Re: Argument from authority [was: When is Hard SF not Hard SF?

Aqua wrote:

Gerry Quinn wrote:
In article <chotv5-992.ln1@xxxxxxxxxxxxxxxxxxxxxx>,
aqua@xxxxxxxxxxxxxxxx says...

[3] Note that I'm from the "IQ is probably a statistical artifact"
school

Could you elaborate on this surprising notion?

I don't have time personally, but I invite you to read this:
http://www.cscs.umich.edu/~crshalizi/weblog/523.html

Aqua

I am sorry but the author does not appear to me to be saying what you
are claiming him to be saying, and more importantly, this isn't in any
way implied by what he is saying.

Now I have to premise that statistics is by far the part of
mathematics I know the least, but still I know enough that I was able
to follow his arguments. I also want to point out that I have no
particular attachment to the idea of g, given that on the contrary I
am somewhat affectionate to the evolutionary psychologist's claim that
the brain (which for some unknown reason they insist in calling mind)
is made of distinct specialized logical modules.

First of all, g is an artifact, not IQ as you wrote. g is the stuff
that supposedly IQ measures. I suppose this was just a typo.

He is saying that nonzero correlation with g of correlated test
results is implied by construction. There could be several independent and
equally important causes and you would still be able to extract g.
All of this does not, however, mean that IQ, defined operationally, is
meaningless, and Shalizi knows it, in fact he writes:

|A reasonable counter-argument would be something like: "It's true
|that 'intelligence', in the ordinary sense, is a very broad and
|imprecise concept, and it's not surprising the tests don't capture it
|perfectly. But the aspects of 'intelligence' they do capture are ones
|which are vastly more important for economic development than the ones
|displayed by Cowen's friends in San Agustin Oapan, however amiable or
|even admirable those traits might be in their own right." This would
|be a position about which one could have a rational argument. (Indeed,
|I might even agree with that statement,

Second, this is a critique of a line of arguments which he asserts are
prevalent. He is not proving that g is nonsense (nor is he claiming
it), any more than the idea of size of a human being is nonsense in
describing any single measurement of a human body and their
correlations. Rather, he is saying that this is descriptive and not
inferential statistics, and that people who believe in g fail to
understand this. He further says that g fails confirmatory factor
analysis but, unless I have somehow managed to miss it, he does not
show this nor does he link to a proof.

Further, he does not deny nor explains away the presence of all positive
correlations among all diverse tests of cognitive abilities (He does not
prove that "intelligence tests are made to correlate with each other" by
asserting that "new tests are validated by showing that they are highly
correlated with the common factor", particularly as he does not provide
evidence of this), which g certainly describes and allegedly unaided
purports to also explain.

Is he right? I looked for a adequate target for his criticism and,
after a bit of googling, I reached the conclusion that the name of
Arthur Jensen was what I was looking for. So I googled for something
written by him.

http://www.debunker.com/texts/jensen.html

I don't know what to think; he seems to be both guilty

|The g factor may also be termed a theoretical construct, which is
|intended to explain an observable phenomenon, namely, the positive
|intercorrelation among all mental tests, regardless of their
|apparently great variety [...]

|that will eventually explain the nature of g

and also innocent

|The g factor represents this salient fact of nature better than any
|other single factor or any combination of multiple orthogonal factors

|Taken together, the g factor plus smaller group factors (primary
|abilities independent of g) best represent the fact that

|The g factor reflects more of the variance

It would seem (from further googling) than confirmatory factor
analysis did not exist back when Jensen was writing - however it seems
he used another method, namely 'correlated vectors', which many
however find wanting.

He presumably thought he needed not anyway, since he also says

|At an even more basic level, there is now considerable evidence that
|g is correlated with the amplitude, latency, and complexity of average
|devoked potentials in the brain, as measured by means of EEG apparatus
|and electrodes attached to the scalp (e.g., Eysenck, 1981; Jensen,
|Schafer, & Crinella, 1981).

Which if true, it is my understanding, would justify g even on
Shalizi's own grounds.

Bottom line: the essay does nothing to prove that IQ is a statistical
artifact. It does make its case for g as far I can understand, which case
is relevant if, and only if, exploratory factor analysis is the only
justification put forward by the supporters of g. Whether this is true or
not isn't clear to me. (The above quote could be an outlier.)

Interesting site by the way, and interesting person, in substantial part
because he is clearly highly intelligent.


--
Delete the numbers to get a working address.
.

Relevant Pages

  • Re: Legalization: only way to beat "Terrorisms Harvest"
    ... >>These correlations are not significant. ... are you saying that because you don't ... a bunch of prose claiming things that I ... the people who carried the laws forward ...
    (sci.econ)
  • Re: True random numbers
    ... >> There is an enormous amount of confusion about the meaning of the term ... > Consultation of my nearest statistics text then surprised me when I ... > underlying physical theory does not introduce such correlations, ... > any pseudorandom number generation algorithm could be. ...
    (comp.lang.fortran)
  • Re: Presenting unorthodox factor analysis findings
    ... and usually correlations are used. ... the covariances of the factors are constant; ... are those of the Statistics Department or of Purdue University. ... Herman Rubin, Department of Statistics, Purdue University ...
    (sci.stat.edu)
  • Re: PCA C code contradictions
    ... how I advised simple correlations to be used. ... the formula for the REAL correlation matrix! ... Most typos are inconsequential, as the typo in the title above. ... that betray the claim of the misspeller as an expert in statistics. ...
    (sci.stat.math)
  • Re: Arizona is the Dumbist state in the US
    ... > is all the major open issue of Cognitive Psychology and Artificial ... > Correlations and statistics are irrelevant. ... > there has to already be something to validate it AGAINST. ... is that one uses statistics to define (and ...
    (sci.physics)