Jensen's elaborate thesis on g can be shown to be based on several fallacious premises. IQ tests are merely clever numerical surrogates for social class. The numerous correlations evoked in support of g arise from this. His 'genetic' arguments are based on a highly simplistic, and outmoded, model of genes. And his model of "race" is based on evolutionary misconceptions.
1. In this book Jensen (1998, 1999) pursues his well-known arguments about g, a 'general, cognitive factor'. But it isn't difficult to show that what is cognitive is not general, and what is general is not cognitive. Scores on standardised psychometric tests intercorrelate partly because they have been subjected to considerable construction engineering on the basis of common criteria. Jensen himself has noted how 'every item is carefully edited and selected on the basis of technical procedures known as "item analysis", based on tryouts of the items on large samples and the test's target population' (1980:145). Even so, because item designs tend to be intuitive, and the criteria for item selection statistical and pragmatic, rather than theoretical, there is sill much puzzlement about what the common factor actually is. Other cognitive theory might help us in this regard.
2. For example, a prominent line of study in recent years has shown how different patterns of cognition arise, not from individual computations, but from an internalisation of the cultural 'tools' (patterns of activity, knowledge and reasoning) dominant in the social world in which people grow up and/or currently operate. 'The structure of thought depends upon the structure of the dominant types of activity in different cultures' (Luria 1976: xiv-xv). Because test constructors come from a narrow social group, it follows that test items will contain information structures which will match the background knowledge of some children more than that of others. This cognitive match/mismatch will apply even more critically to non-verbal items than to verbal items.
3. Take, for example, the Raven's test which Jensen says is almost a pure measure of g. According to Carpenter et al (1990: 408), after an examination of Raven's personal notes, 'the description of the abilities that Raven intended to measure are primarily characteristics of the problems, not specifications of the requisite cognitive processes... he used his intuition and clinical experience... without regard to any underlying processing theory'.
4. Inevitably, Raven's 'intuition' will have included his own cultural tools, and illustrating their incursion in the test is not too difficult. Much of middle class culture is based on the manipulation of symbols (e.g. words, numbers) in two-dimensional array on paper. Typical cultural tools are record cards, tables with rows and columns of totals and subtotals, timetables, and so on. These nearly all require the reading of symbols from top left to bottom right, the induction of additions, subtractions and substitutions across columns and down rows, and the deduction of new information from them. These are precisely the kinds of manipulations (or 'rules') that found their way into Raven's items.
5. So although the symbols are experience-free, the rules governing their changes across the matrix are certainly not, and they are more likely to be already represented in the minds of children from middle class homes that in others. Performance on the Raven's is not a question of inducing novel rules from meaningless symbols, but ones which are culturally rooted; each item presents a recognition problem before it is a reasoning problem. The latter is easy when the former has been achieved.
6. This has been shown in a vast variety of tasks in which subjects can map the covariation relations in the task onto relations in their background knowledge (reviewed in Richardson 1999). These include the Wason selection task; computerised 'games' governed by complex 'rules'; pragmatic reasoning schemes; analogical reasoning tasks; class-inclusion and scientific reasoning tasks; categorisation tasks; and modified Raven's matrices. All of this explains why the Raven's (and other non-verbal tests), often referred to as culture-free etc., are, in fact, the most enculturated of all tests.
7. So relative acquisition of relevant background knowledge (which will be closely associated with social class) is one source of the elusive common factor in psychometric tests. But there are other, non-cognitive, sources. Jensen seems to have little appreciation of the stressful effects of negative social evaluation and systematic prejudice which many children experience every day (in which even superficial factors like language dialect, facial appearance, and self-presentation all play a major part). These have powerful effects on self concepts and self-evaluations. Bandura et al (1996) have shown how poor cognitive self-efficacy beliefs acquired by parents become (socially) inherited by their children, resulting in significant depressions of self-expectations in most intellectual tasks. Here, g is not a general ability variable, but one of 'self-belief'.
8. Reduced exposure to middle-class cultural tools and poor cognitive self-efficacy beliefs will inevitably result in reduced self-confidence and anxiety in testing situations. There is a well-known association between IQ test performance and test-anxiety. In his meta-analysis of 562 studies, Hembree (1988) found that High Task Anxiety (HTA) subjects hold themselves in low esteem, fear exposure to negative evaluation, experience greater emotional reaction to testing situations, and more encoding difficulty and other cognitive interference when tested. It will not do for Jensen to attempt to dismiss the role of task-anxiety by reference to the old Yerkes-Dodson Law (which is about drive) and a study involving a small group of university students!
9. In sum, the 'common factor' which emerges in test performances stems from a combination of (a) the (hidden) cultural content of tests; (b) cognitive self-efficacy beliefs; and (c) the self-confidence/freedom-from-anxiety associated with such beliefs. In other words, g is just an mystificational numerical surrogate for social class membership. This is what is being distilled when g is statistically 'extracted' from performances. Perhaps the best evidence for this is the 'Flynn effect,' (Fkynn 1999) which simply corresponds with the swelling of the middle classes and greater exposure to middle-class cultural tools. It is also supported by the fact that the Flynn effect is more prominent with non-verbal than with verbal test items - i.e. with the (covertly) more enculturated forms.
10. Once we see g as a variable of class-cultural characteristics, instead of a mystical biological power, the many other correlations which Jensen reports are demystified. We also see the diverse ways in which correlations can be interpreted. It is not the least bit surprising that g also correlates with head size, brain size, stature, general health, and so on, which, through nutritional, endocrinal, and other aspects of social privilege/exclusion, are also correlates of social class.
11. Jensen relies heavily on the (weak) associations between performances on Elementary Cognitive Tasks and IQ. But such performance will be much influenced by task-anxiety, as numerous studies on speeded tasks have shown (Hembre 1988). Since HTA produces more erratic reaction times, this would explain why the biggest correlate of g is not mean (or median) speed of response, but response variation. It also explains the lack of correlation between Nerve Conduction Velocity (however crudely measured) and RT.
12. Although the correlation between IQ scores and school performance is one deliberately built into tests, it produces large 'knock-on' effects, such as a built-in correlation with occupational status. Further correlations are built in by the fact that g also reflects cognitive self-efficacy beliefs and self-confidence/freedom-from-anxiety. This will explain the (weak) correlation between IQ and rate of learning (or job training), and also why such associations crease with task complexity.
13. When we turn to job performance the picture becomes very murky, not least because of serious methodological problems and contradictory findings. The 'job performance' measure used in nearly all studies is that of supervisor ratings. But supervisors can be rather subjective, use widely different criteria, with 'halo', age-related, and other effects. In the Schmidt et al (1986) study, supervisor ratings had very low correlations (around 0.3) both with subjects' job knowledge and actual work samples! The (weak) associations between statistically abstracted g and job performance may, again, stem from differences in self-concept, self-confidence, anxiety etc., rather than from an 'ability' variable. This interpretation is supported by reports that, when workers have been in the job for some time, performance is completely uncorrelated with IQ (Hulin et al 1990). Jensen dismisses that idea, citing a meta-analysis by Schmidt et al (1986). But that study was conducted on military personnel, in which, as the authors themselves suggest, job performance involves 'standard operating procedures' routinized by 'thorough, detailed training programs'. Besides, the Manual to the Raven's Progressive Matrices (RPT) Test (which Jensen sees as a test of 'pure' g ) warns us that 'the predictive validity of the RPM... to success within an occupation... is relatively low' (Raven et al 1993, 41).
14. Although a multitude of imponderables remain in correlational data of this kind, it seems reasonable to suggest that IQ predicts little that isn't already built into the test directly or indirectly by virtue of its being a surrogate for social class. It should also be obvious that people who, from a very early age, have reduced self-expectations and self-esteem, and fewer chances of self-fulfilment, are also, in the long run, going to exhibit more social pathology.
15. It is very worrying to find a simplistic 'Mendelian' model of independent and additive genes still being urged upon us by Jensen. The 'genetic beanbag' view is clung to because it furnishes the only paradigm in which Jensen and coworkers can work 'genetically'. In particular, it furnishes the famous 'expected' correlations for relatives (e.g. monozygotic versus dizygotic twins) which form the basis of 'heritability' estimates, even though doubts about the model for complex characters have frequently been expressed (see e.g. Barton & Turelli 1989).
16. Indeed, recent molecular biology has shown better than ever how genes for evolved characters have become intricately tied in with adaptable regulatory systems across the genome as a whole. Under these regulations, variable alleles can be utilised for common ends, or common alleles utilised for divergent ends, as developmental needs dictate. Up to 90% of genes are regulatory in function, and not structural alleles at all (Jensen's claim that humans have 100,000 polymorphic genes seems ridiculous). Phenomena such as canalization, divergent epigenesis, exon-shuffling (which modifies gene-products to suit current developmental needs), and even developmental modification of gene-structures themselves, now make a nonsense of the idea of a one-to-one relationship between incremental accumulations of 'good' or 'bad' genes, and increments in a phenotype (see e.g. Rollo 1995). This makes the objective of most twin and adoption studies surrounding IQ a red herring, because it is attempting to 'prove' a genetic model that no one can seriously believe in.
17. Jensen argues that g has evolved as a 'fitness' character. Yet it is the logic of natural selection that fitness characters come to display little if any genetic variation. This has been repeatedly confirmed in artificial selection experiments, and in the wild. The self-defeating logic of Jensen's argument is obvious. Indeed, I find it amazing that, at the end of the twentieth century, complex, sophisticated edifices like this are being constructed on such patently erroneous foundations.
18. Jensen argues, in effect, that cognitive 'races' exist because genes related to human cognitive systems will have been subjected to diversifying selection in the same way as some superficial physical or physiological characters. He suggests that northern migrants would have faced particularly difficult conditions. As a result, groups of African descent will have lower frequencies of genes for superior cognitive abilities, compared with those of Caucasian or Mongoloid ancestry.
19. This completely misses the point. Our African hominid ancestors themselves evolved as a social-cooperative species in order to deal with conditions of extreme environmental uncertainty, as the climate dried, forests thinned, and former forest dwellers were 'flung out' onto the open savannah or forest margins. It is crucial to point out that when even as few as two individuals cooperate they create a new, social environment that is vastly more complex than anything experienced in the physical world. It is that complexity on the social plane which rapidly impelled the tripling of brain size and furnished the unique cognitive capacity for dealing with complexity in general - in the physical world as well as the social.
20. The uniquely adaptable, highly selected, socio-cognitive system that resulted was a prerequisite, not a consequence, of human migration patterns. Although inhabiting every possible niche, humans have only a quarter of the genetic variation of highly niche-specific chimpanzees (Kaessmann et al 1999). The system operates on a completely different plane from blind genetic selection - one which can 'model' the world conceptually, and anticipate and change it. If our heads get cold we invent hats, rather than wait for natural selection to reshape our skulls and increase the size of our brains (which is what Jensen suggests in one particularly questionable y line of argument). As Owens & King (1999) point out, what minor genetic differences exist are 'quite literally superficial... the possibility that human history has been characterised by genetically homogeneous groups ("races") distinguished by major biological differences, is not consistent with genetic evidence'.
21. Owens & King also point out that 'Of course prejudice does not require a rational basis, let alone an evolutionary one, but the myth of major genetic differences across "races" is nonetheless worth dismissing with genetic evidence' (453). This culmination of Jensen's thesis, then, is as hollow as the conceptual foundations on which it based. It really is time this negative and fatalistic model of humanity was put behind us once and for all.
Bandura, A., Barbarinelli, C., Caprara, G.V. & Pastorelli, C. (1996). Multifaceted impact of self-efficacy beliefs. Child Development, 67: 1206-1222.
Barton, N.H. & Turelli, M. (1989). Evolutionary quantitative genetics: how little do we know? Annual Review of Genetics, 23: 337-370.
Flynn, J. R. (1999) Searching for justice: the discovery of IQ gains over time. American Psychologist, 54:5-20.
Hembree, R. (1988). Correlates, causes, effects, and treatment of test anxiety. Review of Educational Research, 58: 47-77.
Hulin, C.L., Henry, R.A. & Noon, S.L. (1990). Adding a dimension: time as a factor in the generalizability of predictive relations. Psychological Bulletin, 107: 377-389.
Jensen, A.R. (1980). Bias in mental testing. New York: Free Press.
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
Kaessmann, H., Weibe, V. & Paabo, S. (1999). Extensive nuclear DNA sequencing diversity among chimpanzees. Science, 286: 1159-61.
Luria, A.S. (1976). Cognitive development: its social and cultural foundations. Cambridge, M.A.: Harvard University Press.
Owens, K. & King, M-C. (1999). Genomic views of human history. Science, 286, 451-453.
Raven, J., Raven, J.C. & Court, J.H. (1993). Manual for Raven's Progressive Matrices and Vocabulary Scales: Section 1. Oxford: Oxford Psychologists Press.
Richardson, K. (1999). The making of intelligence. London: Weidenfeld & Nicolson.
Rollo, D.C. (1995), Phenotypes: their epigenesis, ecology and evolution. London: Chapman and Hall.
Schmidt, F.L., Hunter, J.E. & Outerbridge, A.N. (1986). Impact of job experience and ability on job knowledge, work sample performance, and supervisor ratings. Journal of Applied Psychology, 71: 432-439.