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

Re: Pro Basketball Prospectus

Expand Messages
  • Dean Oliver
    ... matchups, which ... coefficients, ... regressions that ... weights and I ... solution.... which ... similar ... coefficients that actually ... close to
    Message 1 of 2 , May 26, 2003
    • 0 Attachment
      --- In APBR_analysis@yahoogroups.com, bchaikin@a... wrote:
      > if you can truly reproduce his predicted-wins-to-real-wins
      matchups, which
      > appeared to be very close, but with vastly different parameter
      > both sets should easily be testable...
      > What do you mean by "testable"?
      > you stated earlier that "...I found alternative rational
      regressions that
      > were equally mathematically predictive that yielded very different
      weights and I
      > felt that it implied inconsistency and a poorly formulated
      solution...." which
      > i took to mean you were able to come up with a formula using
      > logic/idealogy as berri but with vastly different parameter
      coefficients that actually
      > came up with predicted wins coming out very close to or just as
      close to
      > those that berri did to the real life team win numbers. we can
      simply look at both
      > his methodology and yours, which you are saying is similar to his,
      and see if
      > indeed some statistical "parlor trick" is the reason behind the
      accuracy or
      > if there is something inherent in both his formulas and yours that
      is real
      > world...

      I was hoping you had some idea of a great test. His original
      evidence that his method was right was that the Bulls got so much
      better in 1996 when Rodman joined the Bulls, yet didn't drop so much
      when Jordan left for 1994 -- that meant that his results showing
      Rodman so much better than Jordan were right. I pointed out that the
      Bulls lost Rodman several times over a couple years and didn't miss
      him, having almost the exact same awesome winning percentage without
      him, but were significantly worse without MJ. That was not a
      sufficient test and I would argue that it is just one of many lines
      of evidence that I could make -- not proof, just lots of reason to

      The other formulations -- there are a few that are really easy.

      He framed his regression as

      Team wins = f(pts, dpts)

      With the results, he built up individual wins. I noted that, just as
      logically, he could have done

      Team win% = f(pts, dpts)

      and, with the results, build up individual win%'s.

      Or, he could have framed his regression as

      Team wins = f(pts/poss, dpts/poss)

      and then regressed pts/poss on different stats. He did that and got
      very different weights on his statistics, though the method was
      equally predictive.

      It didn't matter to me which of these was "right" -- I just saw no
      theoretical basis for any of them being right and, since they were
      all different, it seemed like a weak method.

      > manley/heeren/hoopstats are all great reads....is your book out yet
      and if so
      > what's it called?...

      My book comes out in October. It's called Basketball on Paper and
      has the endorsements of Dean Smith, my contact with the Sonics, and,
      hopefully, a couple of other big names. I'll certainly let people
      know when it's available for pre-order. It's published by the same
      people who do JohnH's book because mine is a monograph, not an annual
      as John's is. Mine is a philosophical-coach's-rule-of-thumb-
      strategical-fan-insight-scientific book with as much humor as I
      possibly can introduce. And, oh yeah, the cover is really cool (a
      friend designed it).

    Your message has been successfully submitted and would be delivered to recipients shortly.