Herrnstein's Sleight-of-Hand

In the November 27 issue of The Crimson, under the title "The Ersatz Controversy," Richard Herrnstein has once again repeated the litany of "I.Q.'s substantial heritability" and "estimates of I.Q.'s genetic ingredient" that has become such a part of the popular journalism on this issue. In this particular piece Herrnstein tells us that he knows of "no published technical refutation, nor even of any major criticism, by a quantitative geneticist" of I.Q.'s substantial heritability. Since his argument about genetics and social class is deeply rooted in his repeated assertion that I.Q. is highly heritable (80 per cent is the usual figure offered), it is natural that he wants us all to believe that quantitative geneticists, the people who really know, concur.

Herrnstein is engaged in a bit of sleight-of-hand, in which he over and over again tried to dazzle us with that shining crystal "fact" of "high heritability of I.Q.," hypnotizing us into accepting his argument. I want, briefly, to break the spell by showing that the "high heritability of I.Q.," is a non-fact, at least in the context of discussion of social class, and that indeed such phrases as "I.Q.'s substantial heritability" or "the heritability of I.Q. is 80 per cent," despite their appearance as English, are actually scientifically meaningless garbage which have not been refuted in technical journals because there is nothing to refute.

But how can that be? Surely if anything is well established it is that I.Q. is highly heritable. Let me explain.

For any trait in any population at any time there is more or less variation. This variation arises because different individuals differ from each other genetically and because they have experienced up to that moment different environmental histories. In an attempt to partition the causes of the variation, geneticists have introduced the concept of heritability. Unfortunately, there are two different quantities, both of which are called "heritability" in genetics, but which have quite different meanings and consequences. There has been a considerable confusion between the two in the popular literature of the subject, a confusion that has considerable consequences and to which I shall return. For the moment, let us consider what I shall call "descriptive heritability." Descriptive heritability is defined as the proportion of all the variation for a trait in a particular population at a particular time, that results from genetic differences among individuals in the population. As defined and as estimated, the heritability applies only to a particular population of individuals at a particular time. Thus it is nonsense to speak about "the heritability of I.Q." We must refer to "the heritability of I.Q. performances among North German schoolchildren in 1973." That same trait may have a different heritability, higher or lower, in another population at another time. Now any population, say the white population of the United States, is itself heterogeneous both genetically and environmentally. That is, it is made up of many subpopulations, which can be defined along any arbitrary lines, say sex, class, occupational status, degree of relationship, etc. Within each of these subpopulations, the trait in question varies and there is also average variation between subpopulations. The heritability of any trait can be measured within each subpopulation and that heritability is in general not the same as the heritability of the population as a whole and is unrelated to the heritability of the difference between groups. That is because the subpopulations may differ from each other for entirely different reasons than do individuals within subpopulations. In fact, the heritability within subpopulations tells us nothing about the cause of differences between populations. The trait could have 100 per cent heritability within classes, yet no genetic component between classes, for example!

The error of confusing the heritability within a population with the causes of differences between populations was clearly made by Arthur Jensen in his famous article in the Harvard Educational Review, when he tried to infer from heritability studies within the American white population the causes of differences between races. This elementary blunder would not be tolerated in a freshman class in statistics or genetics. We may well wonder how it came to be made by a professor! Precisely the same error is made in arguments about the genetic inferiority of the working class. By referring over and over again to the "high heritability of I.Q.," as if I.Q. had a heritability which was a fixed property of the trait, Herrnstein completely obscures the fact that all measures of heritability of I.Q. are estimates of the heritability within social classes, and indeed within families to a very large extent. In no case ever reported is there an estimate of heritability that can be referred to the whole white population of the United States, or to any large random sample of persons spanning a representative spectrum of social classes or environments. The reasons for this is that estimates of heritability in human populations must make use of the resemblances between relatives. But relatives share both genes and social environment, so great emphasis has been put on the resemblances and differences between identical twins raised apart, or between foster children and their foster parents. When these studies are examined, however, it turns out that neither the separation of twins nor the adoption of children is anything like random over social classes. Indeed, "separated" identical twins are typically raised in the same family or virtually next door to each other. The heritabilities estimated from such data are utterly irrelevant for understanding the differences between classes.

But the problem is even deeper than that. Suppose it were possible to distribute identical twins or foster children completely at random across social class lines. An estimate of heritability based on such data would still not tell us anything about the cause of differences among classes. Imagine that the heritability of I.Q. performance were very high within classes, let it be 100 per cent, but that all differences between social classes were the result of social arrangements. Even in such an extreme case, there would be a high correlation between children and their biological parents and the heritability estimated would be a kind of average between 100 per cent and zero per cent. In fact, if the I.Q. distribution of different social classes were broadly overlapping, as they are, the estimated heritability would be very high, despite the total lack of any genetic causation of the between-class difference. The methodological problems of estimating the genetic component of the differences between groups in human populations is as yet unsolved and no experiment performed or proposed has ever come close to doing so.

Thus we see that the "syllogism" that deduces the genetic inferiority of the lower classes from the "high heritability of I.Q." is not a syllogism at all or even a reasonable inference. It is an illogical non-sequitur which completely misuses technical concepts. It is not worth a moment's serious consideration.

The second piece of obscurantism practiced by the publicists of genetic determinism, also not worthy of refutation in a technical journal because it is explained in elementary textbooks of quantitative genetics, is the confusion of the "descriptive heritability" about which I have been writing, and the second quantity, which I shall call "predictive heritability." While descriptive heritability is that proportion of the variation in a population ascribably to all genetic differences, predictive heritability is the proportion of variation ascribable to a special fraction of the genetic variation. It is always smaller than the descriptive heritability, and it is very difficult to estimate in any human population. Only one, worthless, attempt to do so is known to me. Now, all of the predictions made about the effects of differential patterns of fertility, assortative mating patterns and environmental assortment depend critically upon the "predictive heritability" and not upon "descriptive heritability." Yet the heritability-mongers continue to make pseudo-scientific predictions with the wrong numbers.

What has happened is that an exact scientific notion, "heritability," has, through ignorance or design, been perverted in popular media, and wholly unjustified conclusions have been drawn from this perversion by mystification, arbitrary redefinition of terms and misdirection of attention. There has been no refutation in the technical literature because the technical literature is not devoted to correcting the misapplication of science by amateur publicists. What we have is not an "Ersatz Controversy," but a real controversy affecting the lives of millions of people, created by the purveyors of an ersatz science.

R.C. Lewontin is a professor of Biology.