I think many of the responses to the IQ and g discussion have mistaken
models of cognition and validity that underlie them. I will just
summarize these by making two points. The first is that measurement
devices, such as IQ tests, are valid or reliable and do not embody
multiple validities or reliabilities. Since there are a few ways
validity and reliability are estimated, the tendency is to think that a
measurement tool embodies multiple validities and reliabilities. There
is only one validity and this is determined by 1) the logical integrity
of the theory that defines the construct and the 2) mapping of the
construct on to a measurement device. The measurement device can
include a multiple choice test, recall test, self-report questionnaire
etc., whatever matches the theoretical model of the construct.
Empirical studies are then conducted to test hypotheses about the
mapping of the theory to the device. These studies have been given
different names that we are all familiar with, such as criterion
validity, concurrent validity etc. This suggest that there are
different validities for each test. There is only one validity,
construct validity, and this is estimated using a variety of empirical
studies. However, no empirical study can provide evidence for the
validity of a measurement device if the theoretical model defining the
construct is vague. Since this is a teaching forum, the example I
present in class is a set of validity coefficients for a test I refer to
as the Spelling Test from the Wide Range Achievement Test. These
include factor analyses and prediction studies, using the test to
predict grades and other criteria. I then reveal the test items. The
first item is the first item from the Arithmetic subtest of the WRAT,
something like 7+5=?. The correlations I presented were all studies of
the Arithmetic subtest. They appear very convincing and they are
generally in the same range as the correlations of the Spelling
subtest. The point is that the constructs were extremely different and
the correlation patterns were indistinguishable for both constructs.
The only way I know this is because I have theoretical models of
Spelling and Arithmetic that are clear and distinct. Validity (i.e.
Construct Validity) is in the theoretical understanding of the theorist,
not determined by empirical studies. IQ and g are not theoretically
clear. Their validity is consequently unknown even if the device called
an IQ test correlates with other measurements in expected directions and
magnitudes. Once you get to the level of IQ battery subtests, many of
these problems become clear. Just as an example, it is clear from item
examination and factor analyses that the Information, Vocabulary,
Similarities and Comprehension subtests of the WAIS just measure a
better-defined construct called Semantic Knowledge. If you are familiar
with the subtests, just think about this as a theoretical possibility.
For example a Comprehension subtest item, "Why do we pay taxes?"
requires the semantic knowledge associated with the word "taxes". It is
just another way of asking for the definition of the word "taxes".
These subtests are grouped by the test developer under a construct
called Verbal Intelligence. All the Performance subtests group together
because they are timed tests. They may also measure other constructs but
their common variance is based on the subject solving problems quickly.
However, the grouping is given the name Performance Intelligence. The
odd couple, Arithmetic and Digit Span, group together because they share
variance on Sustained Attention. Kaufman called the grouping Freedom
From Distractability. The General IQ score is just the average of all
these scores in comparison to the population average of the scores. No
factor analysis has ever supported averaging all the subtests. This
would require that Semantic Knowledge correlate highly with Sustained
Attention etc. The constructs of Semantic Knowledge, Sustained Attention
and Timing are much better defined than the constructs Verbal
Intelligence, Performance Intelligence and Freedom From
Distractability. IQ has no clear definition as a measurement construct;
Semantic Knowledge does. The WAIS should be called the Semantic
Knowledge, Sustained Attention and Timing Test. There is no g in the
WAIS that I can discern. The correlations of the WAIS IQ scores with
other tests, grades etc., exists because there is a correlation of
semantic knowledge with these other measures. Semantic Knowledge exists
but intelligence does not. In as much as semantic knowledge is acquired
through reading and education, the correlations of the WAIS with any
other measure is just the correlation of one measure of education with
other education measures (e.g. grades), or other criteria that are also
influenced by education (e.g. occupation success, salary etc).
Somehow, psychologists were given free reign to make assessments and
characterize the thinking abilities of millions of people without a
clear theory of thinking abilities and how to measure them. Across the
Delaware river sits the battleship New Jersey, the first ship whose crew
was chosen and placed based on IQ tests. They could have used a
Phrenology examination and derived the same results. IQ tests are the
biggest flimflam every propagated on the American public. We must
abandom them in order to make progress in cognitive assessment.
Mike Williams
---
You are currently subscribed to tips as: [email protected].
To unsubscribe click here:
http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=36029
or send a blank email to
leave-36029-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
