[ExI] The downsides of high IQ

rex rex at nosyntax.net
Thu Apr 16 20:04:30 UTC 2015

Anders Sandberg <anders at aleph.se> [2015-04-16 08:41]:
>   William Flynn Wallace <foozler83 at gmail.com> , 16/4/2015 4:38 PM:
>     Don't bet the house on any three point difference in IQ.  It is well
>     within the range of error.  In fact, if retested, the percentage of
>     people whose scores would change by more than three points is very
>     high.  The standard error of measurement for the Weschler IQ test is
>     about three.  However, that is an average.  In fact, as the score
>     departs more and more from 100 the error gets greater and greater.
>     Three IQ points is trivial and is likely not to be the cause of anything
>     other than test unreliability. Bill W
>   You are missing the point. Individual IQ may not be that sharply defined,
>   but that link talked about a population mean. 


>   The bigger problem with the claim that a 3% increase has a huge effect is
>   simply that extrapolating linearly far from the current mean will not
>   work. Obviously a 10 point increase will not reduce high school drop-outs
>   to zero: the numbers just give a bit of evidence of the sensitivity.

But it's not a claim; it's a count. That is, there's no theory or prediction
involved. Instead, H&M counted the fraction of say, HS dropouts in the IQ 97,
IQ 100, and IQ 103 groups, and found the percentage changes reported.

>   Except that untangling causation is *tough*: these are domains where
>   feedbacks go both ways, especially when analysing "national IQs". 
>   But I think nearly all of the scientific literature supports that if
>   something could give on average a few extra IQ points it would have a
>   measurable positive effect. Not necessarily earth-shattering, but still
>   significant. And the tail effects are fascinating: a small boost would
>   increase the number of 140+ geniuses enormously, with high variance
>   effects depending on what they do. 

Yes. Correlation is not causation, and some people will be winners and
others losers regardless of the mean IQ, so the changes in the counts
that would result from actually altering the group IQ would be less
extreme than the observed counts were. Nevertheless, as you say, the
effects would be significant and are consistent with other research on
the effects of group IQ on social outcomes. 

And yes, the tail effects of "small" differences in the SD of Gaussian
distributions are astonishingly large. The URL below has a graph of the
IQ ratio of two populations which have slightly differing means and SDs.
Near the mean, the ratio is about 1:1, but it grows with IQ, and at IQ 170
there are more than 15 times as many of one group in the high-IQ fraction.

This is a statistical property of normal distributions that has
nothing to do with the particular trait being measured.


Bottom line, if groups differ in mean and SD of IQ, and if some occupations
select for high IQ, then disproportionate representation is expected to occur
even in the absence of discrimination.

"...I paid a visit to Schrodinger in his Vienna apartment before his death...
There were no cats. I was told he did not like cats." -quantam leaps,

More information about the extropy-chat mailing list