More on GREs (but, mercifully, not *much* more)

Evan W. Steeg steeg at cs.toronto.edu
Sat Feb 12 16:03:42 EST 1994


  Just to wind down my diatribe against GREs and standardized
testing.... :-)

  W.R. Pearson made (in a recent posting to which I no longer have
spool access) a good point about the rational use of GRE scores in a
limited way within a broader set of evaluation criteria.  He discussed
the idea of the GREs being necessary to an initial filtering process,
given the economic constraints on departments and admissions
committees (you can't fly hundreds of applicants to Virginia from
around the world in order to interview them).

  This seems eminently reasonable to me -- all the more so as long as
admissions officers don't set the GRE thresholds ridiculously high and
as long as they remain aware of the risk of missing potentially great
researchers among the bad test-takers.

  That's really the main point I was trying to make: It doesn't really
make sense to pick potential marathon runners on the basis of their
times for the 100-metre dash.  But if you have to, you have to.  Make
the best of it.  Remember, though, that the issues of "good
predictors" in statistics are much more complex than the
greed-motivated public relations staff of the Educational Testing
Service would have you know.  To make one last trip to the well of
sports metaphors....  100-metre dash speed may correlate positively
with marathon speed in one population while these variables may show
zero or even negative correlation in other populations.  E.g., in the
population of North America, much of the statistical work in the
correlation is in essentially separating the athletes from the couch
slobs, and so sprint speed and distance running (and weighlifting
strength, for that matter) will tend to go together; but in a sampling
of Olympic athletes, almost nobody will score among the best in *both*
sprint and distance, and so one variable will *not* be a good
predictor of the other.  Similarly for the GREs: Such tests may be
fine for separating the "smart" from the "stupid", but when you're
already sampling from the upper echelons of academia, you run into the
effects of all the different very smart people having different
combinations of intellectual and organizational skills, all of which
contribute to overall academic success, but only one or two of which
are measured even remotely accurately by GRE-type tests.

  Evan


-- 

Evan W. Steeg (416) 978-5182              steeg at ai.toronto.edu 
Dept of Computer Science                  steeg at t13.lanl.gov 
University of Toronto,           
Toronto, Canada M5S 1A4         



More information about the Bioforum mailing list