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

Re: [APBR_analysis] Re: All-Time Lakers Team

Expand Messages
  • Michael K. Tamada
    On Mon, 23 Jul 2001, Mike Goodman wrote: [...] ... [...] ... What you want to use is multivariate regression analysis also known as ordinary least squares
    Message 1 of 5 , Jul 23, 2001
    • 0 Attachment
      On Mon, 23 Jul 2001, Mike Goodman wrote:


      > Turnovers can be predicted, too, and the factors are many. The
      > traditional belief is that assists are most closely related to
      > turnovers, but in fact scoring and rebounding are, also.
      > If you have any idea what I am referring to when I talk about
      > standardized rates (and I have posted a few), the turnover formula is
      > this: TO = .08(Sco)+.07(Reb)+.16(Ast)+.05(Stl)+.10(Blk)-.005(MPG).


      > I would love to know how to actually plug in a few hundred player
      > stats and have my computer generate these correlations; all I have
      > managed is to tinker with the numbers until a good average is
      > achieved. After a tinkering, I just check the extremes at either
      > end, trying to minimize.

      What you want to use is "multivariate regression analysis" also known as
      "ordinary least squares regression". I believe that Excel will only do
      univariate regression. There are however freeware regression programs
      available; I don't use any of them because I've got paid-for programs but
      I know they are out there ... I know there was a shareware or freeware
      econometrics program available at Penn State University's website. Also
      there is a package called "R" which is a shareware or freeware version of
      "S", a package widely used by statisticians. However S, and I imagine R,
      are aimed more at theoretical statisticians and people who need to develop
      and program their own statistics, rather than being aimed at users who
      simply want to crunch some numbers using standard techniques.

      The technique you describe is a standard one for filling in missing data;
      i.e. run regressions to come up with equations predicting what a player's
      turnovers per minute will be.

      Obviously the technique becomes shakier as the amount of missing data
      increases, in particular for years prior to 197? when there are NO data at
      all on turnovers. Then you have to make assumptions that the turnover
      equations for, say, 1957, are the same as the ones for 197?-2001. In
      other words, extrapolation is a lot more difficult than interpolation, and
      for years with no turnover data whatsoever, we're extrapolating rather
      than interpolating.

      So the equations should be double-checked by, e.g. looking at
      season-by-season data to see if there are time trends. E.g. I believe
      that offensive rebounding percentages gradually increased during the
      1970s and 1980s. I believe that turnover rates (certainly per minute, and
      possibly relative to scoring, rebounding, etc.) declined in the 1980s and
      1990s. And for sure, field goal percentages rose for decades, until some
      time in the 1990s when they started declining.

      So the equations for predicting turnovers in the "modern" NBA may not work
      for predicting turnovers in the NBA of the 1950s.

      On the bright side, OLS will be much much faster AND lead to better, more
      accurate equations than fiddling around by hand.

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