IQ: the narrow gauged failed way

There are those who believe that intelligence can be measured and that those measurements can be used to explain the disparities in achievement between classes and ethnic groups. They are indulging in a fallacy, says Steven Rose

The nature of intelligence, its measurement, distribution between individuals, classes and races, and above all its biological meaning, has been the focus of scientific and numerological fact and fantasy for more than a century now. Ameliorists, eugenicists, racists and biological determinists have battled through spurious dichotomies of nature and nurture ever since Francis Galton wrote Hereditary Genius back in 1869.

Buried in the aftermath of 1945 with the Unesco statements on race, the arguments re-emerged a century after Galton with Arthur Jensen's claims that race differences in IQ scores in the United States were the consequence of innate (heritable) differences between blacks and whites, rather than of an inadequate educational system in a racist society. In Britain, the followers of Galton, Karl Pearson and Cyril Burt, placed class rather than race at the core of the argument. A decade of vigorous polemic, which included the exposure of the untrustworthiness of Burt's twin-based research, was followed by a period of quiescence, only to be punctuated in the 1990s by the shrill voices of a small group of self-declared "scientific racists" on both sides of the Atlantic, largely funded by the US-based Pioneer Foundation and with Richard Herrnstein and Charles Murray's Bell Curve as their New Testament.

There is, however, a weary sense of deja vu about their arguments, which have scarcely changed since Galton. In summary they claim: a. There is a thing called intelligence, which intelligence tests measure, yielding a single figure, IQ, the ratio of any person's "intelligence" to that of the general population.

b. Intelligence is a relatively fixed property of the brain, and individual differences are largely heritable.

c. Populations differ in their average IQ scores. In the US blacks score on average lower than whites, and in both the US and the United Kingdom the working class lower than the middle class.

d. These differences are so great that "the environment" cannot account for them, and therefore differences between groups are also inherited.

Propositions a, b and d are false. Proposition c is true, though for scientifically trivial, albeit socially profound, reasons.

IQ tests were originally devised, by Binet in Paris, as an aid to teachers in comparing the performance of any individual child with the average for his or her class, so as to be able to offer remedial education. Brought to the US by Terman as the Stanford-Binet test in 1916 and by Burt to England, tests were extended to adults and their purposes profoundly shifted. Now they were explicitly used to probe the relative ability of non-English speaking immigrants to the US or working-class children in England to comprehend such important matters as the nicknames of baseball players or the design of tennis courts. On the basis of such groups' poor performance on these tasks, the new generation of psychometricians made alliance with eugenicists to declare that intelligence was a fixed property of individuals, lower in non Anglo-Saxon immigrants to the US who were hence diluting the superior WASP stock. In Britain, tests were instruments of the policy of selective education. This sorry history has been told many times, yet psychometry has signally failed to outgrow its past.

Newer, more "culture-fair" tests have been devised since, combining vocabulary, numerical and analogical reasoning and pattern recognition, but they continue to ask children to make judgements about class and acceptable social behaviour. Standardised against the Stanford-Binet, they essentially replicate its values. Yet the tests are elastic instruments. When in the 1930s it became clear that girls were performing better than boys on some items, the tests were "adjusted" by adding and subtracting items until on average the genders performed equally well. No such adjustments have been made for class or race differences, though it is possible to devise tests on which, for example, working-class children do better than middle-class ones. One such tested children on items which included several items of misinformation. Working-class children could cope better with such tests, which perhaps mirrored more closely their day-to-day experience. The point is that tests are essentially social, not biological instruments, adapted to the goals of the testers.

Much is made of the fact that tests distribute the population according to a (nearly) Gaussian bell curve. Yet this is exactly what they are devised to do, and tests that give different distributions are rejected, just as the university marking system is devised to produce some 10 per cent firsts and 10 per cent fails with the rest distributed between. Bell curves are statistical devices, not biological facts. Standard IQ tests correlate with school performance, and social and economic status. Herrnstein and Murray argue that the IQ score drives both educational and economic success, but in a world of unequal education and imperfect social mobility, where wealth and privilege are indeed inherited, the chain of causation is to say the least obscure.

Nonetheless, psychometricians insist that IQ tests measure some real, underlying and fixed feature of an individual's biology, which they call "crystallised intelligence", given, following Charles Spearman, the mystical designation "g". (I sometimes wonder whether this appropriation of one of physics' most hallowed symbols, that for gravity, isn't a deliberate attempt to scientise psychometry's numerology.) The idea that all of a person's multiple abilities, cognitive capacity, creativity, verbal facility, empathy, capacity to communicate, speed of reaction, artistic or musical talent can be reduced to a single number, and the entire population ranked upon such a scale as if regimented by height, would strike most people as absurd. Indeed, it is only achieved, as the authors of The Bell Curve make clear, by discarding most aspects of what non-psychometrically orientated psychologists would regard as intelligence, seen as irrelevant to the real g - that which the tests measure.

To a neurobiologist like myself this insistence is closer to that of Galileo's contemporaries who persisted in believing in Ptolemaic epicycles than it is to modern science. We speak of the brain processes of memory, arousal, attention, perception as among the multiple aspects of cognition, and recognise intelligence as a process, probably non-quantifiable, of engagement by individuals with the social and natural worlds that surround them. Psychometricians cling to their pathetic bit of elastic and tell us it is a ruler with which they can control the world.

These days psychometricians do not make the mistake of claiming to be able to partition any individual's intelligence - or IQ score - into genetic and environmental components. What they do claim, however, is that they can calculate the contribution made by genes and environment to the variance in IQ scores between individuals, by means of a measure known as heritability. This figure, based on comparing IQ scores of more or less closely related individuals, comes to somewhere between 0.6 and 0.8 - that is, up to 80 per cent of the difference in IQ between individuals is claimed to be genetic, even discounting Burt's infamous twins.

But it is important to be clear what such a figure means. Plant a field with a relatively genetically homogeneous variety of wheat, treat patches with varying combinations of fertilisers, soil quality, availability of water, etcetera, and crop yield will also vary. How much of that variance is due to genetic differences and how much to the different environments? In an absolutely uniform environment, of course - were such a thing possible - all the variance would be contributed by the genes, and with absolutely identical genes, all the variance would be contributed by the environment. But this never happens. Genotypes and environments both vary and the purpose of heritability estimates is to try to tease them apart. To do so, however, it is necessary to make some simplifying assumptions. Variance is assumed to be largely additive such that the genetic and environmental components sum to nearly 100 per cent. The remainder, which to make the mathematics work, has to be a rather small proportion of the total, is considered the product of an interaction between genes and environment. Thus, if V is the total variance, G the genetic and E the environmental contribution, then: V = G + E + (GxE).

If genotypes are distributed randomly across environments, heritability defines the proportion of the variance which is genetically determined. However, if there is a great deal of interaction between genes and environment, if genes interact with each other, and if the relationships are not linear and additive but interactive, the entire mathematical apparatus falls apart. In consequence the estimates can only be made in very special cases, from which the majority of traits of interest, outside the special world of plant and animal breeders, are likely to escape. Furthermore, without going into the technical details of the maths, the figure derived for the heritability is itself dependent on the environment - that is, if you change the environment, the heritability estimate changes.

More than any other aspect of genetics, heritability estimates have been persistently misunderstood. And their misapplication perhaps explains why psychometricians claim that everything from voting to marriage intentions shows significant heritability. Relatively high estimates (above 35 per cent) have been provided for such an odd list of attributes as attitudes to the death penalty, sabbath observance, working mothers, military drill, white superiority, cousin marriage, royalty, conventional clothes, apartheid, disarmament, censorship, "white lies", jazz and divorce. Even nudist camps and women judges rate around 25 per cent, so it is I suppose a matter of some surprise that there appears to be virtually zero heritability for pyjama parties, straitjackets and co-education.

The most parsimonious explanation for this bizarre set of statistics is that they demonstrate the inappropriateness of attempting to apply a mathematical formalism devised for plant and animal breeding to such dubiously "objective" measures as the diversity of human social behaviour and attitudes. The genetic claims of the psychometricians draw sustenance from the advances being made in identifying specific genes by way for instance of the human genome programme, but they are as remote from such modern genetics as Minoan B from the Internet.

That there are differences in average IQ scores between races and classes is incontrovertible. That they have anything to do with genetic differences between populations is strictly unprovable. Such evidence as exists points in the reverse direction. Not only has there been a secular increase in IQ scores across cohorts in the decades since the measurements began - an increase which genetic models have a hard time explaining away - but the differences between the populations are steadily declining. In any event, heritability estimates simply cannot be applied between populations. Even if IQ were an objective measure like crop or milk yield, its heritability estimate would apply at best only to differences within a freely interbreeding population. Until such time as humans live in a society in which social barriers restricting relationships between individuals from different ethnic and social groups no longer exist, the estimates are as scientifically meaningless as they are socially and politically pernicious.

Steven Rose is director of the Brain and Behaviour Research Group at the Open University.