j
(but not jerry ;-)
sent from phone. please excuse tipos or bad autoconnect.
On Jan 12, 2012, at 2:49 PM, Wojtek S <wsoko52 at gmail.com> wrote:
> Jerry F: "but it doesn't help me see the statistical flaws."
>
> [WS:] Try Stephen Jay Gould "The Mismeasure of Man" he is pretty good
> at explaining the fraudulent use of statistics behind IQ testing and
> its derivatives - "scholastic aptitude" and similar bullshit. And he
> knows the science.
>
> In a nutshell, Gould argues that statistical techniques, no matter
> how sophisticated they look, can only go as far as the assumptions
> underlying the calculations, and those assumptions are often
> questionable or outright fraudulent. He quotes the example of factor
> analysis which is a regression-based technique of estimating "latent"
> factors or commonalities among multiple empirical indicators.
> Computationally, the problem is expressed as a system of multiple
> equations of the form y=a+bx+e where x is empirically observed and
> other parameters are estimated i.e. y is the "latent" factor a is the
> intercept, b is regression coefficient and e is the error term. In
> order to be mathematically solvable, certain assumptions must be made
> about the parameters to be estimated i.e. y's and e's which in
> practice means specifying how many factors you want there to be. For
> example, if you have 30 empirical indicators, you tell the machine to
> find k number of latent factors where k ranges from 1 to 30, and the
> machine will find them no matter what because it was programmed to do
> so.
>
> The fraud part comes in the interpretation of the machine output. The
> IQ fraudsters tell the machine to find a single factor (general
> intelligence) because they a priori believe that human intelligence is
> uni-dimensional and as such can be hierarchically arranged. That it
> its main political advantage - it scan be used as
> pseudo-scientifically justified measure of "moral worth" or "merits."
> The IQ fraudsters use the results as a proof that such a single factor
> exists as supposedly "proved" by statistics. In reality, however,
> statistics do not prove that a single factor exists - they only show a
> solution for one factor because only was factor was assumed, but they
> can show a solution for any number of factors as long as the number of
> factors is smaller than the number of variables in the equation.
>
> In substantive terms it means that human IQ or "aptitude" or "merits"
> or any kindred bullshit measured by the testing-industrial complex can
> be assumed to be either one dimensional or multi-dimensional and the
> number of dimensions depends on what the researchers want it to be. A
> single dimension or may be two dimensions (verbal vs. quantitative)
> are preferred, because they can be hierarchically arranged and thus
> serve as "scientific" indicator of an individual moral worth.
> Multiple dimensions (e.g. verbal, quantitative, emotional, social,
> psychomotor, visual, aural, etc.) do not yield themselves to
> hierarchical arrangements. If one person is high on quantitative but
> low on verbal and emotional, and another high on emotional and social
> but low on quantitative - who is "better" or more "meritorious"?
>
>
> Another noteworthy point that Gould makes is that the real danger of
> this "measurement of man" is posed not by outright fraudsters who
> "cook their numbers" but by reputable scientists who use correct
> statistical procedures but interpret them in way that is supportive of
> their hidden agenda. Cooking the numbers can be easily exposed, but
> biased interpretation of mathematically correct results generally
> passes for bone fide science.
>
> It follows that the way to expose fraud behind measures of human worth
> is not to be distracted by the statistical technique, which in most
> cases passes the smell test, but go for the jugular of the people who
> produced them, and expose their implicit assumptions and agendas,
> options they did not consider, etc.
>
> Wojtek
> ___________________________________
> http://mailman.lbo-talk.org/mailman/listinfo/lbo-talk