“Science or Superstition?” Cognition

Dr. David Layzer

At the very outset we have to ask, is IQ a valid measure of intelligence? Jensen and Herrnstein assure us that it is. ‘The most important fact about intelligence is that we can measure it, ” says Jensen, while Herrnstein remarks that the “objective measurement of intelligence” is psychology’s “most telling accomplishment. ”I find these claims difficult to understand. To begin with, the objective measurement” does not belong to the same logical category as what it purports to measure. IQ does not measure an individual phenotypic character like height or weight; it is a measure of the rank order or relative standing of test scores in a given population. Thus the statement, “A has an IQ of 100″ means that half the members of a certain reference population scored lower than A on a certain set of tests and half scored higher. “B has an IQ of 115″ means that 68% of the reference population scored lower than B and 32% higher, and so on. (IQ tests are so constructed that the frequency distribution of test scores in the reference population conforms as closely as possible to a normal distribution – the familiar bell-shaped curve - centered on the value of 100 and having a half-width or standard deviation the square root of the variance of 15 points). To call IQ a measure of intelligence conforms neither to ordinary educated usage nor to elementary logic.

One might perhaps be tempted to dismiss this objection as a mere logical quibble. If IQ itself belongs to the wrong logical category to be a measure of intelligence, why not use actual test scores? One difficulty with this proposal is the multiplicity and diversity of mental tests, all with equally valid claims. (This is part of the price that must be paid for a strictly “operational” definition of intelligence.) Even if one were to decide quite arbitrarily to subscribe to a particular brand of mental test, one would still need to administer different versions of it to different age groups. An appearance of uniformity is secured only by forcing the results of each test to fit the same Procrustean bed (the normal distribution). But this mathematical operation cannot convert an index of rank order on tests having an unspecified and largely arbitrary content of ”objective measure intelligence”.

Similarly, the inference that IQ is a measure of intelligence depends on certain assumptions, namely: (a) that there exists an underlying one-dimensional, metric character related to IQ in a one-to-one way, (b) that the values assumed by this character in a suitable reference population are normally distributed.

If these assumptions do not in themselves constitute a theory of human intelligence, they severely restrict the range of possible theories. Once again, we see that the “operational stance”, though motivated by a laudable desire to avoid theoretical judgments, cannot in fact dispense with them. The choice between a theoretical approach and an empirical one is illusory; we can only choose between explicit theory and implicit theory. But let us examine the assumptions on their own merits.

The first assumption is pure metaphysics. Assertions about the existence of unobservable properties cannot be proved or disproved; their acceptance demands an act of faith. Let us perform this act, however - at least provisionally - so that we can examine the second assumption, which asserts that the underlying metric character postulated in the first assumption is normally distributed in suitably chosen reference populations. Why normally distributed?

A possible answer to this question is suggested by a remark quoted by the great French mathematician Henri Poincare: “Everybody believes in the normal distribution: the experimenters because they think it can be proved by mathematics, the mathematicians because it has been established by observation. ” Nowadays, both experimenters and mathematicians know better. Generally speaking, we should expect to find a normal frequency distribution when the variable part of the measurements in question can be expressed as the sum of many individually small, mutually independent, variable contributions.

This is thought to be the case for a number of metric characters of animals such as birth weight in cattle, staple length of wool, and (perhaps) tentacle length in octopuses. It is not the case, on the other hand, for measurements of most kinds of skill or proficiency. Golf scores, for example, are not likely to be normally distributed because proficiency in golf does not result from the combined action of a large number of individually small and mutually independent factors.

What about mental ability? Jensen and Herrnstein believe that insight into its nature can be gained by studying the ways in which people have tried to measure it. Jensen argues that because different mental tests agree moderately well among themselves, they must be probing a common factor (Spearman’s g). Some tests, says Jensen, are “heavily loaded with g”, others not so heavily loaded.

Thus, g is something like the pork in cans labeled “pork and beans” Herrnstein takes a less metaphysical line. Since intelligence is what intelligence tests measure, he argues, what needs to be decided is what we want intelligence tests to measure. This is to be decided by ” subjective judgment” based on “common expectations” as to the “instrument.

In the case of intelligence, common expectations center on the common purposes of intelligence testing - predicting success in school, suitability for various occupations, and intellectual achievement in life. ” Thus, Herrnstein defines Intelligence “instrumentally” as the attribute that successfully predicts success in enterprises whose success is commonly believed to depend strongly on intelligence.

That is intelligence is what is measured by tests that successfully predict success in enterprises whose success is commonly believed to depend strongly on what is measured by tests that successfully predict success in enterprises whose success is commonly believed to depend strongly on.