PRETEND INTELLIGENCE 89their report, on which most of this would-be validity is based, Hunter and
Schmidt acknowledged that crucial bits of information like test reliabil-
ity and range restriction, were only sporadically available. Instead, these
crucial gaps were fi lled in by generalizing from those scores actually
available. Such guesstimates have obvious dangers.
Th ese prob lems of missing data are hardly surprising, given that most
of the studies in question were done between 1920 and 1970. In 1989,
a committee set up by the U.S. National Acad emy of Sciences commis-
sioned new meta- analyses on more recent studies than those of Schmidt
and Hunter. Th ese analyses found much lower correlations than those
reported above. As report authors John Hartigan and Alexandra Wignor
(themselves leading statisticians) noted in their report, “Th e most striking
fi nding... is a distinct diminution of validities [i.e., IQ- job per for mance
correlations] in the newer, post 1972 set.” Th e corrected correlations came
out to be around 0.25, rather than the widely cited 0.50 from the corrected
correlations of Schmidt and Hunter.^17
Th e committee described the diff erences as “puzzling and obviously
somewhat worrisome.” But they noted how the quality of data might
explain it. For example, the 264 newer studies have much greater numbers
of participants, on average (146 versus 75). It was shown how the larger
samples produced much lower sampling error and less range restriction,
also requiring less correction (with much less possibility of a false boost
to observed correlations). And there was no need to devise estimates to
cover for missing data. So, even by 1989, these more recent results are in-
dicative of the unreliability of those usually cited. But it is the earlier test
results that are still being cited by IQ testers.
Fi nally, it seems that even the weak IQ- job per for mance correlations
usually reported in the United States and Eu rope are not universal. For
example, in a study reported in 2010, Eliza Byington and Will Felps found
that IQ correlations with job per for mance are “substantially weaker” in
other parts of the world. Th ey include China and the Middle East, where
per for mances in school and work are more attributed to motivation
and eff ort than to innate cognitive ability.^18
Th ese refl ections on correction methods all add to the impression of a
large amount of guesswork involved in arriving at corrected correlations
between IQ and job per for mance as well as overzealous claims about test
validity.
This content downloaded from 139.184.14.159 on Tue, 17 Oct 2017 13:52:12 UTC