Genes, Brains, and Human Potential The Science and Ideology of Intelligence

86 PRETEND INTELLIGENCE

between IQ and occupational level (and therefore, income) is also self-

fulfi lling. Again, the mea sures are not in de pen dent.

Th e really critical issue, therefore, surrounds the question of whether

IQ scores predict individual diff erences in the more in de pen dent mea sure

of job per for mance. So let us take a much closer look at that issue.

IQ AND JOB PER FOR MANCE

Th e idea that IQ tests are valid because they also predict job per for mance
is remarkably widespread and durable. Adrian Furnham prob ably refl ects
most views when he claims that “ there is a large and compelling lit er a-
ture showing that intelligence is a good predictor of both job per for-
mance and training profi ciency at work.” Fritz Drasgow describes the
correlation as incontrovertible. John Hunter and Frank Schmidt, in the
1980s, even attached a dollar value to it when they claimed that the U.S.
economy would save $80 billion per year if job se lection were to be uni-
versally based on IQ testing.^13
Th e prob lem is that the “facts” reported turn out to be rather question-
able. A large number of studies prior to the 1970s reported that direct
correlations between IQ and job per for mance were very low (around 0.2–
0.3). Obviously, this result was very disappointing to IQ testers. Th en two
statisticians, Frank Schmidt and John Hunter, considered the possibility
that those many results were error prone. Th e errors, they said, arise from
three main sources. Th e small correlations could be due to small sam-
ples, just as opinion polls with small samples can bias results. Th en the
rough- and- ready mea sures oft en used could be so unreliable as to give
diff er ent scores on diff er ent occasions, which weakens the true correla-
tions. Fi nally, the small samples might have yielded reduced ranges of
scores (of IQ or job per for mance or both). Such restriction of range can
also reduce correlations.
Th ese arguments are quite reasonable, so Hunter and Schmidt at-
tempted to correct the original correlations using statistical methods and
pooling the results (using meta- analy sis, as described in chapter 2). Th is
doubled the correlations to around 0.5–0.6. Nearly all studies cited in
favor of IQ validity are either drawn from the Schmidt and Hunter meta-
analyses or from others using the correction methods developed by them.

This content downloaded from 139.184.14.159 on Tue, 17 Oct 2017 13:52:12 UTC

http://www.ebook3000.com