Discussion: View Thread

Data Validity (was Re: NBA Player Heights)

  • 1.  Data Validity (was Re: NBA Player Heights)

    Posted 05-03-2000 05:57
    Ok as I see it, the issue at hand is to transform guano into fertilizer...
    (poor data into useful information)

    Agree most likely our data will have particular biases and errors incorporated
    into it.
    Two questions surface in my mind.
    - "What manipulations and processes can I perform to eliminate the biases and
    errors?"
    -"What steps are required to get more overall precision and reliability without
    increasing the precision of the individual components?"

    If we can construct tools to provide higher precision than the components used
    and eliminate the flaws contributed by individual components in mechanical
    systems, why can't we create useful appraisal processes given the existing flaws
    of the measurement tools?

    Saludos

    Esteban

    -------------------------------------
    a veneer

    I think if we were to talk about this further we'd have to go into all
    kinds of things that are rather esoteric, like measurement error,
    and so forth, which also have statistical components.

    The NBA example was, as you said, an easy one, and purer to
    illustrate the fallacy of normal distributions. But the principle, as far
    as I can figure is the same. In fact it's moreso.

    If you have an unreliable measurement instrument, or one of
    questionable validity, ANY use of math or statistics is going to give
    you bad results, GIGO. Which is why, in fact it is virtually pointless
    to mathematically or statistically manipulate (operate on) appraisal
    data, unless it is measures more precisely.

    In a sense though the issue in practice wouldn't be the normality of
    the curve (which most of us mortals aren't going to have the
    resources to test in real life(, but whether ANY mathematical
    functions should be used on poor data.

    So, you have two issues. Can we assume a normal distribution
    without verfying (No) and Is our data solid enough to use statistical
    functions on them (often no).

    ...but the reason people continue to do that and pretend the data is
    good is that it provides a veneer of scientificiousity to the process,
    which of course employees recognize as complete guano.


    Visit the work911.com supersite at http://www.work911.com
    for work related articles, or to find almost anything including
    book reviews and suggestions, discussion lists and more.