Richard J. Herrnstein and Charles Murray
The Bell Curve: Intelligence and Class Structure in American
Life
New York: Free Press, 1994. 845 pp. ISBN 0-02-914673-9. $30.00
Reviews by Thomas J. Bouchard, Jr., and Donald D. Dorfman
"We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain inalienable Rights, that among these are Life, Liberty, and the pursuit of Happiness." With these words Jefferson introduced one of America's most treasured documents, the Declaration of Independence. Successive generations of Americans have not only embraced Jefferson's noble sentiments, they have embellished them. Equality of political rights and legal standing has been expanded into a belief in literal equality; today, differences in outcome are taken as prima facie evidence of unequal opportunity. In an egalitarian society such as ours, the existence of significant and enduring individual or group differences in intelligence is seen as a challenge to our highest ideals. This challenge is taken up by Richard J. Herrnstein and Charles Murray in The Bell Curve.
The Bell Curve has a simple but powerful thesis: There are substantial individual and group differences in intelligence; these differences profoundly influence the social structure and organization of work in modern industrial societies, and they defy easy remediation. In the current political milieu, this book's message is not merely controversial, it is incendiary. As scholars such as Daniel Moynihan, Arthur Jensen, and E. O. Wilson have learned, the mainstream media and much of the scientific community have little tolerance for those who would question our most cherished beliefs. Herrnstein and Murray have received similar treatment. They have been cast as racists and elitists, and The Bell Curve has been dismissed as pseudoscience, ironically by some commentators who broadly proclaim that their critique has not benefited from a reading of the book. The book's message cannot be dismissed so easily. Herrnstein and Murray have written one of the most provocative social science books published in many years. The issues raised are likely to be debated by academics and policymakers for years to come.
Commentators from across the political spectrum have documented the profound social changes that all industrialized societies are undergoing at the end of the 20th century--erosion of the middle class, loss of well-paying manufacturing jobs, and an emerging information age in which individual success will depend on brains not brawn. The Bell Curve tells a similar story regarding the United States. It differs from other works by focusing on intelligence, rather than education or social class as a causal variable. The authors tell us that true educational opportunity as a function of ability (measured by IQ tests) did not arrive in the United States until about 1950. Until that date only about 55 percent of high school graduates in the top IQ quartile went directly to college. From 1950 to 1960, this number jumped to 72 percent, and in 1980 over 80 percent of graduates in the highest ability quartile went to college. In addition, sorting by cognitive ability continues as students move through college. It also occurs across colleges, with the elite schools selecting the more intellectually talented students. Finally, it continues across careers in the world of work. The authors argue that intellectual stratification through occupations is driven by powerful economic pressures. This argument is based on a number of different and compelling lines of evidence. If Herrnstein and Murray are correct, current social inequalities reflect, in large part, the achievement of a meritocracy based on cognitive ability.
The notion of a meritocracy is not, in itself, an affront to American sensibilities. Social scientists have carefully documented that social mobility does occur from one generation to the next and that cognitive ability is a major factor in determining whether an individual will achieve greater or lesser social status than did his or her parents (Waller, 1971). When each generation resorts in this way, the elements of fairness and opportunity are preserved. If, however, as The Bell Curve asserts, the heritability of IQ is quite high and there is a strong tendency for those similar in ability to marry, there will be less regression toward the mean in the cognitive ability of children of the intellectually talented and, therefore, less intergenerational reassortment. Under these circumstances a meritocracy begins to look like an aristocracy, a perception that is strongly reinforced when the intellectual elite segregate themselves from the rest of society by living in separate neighborhoods, sending their children to private schools, and supporting social institutions that cater to their own unique interests.
The authors do argue that general cognitive ability (i.e., "g") is a major determiner of social status and that variance in general mental ability is largely attributable to genetic factors--propositions that are certainly endorsed by many experts in the field. The book explicitly disclaims, however, that general mental ability is the only determinant of social status, or that g is the sum total of an individual's social worth.
The Bell Curve carefully documents in table after table, graph after graph that cognitive ability has become a more important determinant of social status than social class of origin. Although this may come as a surprise to many, it is consistent with a large body of evidence. Research methodology in the domain of individual differences has changed dramatically in the past 20 years. Many investigators in this domain now accept two major methodological principles: that single studies based on small samples are inherently uninformative and that correlations calculated from data gathered within biological families are seriously confounded. Understanding both of these principles is important when evaluating evidence often brought to bear against The Bell Curve.
Results from a single modest study carry little more weight than does a single anecdote, no matter how compelling the finding. Most social scientists, but certainly not all, have adopted the methodology of meta-analysis, a statistical tool that systematically combines the results from many studies to provide a single reliable conclusion. In a similar fashion, behavioral geneticists combine the results from numerous kinships weighted by their sample sizes to provide the best estimate of the degree of environmental and genetic influence on any particular trait. Any single study is viewed as providing only weak evidence on its own.
The confound generated by data drawn from within biological families provides numerous pitfalls when assessing this book's claims and reviewers' counterclaims. Within a biological family, correlations (e.g., parental socioeconomic status x child's IQ) are ambiguous because the cause of the correlation could be the family environment or the parent's genes. Within biological families, the correlation between parental socioeconomic status (SES) and child's IQ, based on a meta-analysis of the literature, is .333 (White, 1982). However, in studies where genetic effects are held constant, through twin or adoption designs, the correlation drops dramatically (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990; Scarr & Weinberg, 1978). Another striking exemplar of this phenomenon is the IQ correlation between unrelated individuals reared together who share a common family environment but lack a common genetic background. When the cognitive ability of these "unrelated siblings" is measured in adulthood the correlation is zero (McGue, Bouchard, Iacono, & Lykken, 1993). Thus the correlation between parental SES and offspring IQ in biological families is due, in some measure, to genetic endowment. Consequently, when examining the relationship between IQ and a dependent variable, to "hold constant" the SES of biological parents (on the grounds that SES is a competing "environmental explanation") results in an underestimate of the true influence of IQ. As early as 1970, Paul Meehl warned that "the commonest error in handling nuisance variables of the `status' sort (e.g., income, education, locale, marriage) is the error of suppressing statistically components of variance that, being genetic, ought not be thus arbitrarily relegated to the `spurious influence' category" (pp. 393-394). In this book, intended for lay readers as well as academicians, the authors have purposefully provided simple and straightforward analyses of SES and cognitive ability. They have, in many instances, understated the role of cognitive ability by holding SES constant. We can expect to see numerous reanalyses and the presentation of many more complex models derived to support both sides of the debate. The careful reader will remember Meehl's caution when examining the data and drawing conclusions.
Part II of The Bell Curve reviews the role of cognitive ability in areas of social dysfunction. In this section, the data are more complicated, conclusions more equivocal. In spite of claims to the contrary by some reviewers, the book makes it clear that with regard to the issues discussed in this section of the book (e.g., poverty, schooling, unemployment, idleness and injury, family matters, welfare dependency, parenting, crime, civility, and citizenship), IQ "almost always explains less than 20 percent of the variance, . . . usually less than 10 percent and often less than 5 percent" (p. 117). These analyses deal only with non-Latino Whites and make use of the National Longitudinal Survey of Labor Market Experience of Youth (NLSY). This large nationally representative survey, begun in 1979, incorporated the Armed Forces Qualification Test (AFQT). The AFQT provides an excellent measure of g, and the survey contains sufficiently detailed information that questions regarding the influence of g on the outcomes listed above can now be addressed systematically.
I discuss the results regarding poverty as an exemplar. First, it must be noted that the decline in poverty from 1940 to 1970 is dramatic and linear, dropping from over 50 percent to less than 15 percent. It has remained nearly constant since 1970. This means that the rise in crime, drug abuse, and many other discontents over the past 25 years cannot be ascribed to poverty per se. It also means the analyses in The Bell Curve are being carried out on a very different population than would have been used had the analysis been carried out before 1970. Consequently, comparisons with earlier research are problematical. The evidence strongly supports the conclusion that high IQ is an important protective factor, and low IQ is an important risk factor. Parental SES is not nearly as protective or nearly as debilitating. IQ has an effect even when education is held constant. When one looks at poverty among women with children, the situation is quite different. For separated, divorced, or never married White mothers with very low IQs, the probability of being in poverty is almost 70 percent. For the same group of mothers with very high IQs, the risk of poverty is about 10 percent. For married mothers, however, the range is from under 20 percent to near zero. IQ is influential, but marriage is clearly more important. Thus poverty among children is strongly associated with the marital status of their mothers. Holding IQ constant washes out any influence of parental SES for both types of mothers but leaves a large marital effect. Similar empirical demonstrations, with numerous twists and turns, are made regarding the other dependent variables enumerated above.
Part III of The Bell Curve contains the most controversial chapter in the book, "Ethnic Differences in Cognitive Ability." The data reviewed here are neither new nor surprising and find strong support in the current psychological literature (Humphreys, 1988). East Asians, living in Asia or America, score above White Americans in tests of cognitive ability; the best estimate of that difference is about three points with findings ranging from no difference to a 10-point spread in test scores. The difference in measured IQ between African Americans and Whites has remained at about 15 IQ points for decades, although there is some indication of very modest convergence due to fewer low scores in the African American population. Controlling for SES reduces but does not eliminate this difference, and of course, controlling for SES in ethnic group contrasts may eliminate a valid source of IQ variance. Moreover, ethnic differences on cognitive tests cannot be attributed to test bias.
As described earlier, The Bell Curve asserts that differences in cognitive ability between individuals are due in part to differences in their genetic endowment. A great deal of research supports this conclusion (Bouchard, 1993; Pedersen, Plomin, Nesselroade, & McClearn, 1992). The question is, What can we infer from these findings about the origins of ethnic group differences? As any graduate student knows, the source of individual differences in a trait cannot be taken as evidence for the source of group differences in the same trait. A great deal of indirect evidence points to both genetic and environmental contributions to ethnic group differences in IQ. None of this evidence, however, is as firm as the evidence for genetic influence on individual differences in IQ. Many experts in the field (Snyderman & Rothman, 1988) agree with Herrnstein and Murray when they state that "it seems highly likely to us that both genes and the environment have something to do with racial differences. What might the mix be? We are resolutely agnostic on the issue; as far as we can determine, the evidence does not yet justify an estimate" (p. 311).
The Bell Curve closes with a review of the policy implications of their findings. What is the role of the social scientist in the formulation of social policy? I agree with Kendler (1993) that it is clearly within the scientific realm to comment on the likely consequences of competing social policies. Judging the value, as opposed to the costs, of such policies is, however, a matter of political rather than scientific discourse. As Kendler documents, many social scientists confuse these two functions. Herrnstein and Murray have been vigorously chastised for discussing policy implications on the basis of the work reviewed and the data analyzed in their book. Similar assertions are, however, regularly made by many investigators in the social sciences. For example, the implications of specific research projects are regularly found in grant applications where they are used to justify the request for funds. Seldom are the value judgments underlying these implications explicitly stated, but they are easily inferred. Herrnstein and Murray have, in my opinion, been much more "up front" about these matters than many social scientists, and their discussions fall clearly within the boundaries discussed by Kendler. They argue, for example, with regard to affirmative action, "Our contribution (we hope) is to calibrate the policy choices associated with affirmative action, to make costs and benefits clearer than they usually are" (pp. 387-388).
In writing the Declaration of Independence, Jefferson was attempting to give birth to a shared political goal--freedom, as expressed in the right to life, liberty, and the pursuit of happiness. Herrnstein and Murray also address this important theme. They make it clear that a meritocracy need not be a Darwinian jungle and that a responsible society should make a place for everyone. Their description of the ideal meritocracy will not be to everyone's taste, but it is neither more foolish nor more naive than many proposals that have been suggested in the past. Nevertheless, predicting the future is an extremely hazardous enterprise. We have recently seen the virtual collapse of a number of societies that were based on a totally different conception of human nature than that underlying The Bell Curve. Virtually no one predicted this dramatic outcome for one of history's largest social experiments. Undoubtedly, Herrnstein and Murray's arguments are wrong in some of the details, and they may be wrong about the larger picture. Nevertheless, one of the goals of the intellectual enterprise is to question received wisdom, to ask difficult questions, and to seek novel and "better" solutions to both new and old problems. They have succeeded admirably at this task.
This is a superbly written and exceedingly well-documented book. It raises many troubling questions regarding the organization of our society. It deserves the attention of every well-informed and thoughtful citizen.
Bouchard, T. J., Jr. (1993). The genetic architecture of human intelligence. In P. A. Vernon (Ed.), Biological approaches to the study of human intelligence (pp. 33-93). Norwood, NJ: Ablex.
Bouchard, T. J., Jr., Lykken, D. T., McGue, M., Segal, N. L., & Tellegen, A. (1990). Sources of human psychological differences: The Minnesota study of twins reared apart. Science, 250, 223-228.
Humphreys, L. G. (1988). Trends and levels of academic achievement of Blacks and other minorities. Intelligence, 12, 231-260.
Kendler, H. H. (1993). Psychology and the ethics of social policy. American Psychologist, 48, 1046-1053.
McGue, M., Bouchard, T. J., Jr., Iacono, W. G., & Lykken, D. T. (1993). Behavior genetics of cognitive ability: A life-span perspective. In R. Plomin & G. E. McClearn (Eds.), Nature, nurture and psychology (pp. 59-76). Washington, DC: American Psychological Association.
Meehl, P. E. (1970). Nuisance variables and the ex post facto design. In M. Radner & S. Winokur (Eds.), Minnesota studies in the philosophy of science IV (pp. 373-402). Minneapolis: University of Minnesota Press. Pedersen, N. L., Plomin, R., Nesselroade, J. R., & McClearn, G. E. (1992). A quantitative genetic analysis of cognitive abilities during the second half of the life span. Psychological Science, 3, 346-353.
Scarr, S., & Weinberg, R. A. (1978). The influence of family background on intellectual attainment. American Sociological Review, 43, 674-692.
Snyderman, M., & Rothman, S. (1988). The IQ controversy: The media and public policy. New Brunswick, NJ: Transaction Books.
Waller, J. H. (1971). Achievement and social mobility: Relationship among IQ score, education and occupation in two generations. Social Biology, 18, 252-259.
White, R. K. (1982). The relation between socioeconomic status and academic achievement. Psychological Bulletin, 91, 461-481.
"Is there a danger that current welfare policies, unaided by eugenic foresight, could lead to the genetic enslavement of a substantial segment of our population? The possible consequences of our failure seriously to study these questions may well be viewed by future generations as our society's greatest injustice to Negro Americans" (Jensen, 1969, p. 95).
So said Arthur Jensen in 1969 in a Harvard Educational Review article on race and general intelligence. General intelligence is often called IQ for short. In the most controversial parts of The Bell Curve, a book written for the general reader, Richard J. Herrnstein and Charles Murray present much the same theories and general concerns as did Jensen with regard to Black cognitive intelligence, while extending the analysis to include Latinos. They greatly expand on the evidence, present possible causal links between IQ and socially undesirable behaviors, and at the end of the book suggest implications for public policy. They are especially worried about a supposed downward pressure on the distribution of IQ in the United States, which they call dysgenic pressure. Dysgenic is a term borrowed from population biology. As does Jensen, the authors believe that Blacks "are experiencing even more severe dysgenic pressures than Whites" (p. 341). Part of the problem may be differences in reproductive strategies among the races, according to J. Philippe Rushton's theory discussed in the book (pp. 642-643). Herrnstein and Murray mention Rushton's theory that Blacks have the largest genitals and the highest frequency of sexual intercourse among the three major races (p. 642). Consistent with customary academic standards of scholarly objectivity and neutrality, Herrnstein and Murray reserve judgment on whether Rushton is right or wrong: "We expect that time will tell whether it [Rushton's theory] is right or wrong in fact" (p. 643).
In addition to supposed downward pressures on the distribution of intelligence in this country produced by high fertility rates in Blacks, Herrnstein and Murray believe that Latinos are also experiencing more severe dysgenic pressures than Whites (p. 341) and that Latino immigration is putting downward pressure on the distribution of American national intelligence. So should we be worrying about dysgenic pressure on our national cognitive intelligence? They conclude, "Putting the pieces together--higher fertility and a faster generational cycle among the less intelligent and an immigrant population that is probably somewhat below the native-born average--the case is strong that something worth worrying about is happening to the cognitive capital of the country" (p. 364).
The authors present a large number of research analyses that they performed themselves, in which they pit parental socioeconomic status (SES) against IQ on a variety of economic and social behaviors. They conclude that the major cause of economic and social behaviors is IQ, not SES. The authors' research analyses are based on data collected in the National Longitudinal Survey of Youth (NLSY). None of their research analyses on the relation between IQ, SES, and social behaviors has ever been published in peer-reviewed scientific journals. The Bell Curve is written for the general reader and does not assume that the reader has had a course in statistics. The authors have even included an appendix for those readers who are sure they can not learn statistics, titled "Statistics for People Who Are Sure They Can't Learn Statistics" (Appendix 1, pp. 553-567). Scientists first publish their research in peer-reviewed scientific journals, not in books written for the general reader who may not have the technical background needed to detect flaws in data and misinterpretations of data analyses. It is inappropriate for a scientist to do otherwise.
Herrnstein and Murray's research analyses--never published in peer-reviewed scientific journals--investigate the relation of IQ and SES to marriage, to divorce, to illegitimacy, to welfare dependency, and to parenting. They conclude that IQ is the primary problem, not SES: "People with low cognitive ability tend to be worse parents" (p. 232). The authors believe that low birth weight and high infant mortality are indications of poor parenting and are probably caused by "prenatal negligence" (p. 233) on the part of mothers with low cognitive ability rather than inadequate prenatal medical care on the part of society. They also present unpublished research analyses on the relation between crime and low cognitive intelligence, and between civility and high cognitive intelligence. "A smarter population is more likely to be, and more capable of being made into, a civil citizenry" (p. 266), according to the authors.
In the final part of The Bell Curve, titled "Living Together," Herrnstein and Murray propose a solution to the supposed dysgenic downward pressures on our national intelligence caused by the large number of children born to "low-IQ women," and to the recent large-scale Latino immigrations to the United States. They argue that America's current fertility policy "subsidizes births among poor women, who are disproportionately at the low end of the intelligence distribution" (p. 548). They seem to urge eugenic foresight to counteract dysgenic pressure: "We urge generally that these policies, represented by the extensive network of cash and services for low-income women who have babies, be ended" (p. 548). With regard to the supposed dysgenic effects of Latino immigration on national intelligence, their central thought about immigration "is that present policy assumes an indifference to the individual characteristics of immigrants that no society can indefinitely maintain without danger" (p. 549). "But," they conclude, "we believe that the main purpose of immigration law should be to serve America's interests" (p. 549). For those members of groups who will not be excluded from the American dream by eugenic foresight or new immigration laws, Herrnstein and Murray propose "that group differences in cognitive ability, so desperately denied for so long, can best be handled--can only be handled--by a return to individualism" (p. 550).
Who are the authors of The Bell Curve? Are they right? The first author, Richard Herrnstein, was a professor of psychology at Harvard University for 36 years. He died a very short time ago. One would presume that The Bell Curve represents Herrnstein's final summing up of a lifetime of objective scholarly research published in peer-reviewed scientific journals on the genetic basis of IQ. Regrettably, the media seem to be totally unaware of the fact that the deceased Harvard professor never published any scientific research on the genetic basis of IQ and its relation to race, poverty, or social class in peer-reviewed scientific journals in his entire 36-year academic career. Richard Herrnstein's actual area of expertise is the experimental analysis of decision making in pigeons and rats, and he never studied the genetic basis of any behavior in those laboratory animals. The first presentation of his theory on the genetic basis of IQ, social class, and poverty appeared in a magazine article titled "I.Q." published in the September 1971 issue of the Atlantic Monthly magazine. As we all know, scientists publish their data and theories in peer-reviewed scientific journals or in peer-reviewed technical books, not in popular magazines or in nontechnical books written for the general reader.
In 1973, Herrnstein published a nontechnical Atlantic Monthly Press book titled I.Q. in the Meritocracy that expanded on his theory of the genetic basis of IQ and poverty. Herrnstein had never collected data on IQ, so the book drew on the work of others, especially the "data" of Sir Cyril Burt. According to Leslie Hearnshaw (1979), Burt's biographer and distinguished historian of British psychology, Burt had probably invented much of his highly cited data on the genetic basis of IQ. While doing research on Burt's data for an article that I later published in Science (1978), I discovered that Herrnstein had in fact laundered Burt's own descriptions of Burt's widely publicized and highly cited study "Intelligence and Social Mobility" (Burt, 1961). Burt had described his own study "merely as a pilot inquiry" (p. 9) and his data as "crude and limited" (p. 9). Burt had not even reported the number of subjects he had tested in his crude and limited study. In describing Burt's study, however, Herrnstein (1973) failed to tell the reader about the deficiencies that Burt, himself, had mentioned. In addition, Herrnstein (1973) said Burt's sample size was "1,000" (p. 203), later revising that figure to "40,000" in response to criticism. In reply to a critical letter by Jerry Hirsch (1975), Herrnstein (1975) revised his 1973 figure: "It is true that Burt's sample was 40,000, not 1,000 as I said" (p. 436), while failing to acknowledge that Burt had never reported the number of subjects he had tested. Leon Kamin (1974) appears to have been the only psychologist to notice and publicly report that Burt had failed to give the sample size of his celebrated 1961 study of IQ and social mobility. Presumably, Herrnstein and other psychologists who had publicized the results of that study had never noticed that Burt had not reported the sample size of his famous study.
The second author of The Bell Curve, Charles Murray, has a doctorate in political science from the Massachusetts Institute of Technology and is currently a Bradley Fellow with the American Enterprise Institute, a conservative research group in Washington, DC. Murray often publishes his research and theories in The Public Interest (e.g., Murray, 1994), a neoconservative magazine edited by Irving Kristol, also a fellow of the American Enterprise Institute and sometimes considered the founding father of neoconservatism (Atlas, 1995). In an article recently published in The Public Interest, Murray listed the first priority of his political agenda: "And so I want to end welfare" (1994, p. 18). Inasmuch as the media sometimes refer to The Bell Curve as Murray's book, perhaps the book represents Murray's summing up of a body of objective scholarly research that he had published in scientific journals on the genetic basis of IQ and poverty. But like his coauthor Richard Herrnstein, Murray has never conducted or published any research in scientific journals on the genetic basis of IQ and poverty in his entire career.
The Bell Curve is not a scientific work. It was not written by experts, and it has a specific political agenda. Still, it is possible that the major scientific premises of the book may be correct. If two monkeys were put before a typewriter, it is theoretically possible for those two monkeys to produce a Shakespearean sonnet. Perhaps Herrnstein and Murray produced a valid scientific work. I will now evaluate the major premises of The Bell Curve.
In 1972, Leon Kamin exposed the empirical unsoundness of the most important evidence in support of the IQ hereditarian position, Sir Cyril Burt's data (Hearnshaw, 1979). He later published his results in a book attacking Burt's data as well as the secondary sources who publicized those data (Kamin, 1974). In 1979, Leslie Hearnshaw (1979) published a biography of Burt in which he concluded on the basis of personal diaries and other material that it was highly likely that Burt had fabricated some of his most celebrated data. Hearnshaw, distinguished historian of British psychology, delivered the memorial address at Burt's Memorial Service and was later asked by Marion Burt, Burt's sister, to write a full-length biography of Burt. The result was the well-known Cyril Burt: Psychologist (1979). In their discussion of the Burt affair, Herrnstein and Murray suggest that some of Burt's "leading critics were aware that their accusations were inaccurate" (p. 12), suggesting a possible conspiracy against Burt. There is, however, no mention whatsoever of Hearnshaw's book in their half-page synopsis of the Burt affair, and Hearnshaw's book does not appear anywhere in their 57-page bibliography of references. This misrepresentation of the Burt affair by omission of important historical facts is not uniquely associated with The Bell Curve. In 1982, Richard Herrnstein published an article in The Atlantic Monthly in which he attacked the media for misrepresenting the evidence in the IQ controversy (Herrnstein, 1982). In that magazine article, the Harvard professor wrote "that most psychometricians had stopped trusting Burt's data years before, partly because of inconsistencies first noted in a 1974 article by Arthur Jensen" (p. 70), while omitting any mention of Leo Kamin, the psychologist who in reality first noted inconsistencies in Burt's data.
The distribution of IQ test scores cannot be expected to follow a bell curve unless it is constructed by the tester to do so (Dorfman, 1978). The shape of the distribution of IQ test scores will depend on the average difficulty of the test items as well as their intercorrelations. The high item intercorrelations in IQ tests imply that the IQ distribution can take a variety of shapes. The central limit theorem does not apply to random variables with positive intercorrelations (Lamperti, 1966). Frederic Lord (1952), one of the fathers of modern test theory and former president of the Psychometric Society, gave results on this question: "The results given are sufficient to show that the distribution of test scores cannot in general be expected to be normal, or even approximately normal. The question naturally arises as to what possible shapes the frequency distribution fs, as given in (76) [Lord's Equation (76)], may assume. The answer is that this function may assume any shape whatsoever, provided the item intercorrelations are sufficiently high" (Lord, 1952, pp. 32-33). The symbol fs refers to the distribution of test scores.
The book uses factor analysis to infer the existence of a single hypothetical general factor of cognitive intelligence that is presumed to account for most of cognitive performance. One of the problems with factor analysis as a tool for determining the underlying structure of a system is that neither the factors nor the loadings are uniquely defined if you have more than one factor (Lawley & Maxwell, 1963), and it is difficult to determine if you have only one factor. In experimental cognitive psychology, factor analysis is virtually never used as a tool to determine the underlying cognitive structure. It is a tool for correlational cognitive psychology, not experimental cognitive psychology. I inspected the subject index of some well-known texts in experimental cognitive psychology and found that the term factor analysis never appears in the subject index (e.g., see Anderson, 1985; Matlin, 1994; Reed, 1982). Why not? Kendall and Stuart (1966) may provide the answer: "Application of the same technique [factor analysis] to physical systems very often results in weighted sums of variables to which no clear interpretation can be given" (p. 310). In short, "The main difficulty, as a rule, is to know what the results mean" (p. 310), Kendall and Stuart point out.
The most direct way of estimating heritability is from data on monozygotic twins reared apart (MZA) and separated in early infancy (Bouchard, Lykken, McGue, Segal, & Tellegen, 1990). This MZA design allows for the estimation of heritability if the following major assumptions are met: (a) environments are a random sample from the population of environments, (b) genotypes are a random sample from the population of genotypes, (c) there is no genotype-environment correlation, and (d) there is no genotype-environment interaction. If the pairs of MZAs differ in age, then these assumptions will not be met. If these assumptions are met, then the intraclass correlation between IQ scores of MZA twin pairs directly measures heritability. Sir Cyril Burt's (1966) study of 53 MZAs appears to have met the first three assumptions. Unfortunately, Burt's data appear to have been invented (Hearnshaw, 1979). Bouchard et al.'s (Minnesota) survey of MZAs provides the next best data set. Unfortunately, to the best of my knowledge, the detailed case-study records of the Minnesota MZAs have never been released and have therefore not been subjected to public scrutiny to determine the degree to which assumptions have been met and the degree to which the MZAs told the truth to the Minnesota group. Finally, if there is genotype-environment interaction--then the fourth assumption is not met--and heritability is undefined. But this is the most controversial assumption underlying the MZA design. Herrnstein and Murray present no convincing evidence to justify the fourth assumption.
The authors have made a fundamental error well-known by professional geneticists. It is sometimes called "Jensen's error." Jensen made that error in his famous 1969 Harvard Educational Review article. The critical importance of that error was first clearly illuminated by Roger Milkman, a professor of biology at the University of Iowa and a world authority on population genetics and evolutionary biology. The article, "A Simple Exposition of Jensen's Error," was published in the Journal of Educational Statistics in 1978 (Milkman, 1978). Melvin Novick was editor of that journal when Milkman's article was published. Novick, professor of statistics and education at the University of Iowa at the time, later became president of the Psychometric Society. What is Jensen's error? It is that within-race heritability has no implications for between-race heritability. The Bell Curve is therefore flawed with regard to inferring between-race heritability in IQ from within-race heritability in IQ.
Herrnstein and Murray use logistic regression to determine which is more important--IQ or SES--in determining socially undesirable behaviors. Logistic regression is a form of regression in which the dependent variable is binary. In all of their analyses, they assume a simple additive model in which the logit (a transform of the sample proportion) is assumed to equal B0 + B1IQ + B2SES + B3 age + random residual [numbers after Bs should read as subscripts]. They assume no IQ-SES interaction. They use the standardized beta weights to determine the relative importance of IQ and SES in determining the probability of various undesirable or desirable behaviors. Unfortunately, IQ and SES are highly intercorrelated (collinearity).
There are two major problems with Herrnstein and Murray's attempts to determine whether IQ or SES is more important. First, there is the collinearity problem. Weisberg (1985) describes the collinearity problem in linear regression: "When the predictors are related to each other, regression modeling can be very confusing. Estimated effects can change magnitude or even sign depending on the other predictors in the model" (p. 196). Next, there is the problem of deciding that the predictor with the largest standardized beta weight is the most important. Weisberg describes why this approach is faulty: "Unfortunately, this logic is faulty because the scaling depends on the range of values for the variables in the data" (p. 186). Perhaps these are the reasons why Herrnstein and Murray never published their logistic analyses in peer-reviewed journals.
Were Herrnstein and Murray as lucky as the proverbial monkeys at a typewriter? That depends on your point of view.
Anderson, J. R. (1985). Cognitive psychology and its implications (2nd ed.). New York: W. H. Freeman.
Atlas, J. (1995, February 12). The counter counterculture. The New York Times, pp. 32-39, 54, 61-62, 65.
Bouchard, T. J., Lykken, D. T., McGue, M., Segal, N. L., & Tellegen, A. (1990). Sources of human psychological differences: The Minnesota study of twins reared apart. Science, 250, 223-228.
Burt, C. (1961). Intelligence and social mobility. The British Journal of Statistical Psychology, 14, 3-24.
Burt, C. (1966). The genetic determination of differences in intelligence: A study of monozygotic twins reared together and apart. British Journal of Psychology, 57, 137-153.
Dorfman, D. D. (1978). The Cyril Burt question: New findings. Science, 201, 1177-1186.
Hearnshaw, L. S. (1979). Cyril Burt: Psychologist. London: Hodder & Staughton.
Herrnstein, R. J. (1971, September). I.Q. The Atlantic Monthly, 43-64.
Herrnstein, R. J. (1973). I.Q. in the meritocracy. Boston: Atlantic-Little, Brown.
Herrnstein, R. J. (1975). Herrnstein replies. Contemporary Psychology, 20, 436.
Herrnstein, R. J. (1982, August). IQ testing and the media. The Atlantic Monthly, 68-74.
Hirsch, J. (1975). Hirsch on Herrnstein and Jensen. Contemporary Psychology, 20, 436.
Jensen, A. R. (1969). How much can we boost IQ and scholastic achievement? Harvard Educational Review, 39, 1-123.
Kamin, L. J. (1974). The science and politics of I.Q. New York: Wiley.
Kendall, M. G., & Stuart, A. (1966). The advanced theory of statistics (Vol. 3). New York: Hafner.
Lamperti, J. (1966). Probability: A survey of the mathematical theory. New York: W. A. Benjamin.
Lawley, D. N., & Maxwell, A. E. (1963). Factor analysis as a statistical method. London: Butterworth.
Lord, F. M. (1952). A theory of test scores. Psychometric Monographs (Whole No. 7).
Matlin, M. W. (1994). Cognition (3rd ed.). Fort Worth, TX: Harcourt Brace.
Milkman, R. (1978). A simple exposition of Jensen's error. Journal of Educational Statistics, 3, 203-208.
Murray, C. (1994, Spring). Does welfare bring more babies? The Public Interest, 17-30.
Reed, S. K. (1982). Cognition: Theory and applications. Monterey, CA: Brooks/Cole.
Weisberg, S. (1985). Applied linear regression (2nd ed.). New York: Wiley.
Richard J. Herrnstein (deceased) was Edgar Pierce Professor of Psychology at Harvard University (Cambridge, Massachusetts) and is author of IQ in the Meritocracy and coauthor, with J. Q. Wilson, of Crime and Human Nature and, with E. G. Boring, of A Sourcebook in the History of Psychology. Herrnstein is coeditor, with R. C. Atkinson, G. Lindzey, and R. D. Luce, of Stevens' Handbook of Experimental Psychology, Vol. 1: Perception and Motivation, and Vol. 2: Learning and Cognition (2nd ed.); and, with M. L. Commons, S. M. Kosslyn, and D. B. Mumford, of Quantitative Analyses of Behavior, Vol. 8: Behavioral Approaches to Pattern Recognition and Concept Formation, and Vol. 9: Computational and Clinical Approaches to Pattern Recognition and Concept Formation.
Charles Murray, currently a Bradley Fellow at the American Enterprise Institute, is author of Losing Ground: American Social Policy 1950-1980.
Thomas J. Bouchard, Jr., professor of psychology at the University of Minnesota (Minneapolis) and director of the Minnesota Center for Twin and Adoption Research, is immediate past president of the Behavior Genetics Association and American Psychological Association Distinguished Scientist Lecturer for 1995. Bouchard is author of the chapter "The Genetic Architecture of Human Intelligence" in P. E. Vernon (Ed.) Biological Approaches in the Study of Human Intelligence and of the chapter "IQ Similarity in Twins Reared Apart: Findings and Responses to Critics" in the forthcoming R. J. Sternberg and E. L. Grigorenko (Eds.) Intelligence: Heredity and Environment. Bouchard is coeditor, with P. Propping, of Twins as a Tool of Behavior Genetics.
Donald D. Dorfman, professor of psychology and radiology and faculty member in the graduate program in applied mathematical and computational sciences at the University of Iowa (Iowa City), is author of the chapter "Group Testing" in S. Kotz and N. L. Johnson (Eds.) Encyclopedia of Statistical Sciences, Vol. 3 and coauthor, with J. T. Cacioppo, of the chapter "Waveform Moment Analysis: Topographical Analysis of Nonrhythmic Waveforms" in J. T. Cacioppo and L. G. Tassinary (Eds.) Principles of Psychophysiology: Physical, Social, and Inferential Elements.