Jejudo, South Korea
From Skeptic vol. 4, no. 3, 1996, pp. 108-109. The following article is copyright ©1996 by the Skeptics Society, P.O. Box 338, Altadena, CA 91001, (626) 794-3119. Permission has been granted for noncommercial electronic circulation of this article in its entirety, including this notice.
BOOK REVIEW: The Bell Curve Cracks
Inequality by Design-Cracking the Bell Curve Myth, by Claude S. Fischer, Michael Hout, Martin Sanchez Jankowski, Samuel R. Lucas, Ann Swidler, and Kim Voss. Princeton University Press, 1996. 318 pages.
Reviewed by Brian Siano
A mountain peak, and its darker silhouette displaced by 'one standard deviation,' is the image from The Bell Curve that persists in memory. It's the purported shape of the distribution of IQ scores classified by race, derived from the scores of 11,878 people taking the Armed Forces Qualification Test (AFQT), and it appears in TBC's notorious "Ethnic Differences in Cognitive Ability" chapter. (It also appeared in the New Republic's cover story on the book, as well as in the pages of Skeptic,) One could easily imagine the shapes rising into view as Microsoft Excel or SAS scattered the points on the computer screen, like a lost Aztec city shining through a false color enhancement of satellite data. The symmetry of the slopes, of course, spoke for a natural evenness of distribution- the classic "bell curve" one sees for height, life expectancy, shooting craps, and other natural processes. The conclusion: an underlying pattern has been revealed through the ruthless application of statistics.
On page 32 of Inequality by Design, the authors provide a re-casting of the same magic runes. As before, the whites-only Matterhorn of Herrnstein and Murray is present. But beside it is a grey lump, slouching towards the high ends of the scale in defiance of the demanded symmetry. This insubordinate lump is the original distribution of scores on the AFQT.
Looking further in The Bell Curve for an explanation, one finds that answers are somewhat elusive. Herrnstein and Murray could have used centile scores-placing people in the 99th percentile of scores, the 98th percentile, etc. This would have been a lot simpler. In Appendix 2, they state they "we knew from collateral data" that the important IQ stuff "occurs at the tails of the distribution," and "using centiles throws away the tails." In short: the original scores are not a bell curve; but IQ scores must follow a bell curve; all the action in our project happens at the tail ends of a bell curve; therefore, we must derive a bell curve with distinct tails from the unruly data. If you're one of those people who feel that data should shape the theory, this may seem somewhat less than valid. Toss in the fact that intelligence tests are frequently designed to provide bell-curve distributions of scores, and we notice a kind of circular reasoning implicit in the psychometric model used by Herrnstein and Murray. Reshaping the lump into the mountain was, in the words of Inequality by Design, the result of "a good deal of statistical mashing and stretching," demanded by the assumption "that intelligence must be distributed in a bell curve."
The authors take Herrnstein and Murray to task for presenting the AFQT as an intelligence test. This isn't the case: the AFQT was designed to predict performance in the armed forces (no wisecracks, class), and it functions best as a test of the level of schooling the subject has received. The math sections, which make the greatest differences in the final scores, require having had exposure to high school algebra. The National Longitudinal Survey of Youth, which administered the AFQT to its thousands of subjects, measured schooling at a very crude level (number of years and whether the subject was in an academic track), but these two factors correlated well with AFQT scores. Other studies, using the same data, argue that Herrnstein and Murray drastically underestimated the influence of schooling.
(In one passage, the authors argue that if we keep education level constant, age correlates negatively with AFQT score. In other words, older respondents with 12 years of education score lower than younger respondent with the same level. If AFQT measures education level, then this difference is explained by the fact that the older respondents had been out of school longer, and had forgotten some of their education. But if we accept the psychometric claim that the test measures innate ability, then we'd have to believe that people start getting stupider around their midteens.)
What can we say about Herrnstein and Murray's "cognitive elite," that upper 5%? They were people who had had schooling beyond the high school level, disproportionately male (thanks to the weighting in favor of math), and just plain lucky; one or two more wrong answers, and they would've dropped down into the "bright" category. As for the other end of the curve, 27 % had dropped out of school at least three years before taking the test.
The core of Herrnstein and Murray's argument is that IQ is a better predictor of life outcomes than the usual measure of socioeconomic status (SES). Contrary to the forbidden-data claims of The Bell Curve, sociologists have been working with intelligence tests and the AFQT for years. The authors of Inequality by Design "accepted Herrnstein and Murray's evidence, their measure of intelligence, and their basic methodology and then reexamined the results. By simply correcting a handful of errors, we showed that coming from a disadvantaged home was almost as important a risk factor for poverty as a low AFQT score."
Herrnstein and Murray defined SES very narrowly, as four factors: level of education, income, and parents' occupation(s). They then needlessly compiled these into a single index-thus giving them equal weight amongst each other. This is a major error, since NLSY data shows that parental income has a far greater effect than parental education on a child's life outcome.
Also, when such information was missing from the NLSY respondents, they simply assigned them the average value derived from other respondents. (This introduces error, in the case of respondents who are rich or poor, and reduces the statistical association with effect variables.) The NLSY only included four questions about parental SES, which makes it far less reliable than the 105-question AFQT-which stacks the IQ-vs-SES face-off in Herrnstein and Murray's favor.
Herrnstein and Murray also left out several factors known to have effects on a subject's life outcome. The number of siblings, for example, was not incorporated into their analysis. The adult community environment-local unemployment rate, geographic region (rural, urban, suburban)-was also overlooked. Was the subject, at age 14, living in a two-parent household? What about access to quality schools? Bringing these factors into play provides a probability-of-poverty graph that matches Herrnstein and Murray's AFQT curves. In short, the authors conclude that the subjects' "life chances depend on their social surroundings at least as much as their own intelligence... The key finding of The Bell Curve turns out to be an artifact of its method."
Inequality by Design goes on to present a far more detailed analysis of poverty. For example, despite rough parity in IQ scores, women are far more likelier than men to be poor. Using the NLSY-AFQT data, the authors state that "a young woman would have had to score forty-one points higher on the AFQT than a young man of the same age, formal schooling, and background in order for her risk of being poor to have been as low as his." [Italics in original.] Having children increases the risk of being poor. The economic effects of marriage and divorce are more dramatic for adults who grew up in low-income families. Herrnstein and Murray say nothing substantial about gender; instead, they argue that unmarried status is a result of lower intelligence. But the AFQT scores of unmarried respondents were no higher than those of marrieds....
Again, data seems to have lost primacy over theory through much of The Bell Curve. In one spectacular example, Herrnstein and Murray claim that a three-point drop in average American IQ would increase unwed motherhood and incarceration rates by about 10 % (pp. 364-368). But to explain the doubling of incarcerated men in the 1980s, the 150% increase in unwed motherhood since 1970, and the rises and dips of poverty between 1960 to 1992, we'd have to believe that average IQ varies as much as 55 points within a single generation-a claim Herrnstein and Murray explicitly rule out.
At this stage, discussion of The Bell Curve's notorious race-and-IQ chapter is almost a done deal. Inequality by Design echoes Berkeley sociologist John Ogbu's work on the matters of caste systems in societies. Yes, American blacks score one standard deviation below whites on IQ tests. This is more a result of low ethnic or caste status than it is of any biological factor. Polish Jews emigrating to America tested poorly on IQ tests, and Koreans living in Japan have patterns of poverty, crime and school performance similar to those of blacks in America. Low caste status contributes to three factors-socioeconomic deprivation, group segregation, and the stigma of inferiority. Castes assimilated through conquest or capture (Irish in Britain, Maori in New Zealand, Koreans in Japan, blacks in the U.S.) have a harder time assimilating than minorities who immigrated willingly. And when we consider the arbitrary distinctions that mark the Indian "untouchables" and Japanese burakumin as low-caste, we are forced into realizing that race matters only as much as people want it to matter.
When The Bell Curve was published in 1994, few reviewers were equipped to re-evaluate the data in time for publication deadlines. Some admitted ignorance of statistics, while others rested on denunciations of the book's logical fallacies, implied policy demands, the precedent of the previous century's race-science, and the book's reliance on white supremacist-funded research sources. The more thoughtful reviews went over the classic criticisms: the misuse of the concept of heritability and debates over the validity of IQ tests and Spearman's g. A more detailed analysis of the data would face the prospect that The Bell Curve's impact wouldn't last very long, and a rebuttal book published a year or so down the road would be irrelevant.
So we can be thankful that Inequality by Design's detailed and scrupulous re-analysis of Herrnstein and Murray's g-over-all theories has been published. (Welcome to peer review, guys.) But the authors go on to present a good discussion of how economic and social factors shape inequality in the United States. We don't succeed or fail because of our individual IQ or talents alone: nearly all of us are beneficiaries of a vast system of social support, ranging from tax breaks for homeowners, the national highway system, and grammar school immunizations, to the WIC and food stamp programs. True, there are people who do stand on their own two feet, free from reliance on society. Problem is, they also sleep on steam vents.