Jensen (1998, 1999) offers a comprehensive presentation of the argument that the g-factor, as defined by hierarchical common factor models, constitutes the core and major component of human intellective function. Its validity includes matters of evolution and inheritance of g with attendant consequences for both individual and group differences such as racial differences. The future is envisioned as elucidating the details of the genetic and brain elements of g. The uninformed reader would have no hint that there is also a century of work which can be cited against the argument. The g model is not consistent with mainstream twentieth century work in evolution and in genetics. Some of the main points of conflict are examined.
2. The core around which the book is built is the view that Spearman discovered a general factor, g, and claimed that this was the major element in all areas of human intellective function. It is held that, with the research of an intervening century, the legacy of this view is now largely uncontested. Hierarchical factor analysis has succeeded the two-factor theory but g remains with us. Characterising all humans, this general factor must be genetic and evolutionary in origin both on theoretical and empirical grounds. It follows that the future of research on intelligence lies in explicating the details of the gene architecture and brain correlates for g.
3.The opening history chapter is limited to Spearman's contribution to the g heritage. He generated a quarter of a century of vigorous interest. Not mentioned is the following half a century of decline and disinterest (Cronbach, 1990). Neisser and Bouchard (1999) correctly credit the beginning of the resurrection of interest in g to Jensen and his Harvard Educational Review paper (1969). Reviving an historic concept calls for historic context.
4. For historical context, one should understand the milieu within which g was conceived. The word "gene" did not exist in 1904. Just on the verge of rediscovery, Mendel's ideas were generally unknown or forgotten. Evolution was the great new concept of biological science. Genetics was a lesser and only remotely, if at all, related field. It would be a quarter of a century before Fisher (1930) would bring them together. Genetic phenomena were understood to be governed by "blended inheritance" where the genotype was the mid-parent value; offspring inherited the average of their parents' characteristics. Given this linear model, the genetic or "factor" structure of inherited characteristics could be inferred from the correlation matrix of measures of the phenotype. Karl Pearson's Biometrika, was the leading journal for disseminating work following the "biometric" model. Spearman's model followed directly from this "biometric" model. Galton's original concept of a quantitative model for the life sciences - "anthropometrics" - had led to a model more focused on biological processes - "biometrics" - and had now spawned "psychometrics", extending the biometric genetics model to psychological processes. Correlation analysis has remained the core of psychological measurement ever since. The first book on psychological measurement (Thorndike, 1904) confirmed psychometrics as its own field in the same year as Spearman's (1904a,b) basic papers. Thereafter psychometrics developed independently of biometrics.
5. Meanwhile, in genetics, Mendel was rediscovered and the nature of his theory began to be recognised. In what has generally been considered one of the half dozen most important papers in the history of science, Fisher (1918) asked what would be the consequences for quantitative genetic analysis if inheritance were Mendelian rather than biometric. He proved that if inheritance is Mendelian one cannot infer genetic structure from the correlation matrix but one can deduce the correlation matrix if genetic structure is known. With only minor modifications, the analyses of that paper still define the main concepts of population genetics today. The linear models of correlational analysis and their underlying assumptions were replaced by Fisher's analysis of variance model with its components of variance and with the nonlinear concept of interaction necessary to account for such phenomena as dominance, epistasis, maternal effects, and genetic-environmental interaction and correlation effects. The impact went far beyond genetics as correlational analyses of experimental data were rejected to be replaced by components of variance methods ranging from t-tests to MANOVA. Despite Jensen's belief in the genetic basis of g the fact remains that a genetic foundation for g cannot be inferred from any correlational structure, including the correlational structure which defines g.
6. The term factor analysis usually is restricted to the common factor model. This is a linear model of estimates of underlying hypothetical error free variables. It is not a model of the actual empirical data. Present day factor analysis is oriented toward confirmatory analysis which tests whether an hypothesised factor structure is consistent with observed data. Fisher's demonstration that one cannot infer genetic structure from the correlation matrix proved the matrix is not unique. An infinite number of genetic structures are consistent with any given matrix. Confirmatory is a misnomer. The usual logic of hypothesis testing is to frame the hypothesis of interest against the null hypothesis. Rejecting the null confirms the substantive. Factor analysis reverses this logic by interpreting acceptance of the null as confirmation of the substantive hypothesis.
7. In defining terms, Jensen notes an essential empirical data precondition for the estimates of factor analysis: "Factors arise only from the reliable or nonchance correlation between abilities. Now if it were the case that tests were constructed of only those items that happened to be correlated with one another (and items that did not were discarded), factors would indeed be mere psychometric artifacts. That is, factors would be no more than a product of the arbitrary way that ability items are devised or selected for inclusion in psychometric tests (p. 56)." But this is exactly what one does in item analysis! The success of the Stanford-Binet approach is attributable to Terman's (1916) careful attention to internal consistency, making it one of the cornerstones of psychometrics. Trying out items with selection and discard based on item correlations is a major part of the standardised test construction enterprise.
8. Tests are intended to sample performance in some aspect of the test taker's environment. Evolution occurs because individuals with different characteristics perform differently in different environments. Natural selection reflects an advantage for a particular characteristic in a particular environment. For genetic characteristics then, the necessary condition for natural selection is the existence of such genetic-environmental interactions. Thus, from a genetics perspective, a test item samples the performance of an individual in some niche of her world. Accordingly, if performance has any hereditary component then one expects genetic-environmental interactions. In theory, it is possible that some performances have no such interactions, but those of us who work with infrahumans would be hard put to identify a characteristic that has not been found susceptible to some degree of selective breeding. The existence of genetic-environmental (item x genotype) interactions implies that item selection will be weighted by the genetic interactions of the tested population. This has been verified experimentally.
9. The genetic controls necessary for an experimental study are attainable only with infrahuman species. (Some who have proposed human racial groupings for the purpose have not appreciated the fact that racial definition is not genetic definition.) Using six well defined genotypes of rats we formed multiple mixed populations differing in proportions of the six in each (Harrington, 1975/1982, 1984, 1988). Maze performance tests were developed separately for each population using conventional item analysis for selection. New samples from each genotype were tested on all of the tests and also on a set of maze tests defined as the criterion to be predicted. Performance of each genotype on each test was correlated with the representation of the genotype in the test base population. This was true not only for mean performance level but also for predictive validity, which is, of course, statistically independent. From Jensen's statement above, the results are empirical evidence that factors are artifacts.
10. The laboratory experiments have a second implication that is noteworthy. Test performance is correlated with genotype membership in the test base population. Then all tests developed on a given or comparable test base population will share a common genotype representation effect and must be positively correlated with each other. The tests would yield a general factor on factor analysis. Thus standard psychometric test construction procedures create a general factor as an artifact.
11. Jensen sees g as a product of human evolution and inheritance. He observes that behavioural functions involving g involve more brain processes organised in more complex ways than other processes. Both in factor analysis and in neurophysiological organisation, g is a higher order function. In noting the role of intelligence in determining man's place in nature, the message is clear that it is g which has emerged as the higher order function. Jensen seems reluctant to think of it as occurring in lower species at times. At other times he recognises that if it evolved there had to be precursors in other species. A species unique higher intellect function engages a century old argument originally espoused by creationists in opposition to Darwinian theory. When Fisher (1918) showed that correlational analysis failed for Mendelian Inheritance, there ensued a quarter of a century of controversy with Pearson and other defenders of biometric inheritance. One argument was that the higher mental functions of humans were not subject to Mendelian inheritance but to biometric inheritance, and thus subject to correlation matrix analysis. In general the biometricians argued that complex mental functions were higher order functions requiring complex tests and subject to different mechanisms of inheritance requiring different modes of analysis.
12. The chapter on heritability is technically accurate. However it perpetuates the use of heritability coefficients for human data, an inappropriate usage. Read Kempthorne (1978) for the authoritative geneticist's critique of application of the heritability coefficient to human data. Jensen is to be congratulated for making clear the difference between hereditary and heritability - hereditary referring to the biological source of a characteristic and heritability to its variance. He uses the example I thought I had originated: number of heads, hands, or feet is inherited but, since the variability is near zero the heritability is near zero.
13. All but the very careful reader may miss the significance of the coverage of assortative mating as a component of heritability. Whether or not this usage is appropriate, any heritability coefficient for IQ will be high because of the assortative mating component. The place of genetics in evolution was first set forth by Fisher (1930). High selection for one head, two hands, or two feet eliminates other numbers of extremities reducing the associated variance. The "Fundamental Theorem of Natural Selection" is that heritability varies inversely with evolutionary fitness. Jensen believes g is a broad fitness factor too recently evolved to show reduction of variance by selection. Prediction of future evidence of fitness strikes me as soothsaying and as ignoring the question of how it evolved to this point.
14. Brown and Thomson (1921) showed a general factor will occur in a correlation analysis if the observed effects are attributable to a large number of underlying causal influences. Similar views were championed by Tryon (1932a,b, 1935). Guilford (1954, p.476) explained that the sampling theory alternative to g never gained acceptance among psychometricians because: "In criticism of the sampling theory, it may be said that there seems to be little likelihood of demonstrating experimentally the existence of the elements hypothesized. . . It is in repeated compounds that we find the invariances such as we seek in science." That anti-reductionist view was reinforced by the abject failure of an almost monolithic investment of many years of the research resources of the entire psychological community in sensori-motor elements - a disaster in that it yielded nothing (Wissler, 1901). Global approaches carried the day and only a minority followed Brown or Tryon in a belief in many factors or elements. Today the Human Genome Project reflects a massive investment in a pursuit of elements. A third or more of those elements are thought to be related to brain processes. Because of current successes in illuminating many disease and other processes, contemporary neurobiology is driven by molecular biological concepts and research strategies which are diametrically opposed to those global approaches which Guilford saw as dominant in psychometrics.
Brown, W., & Thomson, G. H. (1921). The essentials of mental measurement. Cambridge: Cambridge University Press.
Cronbach, L. J. (1990). Essentials of psychological testing (5th ed.). New York: Harper & Row.
Fisher, R. A. (1918). The correlation between relatives on the supposition of Mendelian inheritance. Transactions of the Royal Society (Edinburgh), 52, 399-433.
Fisher, R. A. (1930). The genetical theory of natural selection. Oxford: Clarendon Press
Guilford, J. P. (1954). Psychometric methods (2nd ed.). New York: McGraw-Hill.
Harrington, G. M. (1982). Intelligence tests may favour the majority groups in a population. In J. Hirsch & T. R. McGuire (Eds.), Benchmark papers in behaviour . Vol. 16. , Behaviour-genetic analysis. New York: Academic Press. (Reprinted from Nature , 1975, 258 , 708-709)
Harrington, G. M. (1984). An experimental model of bias in mental testing. In C. R. Reynolds & R. T. Brown (Eds.), Perspectives on bias in mental testing (p. 101-138). New York: Plenum Press.
Harrington, G. M. (1988). Two forms of minority test bias as psychometric artifacts with an animal model. Journal of Comparative Psychology, 102, 400-407.
Jensen, A. R. (1969). How much can we boost IQ and scholastic achievement? Harvard Educational Review, 39, 1-123.
Jensen, A. (1998). The g Factor: The Science of Mental Ability. Praeger
Jensen, A. (1999). Precis of: The g Factor: The Science of Mental Ability. PSYCOLOQUY 10(23). 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
Kempthorne, O. (1978). Logical, epistemological and statistical aspects of nature-nurture data interpretation. Biometrics, 34, 1-23.
Neisser, U., & Bouchard, T. J. (1999). Two views about the g factor. Contemporary Psychology, 44(2), 131-135.
Spearman, C. (1904a). "General intelligence" objectively determined and measured. American Journal of Psychology, 15, 201-292.
Spearman, C. (1904b). The proof and measurement of association between two things. American Journal of Psychology, 15, 72-101.
Terman, L. M. (1916). The measurement of intelligence. Boston: Houghton-Mifflin
Thorndike, E. L. (1904). An introduction to the theory of mental and social measurements. New York: Science Press.
Tryon, R. C. (1932a). Multiple factors vs two factors as determiners of ability. Psychological Review, 39, 324-351.
Tryon, R. C. (1932b). So-called group factors as determiners of ability. Psychological Review, 39, 403-439.
Tryon, R. C. (1935). A theory of psychological components-an alternative to mathematical factors. Psychological Review, 42, 425-454.
Wissler, C. (1901). The correlation of mental and physical tests. New York: Columbia University.