IQ tests and thermometers

Is intelligence an attribute?

Intelligence is an attribute of persons. Probably for as long as man has been on earth it has been a common observation that persons differ in brightness, in speed of learning, in ability to solve problems, and so on. Parents, teachers, and employers are able roughly to rank children and adults in terms of a subjective impression of brightness or capability, and there is a fairly high agreement among different observers in the rank order they assign in the same groups of children. It is helpful to think of the sub- jective perception of intelligence as analogous to the subjective perception of temperatures, which is also an attribute. Before the invention of the thermometer, temperature was a matter of subjec- tive judgment. The invention of the thermometer made it possible to objectify the attribute of temperature, to quantify it, and to measure it with a high degree of reliability. With some important qualifications, the situation is similar in the case of intelligence tests. The most essential difference is that intelligence, unlike temperature, is multidimensional rather than unidimensional. That is to say, there are different varieties of intelligence, so that persons do not maintain the same rank order of ability in every situation or test that we may regard as indicative of intelligence. It so happens that from among the total spectrum of human behaviors that can be regarded as indicative of some kind of ‘mental ability’ in the broadest sense, we have focused on one part of this spectrum in our psychological concept of intelligence. We have emphasized the abilities characterized as conceptual learning, abstract or symbolic reasoning, and abstract or verbal problem solving. These abilities were most emphasized in the composition of intelligence tests because these were the abilities most relevant to the traditional school curriculum and the first practical intelligence tests were devised to predict scholastic performance. They naturally had a good deal in common with the tests devised for scholastic prediction, since the educational system is intimately related to the occupational demands of a given society. Much the same abilities and skills that are important in schooling, therefore, are important also occupationally. Thus, we find that in industrialized countries practically all intelligence tests, scholastic aptitude tests, military classification tests, voca- tional aptitude tests, and the like, are quite similar in composition and that the scores obtained on them are all quite substantially intercorrelated. In short, there is a large general factor, or g , which the tests share in common and which principally accounts for the variance among individuals. When tests are devised to measure this g factor as purely as possible (i.e., in a factor analysis including a host of other tests it will have nearly all of its variance loaded on the general factor common to all the other tests and have little or no variance loaded on factors found only in certain tests [group factors]), examination of their item content leads to the charac- terization of it as requiring an ability for abstract reasoning and problem solving. Raven’s Progressive Matrices Test is an example of such a test. Tests having quite diverse forms can have equally high loadings on the g factor – for example, the verbal similarities and block design tests of the Wechsler Intelligence Scales are both highly loaded on g. Tests of g can be relatively high or relatively low in degree of ‘culture fairness’. (The question Tn what way are a wheel and a penny alike?’ is probably more culture fair than the question ‘In what way are an oboe and a bassoon alike?’) In short, it is possible to assess essentially the same intelligence by a great variety of means.

Standard IQ tests measure the kinds of behavior in abstract and verbal problem situations that we call abstract reasoning ability. These tests measure more of g – the factor common to various forms of intelligence tests – than of any of the other more special ability factors, such as verbal fluency, spatial-perceptual ability, sensory abilities, or mechanical, musical, or artistic abilities, or what might be called social judgment or sensitivity. But a test that measured everything at once would not be very useful. IQ tests do reliably measure one important, though limited, aspect of human performance. The IQ qualifies as an appropriate datum for scientific study. If we are to study intelligence, we are ahead if we can measure it. Our measure is the IQ, obtained on tests which meet certain standards, one of which is a high g loading when factor analyzed among other tests. To object to this procedure by arguing that the IQ cannot be regarded as being interchangeable with intelligence, or that intelligence cannot really be measured, or that IQ is not the same as intelligence, is to get bogged down in a semantic morass. It is equivalent to arguing that a column of mercury in a glass tube cannot be regarded as synonymous with temperature, or that temperature cannot really be measured with a thermometer. If the measurements are reliable and reproducible, and the operations by which they are obtained can be objectively agreed upon, this is all that need be required for them to qualify as proper scientific data. We know that individually administered IQ tests have quite high reliability; the reliability coefficients are around 0 95, which means that only about 5 per cent of the total individual differences variance is attributable to measurement error. And standard group administered tests have reliabilities close to 0 90.