\doc\web\99\17\satiq.txt
From:           	Lostrow@aol
Date sent:      	Thu, 9 Dec 1999 23:30:09 EST


    The g-loading of the SAT is indeed extremely high, .7, but the g-loading 
of the WISC is marginally higher, .8, for several reasons, one of which was 
given by La Griffe:  the theoretically greater number of test-takers of the 
WISC (in fact, the renorming every decade or so depends on sampling).  IQ 
results are reported as a discrete number with a confidence range of a few 
points either way.  That is one reason for use of a range rather than one 
number.  There are other reasons as well (for example, all test-takers do not 
do equally well on the non-cognitive "contaminents" in a particular test) , 
but few people find them as interesting as I do.  The best work in this area 
remains Arthur Jensen's 1980 work, Bias in Mental Testing, for those whose 
eyes have not yet glazed over.


    Before the renorming in 1995, the SAT was important for another reason.  
The pre-renormed SAT was the best cost-effective instrument available to 
detect high IQs ( up to about 156-160, represented by the top verbal SAT 
score, 800).  The WISC and Stanford-Binet are frequently inadequate to test 
people whose IQs are above 130-132 and who have uneven subtest scores, some 
of which hit the test ceiling.


    The renorming, added to the slight dumbing down of the SAT from the mid 
60's onward, has resulted in the renormed SAT verbal test ceiling cutting off 
at roughly IQ 135-140.  What was a mid-1960's score of 700 (verbal), after 
renorming, is probably above 800, the current top score.  The SAT's standard 
deviation has increased since the 1970's -- it is now 110, reflecting the 
wider range of IQs of current test-takers.  The College Board states that the 
renorming difference is not more than 70-80 points, but anecdotal evidence 
does not support this.
------------------------------------------------------------------------------
----------------------------------------------


 <There, I put forward the Principle of Corresponding Order, from which I
quote:


"If a large group of people takes a set of tests, such that each test
contains at least one part dependent on cognitive ability, then the
test takers will finish in the same rank order on every test. 
La Griffe calls this the Principle of Corresponding Order. It follows
directly if we assume that noncognitive parts of an exam will, on
average, be answered equally well by all test takers, irrespective of
their IQs. The order of finish, therefore, will depend only on the
cognitive-dependent parts, and the rank order of finish will correspond
to the distribution of intelligence among the test takers. The
principle, having statistical roots, becomes more accurate as the
number of test takers grows large."


Best,


Griffe>