Jensen asserts that general intelligence (g) exists; that it is largely biological and immutable; that racial differences in intelligence test scores reflect differences in underlying cognitive ability, are genetic in origin, and are probably immutable. I argue that one must distinguish between factors underlying g. A substantial part of the variation in intelligence is biological, but modifiable. Racial differences in cognitive skills do exist but they can be modified to some extent by social means. The extent to which racial differences are biological is unknown at present, and speculation without evidence is unwise.
2. In 1969 Arthur Jensen published an article in the Harvard Educational Review (Jensen, 1969) that reviewed the then-existing evidence indicating that all of these bad things (at least from the social egalitarian's viewpoint) are true. Jensen's (1998) current book summarises thirty more years of research, which he concludes establishes scientific evidence for the existence and importance of general intelligence, its biological and largely genetic origin, and the putative genetic origins of differences in cognitive competence between racial groups. There are no surprises here for anyone who has followed Jensen's work. Therefore a review of the book provides a chance to comment on the ideas that Jensen and his not-inconsiderable group of followers have espoused.
3. Reactions to Jensen's ideas have, predictably, fallen into two classes. He has been vilified and threatened. 'Jensenism' has, to some, become a modern term almost on a par with Nazism. Jensen himself, and a not-inconsequential group of psychometricians, psychologists, and sociologists, see some of the criticism as attacks on a reputable scientist whose research has turned up some unpleasant truths. In 1988, a full issue of the journal Intelligence was devoted to comments on his work. There he was referred to as a 'King among men' for his unswerving pursuit of the truth. Now, in The g factor, Jensen makes his own case. In evaluating it I shall take a middle ground. (There are times when I feel very lonely doing this.)
4. Jensen's treatise is not a book on Intelligence. It is a book presenting Jensen's view of intelligence. If you want a text with a balanced view, use Mackintosh's (1998) even-handed presentation. However, I do not think that Jensen intended to write a textbook; he wanted to present a case. His case is built around three themes, and several subthemes.
5. Jensen's first theme is that there is a general factor in intelligence. He is undoubtedly correct. Virtually all aspects of cognitive performance are positively correlated, especially in the "lower half" of ability. Truly high performance tends to be more specialised, a point that will be important later. However there can be no doubt about the statistical finding.
6. Jensen explains the statistical finding by assuming that it reflects the influence of a pervasive general intelligence, the g factor. It is not clear that this is the case. After a massive review of the psychometric literature, Carroll (1993) concluded that intellectual performance is best characterised by a three-tier theory. At the bottom are narrowly defined abilities (e.g., speed of word retrieval). Above that there are broad general factors, most prominently fluid and crystallised intelligence, and visual-spatial reasoning. For those not familiar with the terminology, fluid intelligence is, loosely, the ability to deal with new and unfamiliar problems and crystallised intelligence is the ability to bring previously acquired problem solving techniques to bear on the present problem. Because the broad general factors are statistically associated, you can place a single factor at the top. A similar argument was made by R. B. Cattell (1971) and by John Horn (1985; Horn & Noll 1994).
7. The question is whether the correlation between the broad general factors is causal, because of an underlying general intelligence factor, or whether it is a statistical artifact, because the second order factors depend (to different degrees) upon the same underlying factors, such as pattern recognition, working memory, and the control of attention. Jensen favours the causal explanation. Others (notably Horn) do not. I side with Horn, and here is why.
8. In experimental psychology one of the ways of showing that two behaviours are reflections of different causal mechanisms is to show that there are independent variables that affect the two behaviours in different ways. The same logic is used in neuropsychology, where the independent variable is brain damage and the phenomenon is called double dissociation. There are two areas in which interactions between types of intelligence tests and other variables have been well established. The first is ageing. It is well known that over the adult years scores on fluid intelligence tests drop. On the other hand, scores on verbal comprehension and general knowledge tests increase or stay constant until old age. Next, suppose we look at the cohort effect, the fact that, with age held constant, test scores have been rising in the industrial nations over the past fifty years (Flynn, 1987; Hertzog, 1994). Scores on tests that measure fluid intelligence have risen substantially. Scores that emphasise crystallised intelligence, and scores on acquired problem solving skills, such as text comprehension, have shown little if any rise.
9. If a neuropsychologist were to observe these patterns in patients characterised by different types of brain damage, the neuropsychologist would conclude that the behaviours were subsumed by different brain systems. I suggest that these interactions, observed on a societal rather than an individual basis, constitute substantial evidence for distinct second order cognitive talents that are statistically but not causally linked.
10. Another of Jensen's arguments for the causal nature of general intelligence is that intelligence test scores are correlated with a variety of elementary cognitive tasks, most notably reaction time and inspection time tasks. This is consistent with Carroll's conclusions regarding three levels of cognitive abilities. The facts are clear. Jensen goes beyond the facts to conclude that these correlations reflect basic physiological processes associated with neural efficiency. This conclusion bothers me, because there are two major holes in the evidence.
11. First, the elementary cognitive tasks do not seem to be closely correlated with each other. If Jensen's argument is correct they should be. (However I think we need much more evidence before a definitive statement can be made.) Second, much of the evidence for reaction time correlations rests on studies by Jensen and his colleagues using a particular procedure, which he describes in detail. The important thing about this procedure is that Jensen provides many fewer practice trials than is customary in reaction time research. (You have to go back to the original literature to see this, and of course the exact statement varies with the study in question. However, strong practice effects have been observed using Jensen's procedures and apparatus (Widaman and Carlson, 1989). What Jensen's procedure may be picking up is a bit of fluid intelligence as the participant tries to figure out how to work the gadget the psychologist designed. Also, suppose that we accept Jensen's argument that elementary cognitive tasks correlate with intelligence tests because both depend on neural efficiency. If this were true, then the correlations between intelligence test scores and reaction time tasks should go up over practice, for as the participant figures out how to work the apparatus his or her performance should depend more and more on neural efficiency limits rather than strategy choices. But the opposite occurs, over extended practice the correlations between intelligence test scores and elementary task performance go down (Ackerman, 1989). This is a serious embarrassment for Jensen's approach.
12. Having said this, are the different varieties of intelligence important? You bet they are. Psychometric tests of cognitive ability are generally the best predictors that we have of both academic and workplace performance. They far outperform personality tests, and are only equalled by carefully structured interviews or by work samples. Jensen is quite correct in emphasising the importance of whatever is evaluated by the tests.
13. Jensen's second theme, genetics as a cause of intelligence, is his second most contentious point. Massive studies have shown that genetics accounts for about half of the variance in test performance. The effects seem to be highest in childhood and then again in great age. Jensen often refers to 'adults.' This is a bit misleading. The evidence for the post-sixty generations is strong. The data for individuals in their working years (25-60) are sparse, for the simple reason that it is very hard to get a representative sample of working adults. They are too busy to sit down for psychological testing.
14. No environmental variable has ever been found that begins to account for as much variance in test performance as genetic variables. Now one can always explain this away by reference to unmeasured variables, but this really will not do. In my own opinion environmental hypotheses are not going to have much influence until we develop a theory of the environment that approaches the sophistication of our theories of genetics. Put another way, I know what percent of my genes I share with my brother and my cousins. I do not know how to measure the similarity between their social-intellectual environment and my own.
15. As is well known, the expression of genetic traits can be altered by environmental factors. Therefore showing that intelligence is partly genetically determined does not mean that it cannot be changed. But how? Jensen claims, with good evidence, that none of our programmes intended (partially) to alter cognitive skills have had many long term effects. That is, we do not know how to create an infant-pre-school-school- work environment that will increase intelligence.
16. But we do know that some aspects of intelligence can be changed. First, the existence of the cohort effect shows that intelligence can be altered, for some changes in our society are altering it. In this respect, I note, along with Teasdale and Owen (1989) that the primary cohort gains are at the bottom. That is, we do not have as many low scorers as we used to. (Gains have been shown at the top, but they are much smaller.) Also, it is quite clear that schooling affects some of the test scores that Jensen is willing to accept as appropriate tests of intelligence. The case has been best made for the data from the National Longitudinal Study of Youth (NLSY), where it has been shown that an extra year of schooling does influence scores on the Armed Forces Vocational Aptitude Battery (ASVAB) (Cawley et al., 1997). This is hardly surprising to anyone who knows what is on the ASVAB, a lot of its content is related to school subjects. Because Jensen is willing to commit himself to the ASVAB, and to the NLSY data set, when it suits his purposes, he must deal with the schooling effects demonstrated for this test and data set.
17. Parenthetically, I believe that the schooling effects and the cohort effect may be related. Since World War II every industrialised nation has increased its requirements for mandatory education and has gone to great lengths to reduce drop-outs and to offer 'second chance' schooling for adults who do drop out. Given the amount of money spent, one hopes that there would be an effect on people in the lower range of test scores. Apparently such an effect exists.
18. To sum up this part of the argument, Jensen is right in stating that intelligence has a strong genetic component. He is probably wrong in saying that intelligence cannot be altered, but right in saying that we do not know how to alter it. Certainly the present evidence suggests that massive early childhood intervention may have only a transitory effect on cognitive skills. But may I point out that providing every child with a safe, healthy, supportive environment is a good thing to do, regardless of whether or not it affects their IQ test score ten years later?
19. The third theme is race. Jensen's position on racial differences in intelligence is by far the most contentious issue in this book and in his career. While it is hard to separate the scientific, political, and social issues I shall try.
20. Jensen's argument has changed very little since his famous (or infamous, depending on your view) Harvard Educational Review article thirty years ago. He asserts that there is an I.Q. score gap of slightly over one standard deviation unit between White and African- Americans, that the gap is constant, that it reflects real differences in cognitive skills and, in this book, that the gap is largely genetic. He admits that solid proof for the last statement is lacking, but he claims that genetic causation should be the 'default hypothesis' until it is disproved.
21. If you combine Jensen's beliefs about racial differences with his belief about the immutability of intelligence, you are forced to the conclusion that some racial groups are unchangeably more cognitively competent than others. Does Jensen accurately present the data, and is the default hypothesis a useful summary of our scientific understanding?
22. The test score gap between Whites and African-Americans is real. Furthermore, it almost certainly reflects a cognitive-skills gap. The gap appears on tests that are intentionally designed to measure cognitive skills related to the workplace, such as the National Assessment of Educational Progress (NAEP). The gap is important, for validity studies indicate that the tests predict academic and workplace performance of minorities about as well as they predict the performance of Whites.
23. On the other hand, and in contradiction to Jensen's statements, the gap is clearly decreasing. While the gap varies from test to test, I think that the fairest summary of the current situation is that the gap was as much as 1.25 standard deviation units fifty years ago, and is about .8 units today. Therefore the one standard deviation unit statement is a good summary of the overall data, but it does ignore a trend. Also, Jensen cites studies in South Africa, during the apartheid era, showing a gap of more than two standard deviation units between White and Black South Africans. He emphasises the size, I would emphasise the difference between the gap in South Africa in those days and in the United States today.
24. The fact that the gap is decreasing is in agreement with the cohort effect. If the main effect of cohort changes has been to increase the score of the bottom groups, and if African Americans were concentrated toward the bottom of the distribution fifty years ago, then the White-African American gap is going to decrease. Since the decrease is taking place over only a few generations, it is hard to account for the all of the gap on a genetic basis. Some of it might be genetic, but we do not know how much.
25. Jensen rests his case for the genetic hypothesis on three facts about White vs. African American scores, and an inference from a fourth fact. Test items and tests that have the lowest apparent cultural loadings (e.g., the Raven Matrix tests) show the greatest disparity between racial groups. This rules out the hypothesis that the test gap is due to the test's content differentially favouring the white over the African-American culture. Test items and tests that show the highest g loadings also show the greatest disparity between groups. This suggests to Jensen that the gap is due to racial differences in general intelligence. And recall that Jensen sees general intelligence as being largely genetically determined. Finally, the factor structures obtained from testing in the African American population are essentially the same as those obtained in the White population. Jensen combines these observations with the evidence for a genetic basis for intelligence, largely gathered in the White population, to conclude that the White-African American gap is also of genetic origin.
26. One can evaluate the evidence in a slightly different way. To me it appears that the biggest gap is on tests that evaluate fluid intelligence. This is not a play on words. It may be that we can obtain some leverage on the general skills gap by finding ways to inculcate higher degrees of crystallised intelligence, i.e., by better education. If fluid intelligence is low, a teacher who wishes to raise students' crystallised intelligence faces a formidable challenge. Formidable challenges are just that, difficult but not impossible. Whether or not society wishes to expend the resources to overcome this challenge is a social, not a scientific, question.
27. The third fact, about the identity of factor structures, seems to me to be incongruous. Studies completed in both Europe and the U.S. in the past ten years have shown that the general intelligence factor is more pervasive at the low end of the distribution than it is at the high end. Therefore, if the African-American population falls below the White population intelligence tests should be more saturated with general intelligence in the African-American than White populations. However Jensen (pg.374 ff. claims, correctly I believe, that the factor structures are essentially the same. This is consistent with the argument that different underlying causes are responsible for inter-group mean differences and intra-group individual variation.
28. Finally, there is the exclusionary evidence. At present, no one has produced evidence for a specific environmental or social cause that could explain racial disparities in intelligence. This is partly because genetic and social variables are confounded. Should Mother's Education be used as a surrogate for mother's level of acquired cognitive status or mother's genetic makeup? It depends on your point of view. The other part of the difficulty has to do with the theoretical and practical constraints on the studies. We do not have good theories of the relation between social psychology and individual differences; the experimental studies strike me more as investigation of (sometimes insightful) hunches than careful analyses of theoretical implications. It is fairly easy to carry out attitudinal studies in college sophomores, it is much harder to carry out those same studies with adults, in the workplace, where the problem is.
29. In order to carry out a direct test of Jensen's hypothesis we would have to know what genes are related to intelligence test scores, and then determine whether the allele frequencies for those genes differed in different racial groups. There are excellent reasons for locating genes related to test scores and, indeed, some such studies have been carried out. They indicate (no surprise) that there is no one gene for intelligence, but that there may be hundreds, each of which makes a small contribution. I am sure there will be steady, but possibly slow, progress.
30. So, it may be that sometime between 25 and 50 years from now we will be able to compare allele distributions in different racial groups. Jensen says that as a default we should assume that there will be differences. I fail to see any compelling reason for having a default hypothesis at all. And while I would (and do) support the studies to identify the relevant genes, I would not put very much money into the studies to identify different distributions across racial groups. I do not see what the purpose would be.
31. Suppose that we knew all the allele-intelligence associations, and suppose further that we could establish an individual's genotype as easily as we can take fingerprints. Scientifically, any interest will be in the allele associations and the mechanisms by which the genes exert their actions, not on the individual's membership in a particular social group. In terms of social action (including education), any intervention to deal with that individual would again depend upon his or her phenotype, not the statistics on genotypes in populations that the individual might belong to. So why do we need to evaluate, or even need, Jensen's default hypothesis?
32. Only one reason comes to mind. While Jensen is silent on this point, others have used his point of view to attack a variety of affirmative action programs. The attack is based on the argument that the affirmative action programs are only justified if we believe that cognitive skills in various groups are really equal, so that social intervention can realise an unrealised potential. Jensen's point of view is that the potential is largely not there in affected minority groups. If you accept this argument, and you believe that the affirmative action programs are based on an assumption of equality, then expenditures for these programs are not justified.
33. As I have indicated above, I have reservations about the strong statement that group differences in cognitive ability are immutable. On the other hand, there is no compelling reason to believe that no differences exist; and we simply do not know how much effort it would take to give everyone, everywhere, an equal opportunity to realise his or her potential.
34. It follows that we have to make judgement calls about how many resources to develop in programmes that might work. (A programme that does not work is simply a pork barrel project.) These judgement calls occur in the political realm, not the scientific realm. As of the end of the 20th century American and European societies have a few 'affirmative action' programmes for farmers, and offer tax breaks to retailers and shoppers on the Internet. The USA supports a special National Science Foundation programme for women in science and engineering, and provides taxpayer-financed incentives to the owners of professional football and baseball teams. It also provides funding to allow university professors to explore the Antarctic, offers school lunch programmes to poorer children, and does many other things that provide advantages for some citizens at the expense of others. In theory, at least, these programmes are justified on the grounds that they add to the overall social good. That is the criterion an affirmative action programme has to meet; will it add to the overall social good? Jensen's default hypothesis is irrelevant on social as well as scientific grounds.
Ackerman, P.L. (1988), Determinants of individual differences during skill acquisition: Cognitive abilities and information processing. Journal-of-Experimental-Psychology:- General. 117 (3), pp.288-318.
Carroll, J. B. (1993), Human Cognitive Abilities. Cambridge. Cambridge U. Press.
Cattell, R. B. (1971), Abilities: Their structure, growth, and action. Boston: Houghton Mifflin.
Cawley, J., Conneely, K., Heckman, J. & Vytlacil, E. (1997), Cognitive ability, wages, and meritocracy. In Devlin, B., Fienverg, S.E., Resnick, D.P., & Roeder, K., Intelligence, genes, and success. New York: Springer-Verlag, pp.178-192.
Flynn, J.R. (198), Massive gains in14 nations. What IQ tests really measure. Psychological Bulletin, 101, pp.171-191.
Herrnstein, R.J. and Murray, C. (1994), The Bell Curve: Intelligence and class structure in everyday life. New York: The Free Press.
Hertzog. C. (1994), Cohort effects. In R.J. Sternberg (ed.), Encyclopedia of Human Intelligence, Volume 1. New York: MacMillan, pp.273-275.
Horn, J.L. (1985), Remodeling old models of intelligence. In B.B. Wolman (ed.), Handbook of Intelligence. Theories, Measurements, and Applications. New York: Wiley, pp.267-300
Horn J.L. & Noll, J. (1994), A system for understanding cognitive capabilities: A theory and the evidence on which it is based. In D. Detterman (ed.), Current Topics in Human Intelligence: Vol. 4. Theories of Intelligence. Norwood, NJ. Ablex, pp.151-204.
Jensen, A. R. (1969), How much can we boost IQ and scholastic achievement? Harvard Educational Review, 39(1), pp.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
Mackintosh. N. (1998), IQ and Human Intelligence. Cambridge. Cambridge U. Press.
Teasdale, T.W. & Owen, D.R. (1989), Continuing secular increases in intelligence and a stable prevalence of high intelligence levels. Intelligence 13(3), pp.255-262.
Widaman, K.F. & Carlson, J.S. (1989), Procedural effects on performance on the Hick paradigm: Bias in reaction time and movement time parameters. Intelligence 13(1), pp.63-85.