The g factor per se, one of the most important constructs in psychology, has never been mystifying to anyone except those who either don't understand it or wish to influence others to reject it because of its centrality in a broad nexus of educational and social correlates. A careful reading of "The g Factor" (Jensen, 1998a; 1999) should utterly dispel the idea that there is anything in the least 'mystical' or mysterious about g. The fact that the brain variables involved in the cause of g are as yet poorly understood doesn't make it mystical, nor does it nullify what is already known about g at the psychometric, behavioral, and behavior-geneti c levels, including its correlations with many socially significant phenomena. The main thesis underpinning nearly all the points of Richardson's (1999) argument that "IQ tests are merely clever numerical surrogates for social class" is demonstrably false.
2. The IQ of school-age children is correlated, on average in numerous studies, about +.40 with their parents' SES. In adults, by the time they have attained their final career lines, the correlation between IQ and SES is about +.70. School-age children's own IQ is a much better predictor of the SES they attain as adults than is the SES of the parents who reared them. The predictive validity of childhood IQ for predicting adult SES applies also to full siblings who are reared together and hence have the same SES as children.
3. The IQs of adoptees in samples where the adoptive homes span a range of SES from blue-collar working class to professional and managerial occupations show virtually zero correlation with the SES of their adoptive parents, but they show a significant correlation with the SES and IQ of their biological parents. By late adolescence, virtually all of the nongenetic variance in IQ is within-family variance; the between-families component of IQ variance almost completely vanishes (see Jensen, 1998a, Chapter 7). And of course it is the between-families component of variance that would necessarily embrace the effect of SES on IQ if any such effect existed.
4. If IQ differences in the population were entirely caused by differences in SES, and if SES differences resided entirely in the between-families component of the population variance in IQ, then we should expect a much larger proportion of the IQ variance to reside in the between-families variance than in the within-families variance. Yet the fact is that the total IQ variance is apportioned approximately 46 percent within-families and 51 percent between-families. This is based on the average value of the IQ correlation between full siblings reared together, which in all the studies reported in the literature (comprising 27,000 sibling pairs) is +.49. This is equal to the proportion of the total IQ variance that exists between-families, hence the remaining proportion of the IQ variance (i.e., .51) is divided between the within- families variance (.46) and measurement error (estimated as.05).
5. An important question for psychometrics is whether the IQ variance that exists between-families differs in factorial composition from the IQ variance that exists within-families. It is possible to factor-analyze the correlations that exist entirely within-families among a number of diverse tests. Such within-family correlations completely exclude the influences of between-family differences in SES (or any other variables that may differ between families). It is also possible to factor-analyze the between-families correlations among the same battery of tests. The key question, then, is whether the g factor of this battery remains the same g factor when it is extracted from the matrix of SES-free within-family correlations as when it is extracted from the between-family correlations.
6. I investigated this on all the families with two or more children in grades 1 to 6 in a California school district with an especially wide range of SES. I found that the g factor obtained from within-family correlations and the g factor obtained from between-families correlations were virtually identical (methodology explicated in Jensen, 1980a; summarized in Jensen, 1998, pp. 170-173). In other words, a battery of psychometric tests of mental abilities measures one and the same g factor both within-families and between-families: the mental test differences between individuals from different families have the same factor structure, hence the same meaning psychometrically, as the differences between siblings reared together.
7. Moreover, this was true both within and between the white and black populations in this school district. Individual differences in psychometric g, regardless of race or SES, are differences in the same g factor. A well-known and frequently cited adoption study (Capron & Duyme, 1989, 1996), based on a full cross-fostering design, estimated the magnitudes of both the genetic and the environmental effects on the IQs of children whose biological parents were either of low or of high SES and whose adoptive parents were either of low or of high SES. Both the SES of the adoptees' genetic background and the SES of their adoptive environments showed significant effects (with genetic > environmental) on Wechsler Full Scale IQ. However, it was shown that the g factor of the Wechsler subtests was significantly related only to the genetic component of the IQ difference. The SES effect of the rearing environment was not reflected in the g component of IQ. In other words, the SES of the biological parents predicted the g aspect of IQ, while the SES of the adoptive parents affected only the non-g aspect of IQ (Jensen, 1998b). This finding was not surprising in view of the prior discovery that various tests' g loadings significantly predict those tests' heritability coefficients (Jensen, 1998, pp. 182-189).
8. All these items of evidence noted in the preceding paragraphs clearly indicate that the causal pathway mediating the correlation between psychometric g and SES goes in the direction from the level of g to SES, although SES probably has some effect on test specificity, particularly specific aspects of the knowledge content of some verbal tests. Hence SES differences are generally larger on verbal tests involving specific knowledge content than on equally g-loaded nonverbal tests, in which the tests' specificity is less related to the person's SES background.
9. Richardson's (par. 2 - 7) arguments fall back on the disproved claim that mental tests are culturally biased across SES groups and racial and ethnic groups. This is to ignore the important distinction between (1) culture loading and (2) culture bias (Jensen, 1980b). These are separate concepts. Operationally defined measurements of them may or may not be correlated. This is an empirical question, answerable in terms of a host of psychometric criteria (Jensen, 1980b; summarized in Jensen, 1998, pp. 360-367). The consensus of hundreds of studies of test bias is that the standardized mental tests in common use today, whether culture-loaded or culture-reduced, are not culture-biased for any native born, English speaking groups in the United States. The same is true for nonverbal tests in groups for which English is not the primary language. The notion that tests are biased to favor of the ethnicity or social class of the test constructors is contradicted by the fact that Asian groups in Asia and in America perform, on average, at least as well or better on European and United States tests constructed by non-Asian psychometricians compared to the normative groups of European ancestry in the United States population. Chinese school children, many with first-generation immigrant parents living in relatively low SES 'Chinatown' neighborhoods in California cities, out-score white children in higher SES neighborhoods on the Raven Matrices, a nonverbal reasoning test. And Arctic Eskimos have been found to score at least on a par with Scottish and Canadian norms on the Raven Matrices.
10. The claim that trait anxiety or state anxiety in test situations accounts for SES or racial-ethnic differences in test performance is purely ad hoc. Although some small percentage of individuals is anxious in test situations to a degree that hinders test performance, I have found no evidence of group differences in test anxiety with respect to either SES or racial-ethnic groups that could account for the observed group mean differences in test scores. Anxiety as well as other personality variables account for some small proportion of test score variance, lowering the test score's g construct validity to some degree, but no systematic relationship of this variance to group SES or racial group differences has been demonstrated. The Yerkes-Dodson law, which relates task performance to drive or anxiety level and task complexity was originally discovered by manipulating drive level in rats (Yerkes & Dodson, 1908). It has long been known that the Yerkes-Dodson law applies to virtually all forms of sympathetic arousal, including anxiety, which acts much like a drive. Experiments have borne out predictions based on the Yerkes-Dodson law of the relationship between anxiety (or neuroticism) and mental test performance (Eysenck, 1967). The law is highly relevant to the 'stereotype threat' hypothesis of the Black-White difference in test performance (Jensen, 1998, pp. 513-515).
11. There is absolutely nothing that Richardson (pars. 16 - 17) has to say about genetics that in the least contradicts or is incompatible with the massive evidence we now have on the heritability of IQ and especially of g. If Richardson has any real evidence that human variability in intelligence, IQ, or g does not have substantially heritability, it would be headline news in the scientific world. The genetic variance of fitness characters is not necessarily reduced to zero or even to very low values. R.A. Fisher has pointed to human fertility as a prime example of an obvious fitness character, yet it has a narrow heritability estimated at about .40, which is about the same as the narrow heritability of IQ. (The broad heritability of IQ in adults is closer to .70.). Many traits are maintained in the population as balanced polymorphisms, for which there is stabilized selection for different alleles that either enhance or depress the phenotypic expression of the trait. Examples are birth weight, adult height (which has a heritability coefficient over .90), and IQ. In such cases, heterozygous genes have greater fitness than homozygous genes.
12. Even in livestock, after intensive artificial selection over many generations for a given trait (e.g., egg-laying capacity in chickens, milk yield in cows, back fat in pigs, or beef yield in cattle), the traits maintain moderate heritability, which, under relaxed selection, rapidly increases. Selection reduces mainly the additive component of genetic variance, thereby causing the interactive or nonadditive components (dominance and epistasis) to constitute a larger proportion of the total genetic variance. Less advantageous alleles at a given locus are gradually eliminated by natural selection when heterozygosity lends no advantage to fitness. Heterozygosity with moderate to high heritability is maintained in polygenic traits with balanced selection, such as height and IQ. At this point in history, however, arguments that genetic factors do not play a major role in human variation in mental abilities, particularly in the component of test score variance identified as g, can truly be likened to the creationists' rejection of evolution by natural selection.
13. Richardson's ( par. #16) uninformed statement that "Jensen's claim that humans have 100,000 polymorhphic genes seems ridiculous" is simply wrong, although the precise proportion of polymorphic genes is wholly irrelevant to any substantive or theoretical issue in my book. The best present estimate of the total number of genes in the human genome is indeed 100,000, but this is still considered a "soft" figure. And it is a fact that virtually all (i.e., at least 99.9 percent) of the gene loci in the human genome are polymorphic: that is, they have at least two or more alleles with different effects and therefore contribute to the total genetic variance in the population. Some alleles, however, are very rare in the population and some are even lethal before their hosts reach reproductive maturity. So geneticists can claim different proportions of the genes that are polymorphic by using different definitions of polymorphism, such as saying that polymorphic genes are those whose least frequent allele has a frequency in the population of, say, less than 0.1%, or < 1%, or <5%, or < 10%, whereby at each of these increasing frequencies there is a decreasing percentage of all the genes that are classified as polymorphic. Most of the textbooks published before about 1995 are out-of-date on this subject, since the rapid pace of advances in research on the human genome in just the last few years has made it clear that there is much greater genetic variability in the population than was previously expected. The calculated probability is essentially zero that any two individuals ever born since the advent of Homo sapiens, except monozygotic siblings, would have the same alleles at all loci in the human genome.
14. Richardson's opinions regarding the genetics and evolution of human abilities depend upon ignoring the "total evidence rule" and are way out of tune with mainstream science on these issues. For those seeking a dependable, up-to-date, and highly readable introduction to behavioral genetics, I would suggest the volume edited by Plomin and McClearn (1993).
Capron, C., & Duyme, M. (1989). Assessment of effects of socioeconomic status on IQ in a full cross-fostering design. Nature, 340, 552-553.
Capron, C., & Duyme, M. (1996). Effect of socioeconomic status of biological and adoptive parents on WISC-R subtest scores of their French adopted children. Intelligence, 22, 259-275.
Eysenck, H.J. (1967). Intelligence assessment: A theoretical and experimental approach. British Journal of Educational Psychology, 37, 81-98.
Jensen, A.R. (1980a). Bias in mental testing. New York: Free Press.
Jensen, (1980b). Uses of sibling data in educational and psychological research. American Educational Research Journal, 17, 153-170.
Jensen, A.R. (1998a). The g factor: The science of mental ability. Westport, CT: Praeger.
Jensen, A.R. (1998b). Adoption data and two g-related hypotheses. Intelligence, 25, 1-6.
Jensen, A.R. (1999). Precis of: "The g Factor: The Science of Mental Ability" PSYCOLOQUY 10(023). ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1999.volume.10/ psyc.99.10.023.intelligence-g-factor.1.jensen http://www.cogsci.soton.ac.uk/cgi/psyc/newpsy?10.023
Plomin, R., & McClearn, G.E. (Eds.), Nature nurture and psychology. Washington, D.C.: American Psychological Assoication.
Richardson, K. (1999). Demystifying g. PSYCOLOQUY 10(048) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1999.volume.10/ psyc.99.10.048.intelligence-g-factor.5.richardson http://www.cogsci.soton.ac.uk/cgi/psyc/newpsy?10.048
Yerkes, R.M., & Dodson, J.D. (1908). The relation of strength of stimulus to rapidity of habit formation. Journal of Comparative Neurology and Psychology, 18, 458-482.