Palmer and Sean
      Lahman and they both have data to do what I want to do). In '87, I
      scored my first basketball game and that led me down the apparently
      demonic path of points per possession (though it's actually been
      around 50 years in the coaching profession, as Dean Smith did it at
      UNC) and away from linear weights.

      >
      > So, I work on my stuff - and ALWAYS assume I can make it a little
      > better at indicating player value in any given season - given the
      > limitations of the stats we have to draw from.

      There are definite limitations in the #'s, especially prior to 1978
      and the advent of individual turnovers. I see hugely varying
      turnover rates among individuals who play the same position. And
      those really make a big difference in how effective they appear to
      be. Nate Archibald's turnover rate didn't appear to be a stable
      thing and probably varied. He obviously played well, scored a lot of
      points, but if he was turning the ball over a ton, that would explain
      partially why his team didn't win. MJ was so great in part because
      he didn't turn the ball over while using a ton of possessions
      (something MikeG has also pointed out, using a comparison
      with "expected" turnovers as he defined it). I do have a few tricks
      up my sleeve for estimating turnovers from way back when, but the
      uncertainty is pretty big.

      Regardless, making changes in methodology based upon players pre-1978
      is risky. Just too much uncertainty in what the non-measured stats
      were.

      DeanO
    • John Hollinger
      Personally, I found single-number rating extremely handy for the studies in the book. In particular, the Detroit study (searching for fluke years) and the
      Message 2 of 20 , Nov 1, 2002
      • 0 Attachment
        Personally, I found single-number rating extremely handy for the
        studies in the book. In particular, the Detroit study (searching for
        fluke years) and the Indiana study (comparing playoff performance)
        would have been close to impossible without the numerical comparison.



        --- In APBR_analysis@y..., "Dean Oliver" <deano@r...> wrote:
        > --- In APBR_analysis@y..., "Mike G" <msg_53@h...> wrote:
        >
        > > > > Realizing that their own system was not passing the laugh
        test
        > > > ... started taking logarithms of the financial variables before
        > > > calculating z-scores,...
        > >
        > > A logarithm isn't exactly a root is it? The math part of my
        brain
        > > was destroyed in an experiment. I know a log and a root are both
        > > parts of a tree...
        > >
        >
        > A log is different. The log of 10 is 1. The log of 100 is 2. The
        > log of 1000 is 3. (All base 10 log).
        >
        > >
        > > > I read US News' rankings of schools and I read MikeG's rankings
        > of
        > > > players. They are entertaining and they are good for some
        > > > trashtalking ...
        > >
        > > Well, you have the option of talking at the trash level. In the
        > end,
        > > we may be working together toward something pretty sound.
        > >
        >
        > How can we know what is sound? What objective measure will we use
        to
        > say that this linear weights method is "pretty sound"? Is it, as
        > Kevin P said, just making sure that Shaq is #1 among today's
        players?
        >
        > > I like some of what danthestatman has done. Some of it is
        > identical
        >
        > I guess the question is Why would you use something he has done?
        Do
        > you like "some" of what he's done only so far as it's the same as
        > what you've done? Why don't people just use JohnH's weights? Or
        > Doug Steele's weights? What have people learned by looking at
        other
        > people's weights? Why does anyone adopt other people's ratings?
        >
        > > Finally, I think single-number player-ranking is pretty handy
        when
        > > looking at the course of a players's career. Whether or not his
        > > scoring is inflated (for example), you can still see how his
        > playoffs
        > > fared relative to his season, or how one season compares to
        another.
        >
        > Bill James listed a few times when single ratings are handy. I
        don't
        > recall what they were, other than basic trade analysis and first
        cut
        > analysis of player evaluation. Anyone else remember?
        >
        > DeanO
      • Dean Oliver
        ... for ... comparison. I guess I always use individual offensive and defensive ratings, as well as a player s percentage of the team offense. (%age of team
        Message 3 of 20 , Nov 1, 2002
        • 0 Attachment
          --- In APBR_analysis@y..., "John Hollinger" <alleyoop2@y...> wrote:
          > Personally, I found single-number rating extremely handy for the
          > studies in the book. In particular, the Detroit study (searching
          for
          > fluke years) and the Indiana study (comparing playoff performance)
          > would have been close to impossible without the numerical
          comparison.

          I guess I always use individual offensive and defensive ratings, as
          well as a player's percentage of the team offense. (%age of team
          defense is harder to estimate, so I don't use that).

          As a screen, one number gives you a sense and you can do stats on
          it. I guess I like to know reasons. How can you change and predict
          it? The stats from one number can help make predictions (sometimes
          better than with multiple #'s because you don't rationalize a
          number), but it doesn't help you _change_ anything, which is how I
          like to use my stuff.

          Didn't get an answer, though, to how people use each other's weights
          or what they've learned from them.

          DeanO

          >
          >
          >
          > --- In APBR_analysis@y..., "Dean Oliver" <deano@r...> wrote:
          > > --- In APBR_analysis@y..., "Mike G" <msg_53@h...> wrote:
          > >
          > > > > > Realizing that their own system was not passing the laugh
          > test
          > > > > ... started taking logarithms of the financial variables
          before
          > > > > calculating z-scores,...
          > > >
          > > > A logarithm isn't exactly a root is it? The math part of my
          > brain
          > > > was destroyed in an experiment. I know a log and a root are
          both
          > > > parts of a tree...
          > > >
          > >
          > > A log is different. The log of 10 is 1. The log of 100 is 2.
          The
          > > log of 1000 is 3. (All base 10 log).
          > >
          > > >
          > > > > I read US News' rankings of schools and I read MikeG's
          rankings
          > > of
          > > > > players. They are entertaining and they are good for some
          > > > > trashtalking ...
          > > >
          > > > Well, you have the option of talking at the trash level. In
          the
          > > end,
          > > > we may be working together toward something pretty sound.
          > > >
          > >
          > > How can we know what is sound? What objective measure will we
          use
          > to
          > > say that this linear weights method is "pretty sound"? Is it, as
          > > Kevin P said, just making sure that Shaq is #1 among today's
          > players?
          > >
          > > > I like some of what danthestatman has done. Some of it is
          > > identical
          > >
          > > I guess the question is Why would you use something he has done?
          > Do
          > > you like "some" of what he's done only so far as it's the same as
          > > what you've done? Why don't people just use JohnH's weights? Or
          > > Doug Steele's weights? What have people learned by looking at
          > other
          > > people's weights? Why does anyone adopt other people's ratings?
          > >
          > > > Finally, I think single-number player-ranking is pretty handy
          > when
          > > > looking at the course of a players's career. Whether or not
          his
          > > > scoring is inflated (for example), you can still see how his
          > > playoffs
          > > > fared relative to his season, or how one season compares to
          > another.
          > >
          > > Bill James listed a few times when single ratings are handy. I
          > don't
          > > recall what they were, other than basic trade analysis and first
          > cut
          > > analysis of player evaluation. Anyone else remember?
          > >
          > > DeanO
        Your message has been successfully submitted and would be delivered to recipients shortly.