csc@watmath.UUCP (Computer Sci Club) (05/24/84)
Lisa Chabot states that though test scores correlate highly with
something we are interested in, they may correlate even more
highly with something else. True but irrelevant, as long as
the first correlation is still true. (Correlation does not imply
causation.) If the correlation does not change (sometimes not true,
but often true or almost true) then the tests scores can predict
what we are interested in. The strength of the prediction depends
on the strength of the correlation, not on what causes the
correlation. (not exactly true but the counter-examples are of
little interest here)
Tests can be invalidated, or at least made less useful if
a way to bias them becomes widely known (SAT for example).
This can be a problem for those who are trying to come up
with rational admissions standards. The fact does not
invalidate the whole concept of using tests.
William Hugheschabot@amber.DEC (Lisa Chabot) (06/06/84)
The following quotes are from the letter from Mr. Hughes with the following heading: > Subject: Re: Tests and Prediction > Posted: Sun May 20 10:54:59 1984 > Take the much abused I.Q. It is not a very good predictor of individual > preformance but it is not a bad predictor of group preformance. The reason the IQ test is "much abused" is that it has been shown in a number of studies to be heavily culturally-dependent: there are questions dealing with situations unknown to inner-city-raised children, and questions phrased in a vocabulary unpracticed by the same type of children. The choice of spelling the word "performance" as instead "preformance" is rather an amusing pun: the "pre" instead of the correct "per" leads to envisioning the IQ test as measuring how much the children have been preformed, or how much they've been shaped before the test, molded by their own milieu. It is amusing in that it is not something Mr. Hughes probably meant to communicate; we can guess this by his evident faith in IQ tests. > Tests are only useful if they are correlated to some property we are > interested in. ... > If the correlation is strong they become useful predictors. If the > correlation is VERY (and I mean VERY) strong they become useful predictors > of individual preformance. Unfortunately for the researcher, the results of the test may correlate more closely with some property with which the researcher doesn't see or doesn't really want to measure. >(I don't care how well you can write I.Q. tests, I do > care whether you will do well in university). There is a big difference between an ability to do well on IQ tests, and an ability to write IQ tests, and the former does not imply the latter. We never intended to imply the latter from the former. IQ tests can indicate an ability to assimilate white culture, and a good assimilation is very likely necessary to do well in a university. IQ retest scores that are higher than the original may reflect an acquired ease with taking IQ tests or a better assimilation. SAT scores, long considered a good predictor of college performance have been shown by independent studies (independent especially of the corporation that markets them and administers the tests) to be dependent on preparation for the tests. --Lisa S. Chabot UUCP: ...{ decvax | allegra | ucbvax }!decwrl!rhea!amber!chabot USFail: DEC, MR03-1/K20, 2 Iron Way, Marlborough, MA 01752