Arthur R. Jensen (1999) The Galton-spearman Paradigm as a Progressive Research Program. Psycoloquy: 10(083) Intelligence g Factor (20)

Volume: 10 (next, prev) Issue: 083 (next, prev) Article: 20 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 10(083): The Galton-spearman Paradigm as a Progressive Research Program

Reply to Buckhalt on Jensen on Intelligence-g-Factor

Arthur R. Jensen
Educational Psychology
School of Education
University of California
Berkeley, CA 94720-1670


All roads in the scientific study of human abilities lead back to Galton and Spearman. Buckhalt (1999) has clearly identified the main elements of the research programme that they originated, referring to these as the 'central dogma' of psychometrics. This long-running research programme has generated more established facts about human abilities and by far more practically useful applications of its methods than any other efforts in the study of human variation in mental traits. It has emerged theoretically, methodologically, and empirically as the most coherent, clearly articulated, and progressive research program in the development of a true science of mental abilities.


behavior genetics, cognitive modelling, evoked potentials, evolutionary psychology, factor analysis, g factor, heritability, individual differences, intelligence, IQ, neurometrics, psychometrics, psychophyiology, skills, Spearman, statistics
1. As a researcher and a professor in the fields of applied psychology and school psychology, Buckhalt (1999) perhaps has a better appreciation of psychometrics, individual differences, and particularly the dominance of g in the schooling process than do many academicians, whose experience of individual differences is largely confined to college students who are mostly in the restricted range of the top quartile of the IQ bell curve. We also know that abilities are more differentiated in this region of the distribution, that is, at higher levels of g there is also a greater development of group factors, talents, and specialised abilities that constitute the total variance in individual differences, so it is easier to perceive "multiple intelligences" among persons in the upper than in the lower quartile of the g distribution. (Appendix A in Jensen, 1998, reviews the evidence for this phenomenon, discovered by Spearman and more thoroughly studied in recent years.) Analogously, rich people spend their money on a greater variety of things than poor people do.

2. While Buckhalt expresses little doubt about the psychometric basis of g and its behavioural and social correlates, he believes that the evidence for its biological and evolutionary foundations is "less strong." This may be true, since this is the most recently investigated and the least consolidated or fully explicated aspect of the field. Yet I think two lines of evidence make it virtually certain that g is essentially biological, although the mechanisms through which its biological basis operate to produce behavioural differences among people are only beginning to be understood.

3. First, it is important to realise that "intelligence" and "g" do not stand for the same thing. They are very different concepts and confusing them in the least only leads to unnecessary arguments. One way to explicate the difference between intelligence and g is to realise that, in principle, everything that can be known about intelligence could be discovered with research on a single person, i.e., with N = 1. Intelligence refers to a class of the various behavioural capabilities common to all biologically normal members of a given animal species, although here we are particularly interested in Homo sapiens. These capabilities are behavioural, observable functions, such as apprehension of a stimulus, perception, discrimination, generalisation, conditioning, learning, memory, transfer of training, language acquisition, thinking, reasoning, and problem solving. These functions are possessed by all biologically normal human beings, that is, all persons without serious diagnosable brain damage or major genetic or chromosomal anomalies. The 'laws' of each of these functions (also called 'abilities') can be discovered by studying a single individual, just as Hermann Ebbinghaus discovered many 'laws' of learning and memory with himself as his sole experimental subject. Likewise, the brain mechanisms involved in these intelligence behaviours could, in principle, also be studied with N = 1. With N = 1 it would also be possible to study what Sir Charles Sherrington called the integrative function of the central nervous system, that is, those brain processes that allow communication between different functionally specialised modules to accomplish particular complex goal-directed actions.

4. But if we wish to study the conspicuous individual differences in any or all of these functions, we obviously need N 2. Then we will find that measurements of these various "mental" functions do indeed show reliable individual differences, and that these are also all positively correlated to varying degrees across various abilities. And then we discover that the matrix of all the intercorrelations among the N 2 measures of individuals' performances on these ability variables shows different degrees of generality; that is, we can discern clusters, some larger than others, among all the different variables' intercorrelations. Some variables are more strongly correlated with each other and some are weakly correlated. By means of factor analysis we can discover precisely how much of the variance in each measure is independently associated with each of the clusters and how much is not associated with any particular clusters (i.e., group factors or sources of variance that only certain abilities have in common), but is a source of variance common to all of the variables in the matrix -- in other words, the g factor, which emerges from a hierarchical factor analysis or as the first principal factor in a common factor analysis. There is also some residual variance that is unique to each specific ability measure. It should be noted that this g (and all other factors as well) depends upon the existence of individual differences in performance on the various mental abilities I listed above as "intelligence." Thus the psychology of mental abilities (or intelligence) and the psychology of individual differences in intelligence are entirely different concepts. A causal explanation of the one will not serve as the explanation for both.

5. The g factor represents the highest-order common factor among individual differences in a variety of behavioural tests that reflect all or at least a great many of these mental abilities. The g factor is not a direct measure of these diverse abilities per se, but of the individual differences variance they all have in common. An estimate of an individual's level of g, relative to other individuals who took the same battery of tests, is that individual's "factor score" on g, which is a g-weighted average of the individual's standardised scores on each of the tests in the battery, the weights being the tests' g loadings (i.e., the test's correlation with the one source of variance that is common to all of the tests in the battery). A g factor can be found in every battery of tests that one can concoct, provided the test items elicit one or more of the open-ended list of mental functions referred to above as intelligence, and are (1) sufficiently diverse in the psychometric sampling of these functions, and (2) sufficiently numerous to allow reliable measurement. The better these two psychometric conditions are met, the less error or variation there will be in the estimates of g. For several reasons, the most important of them being broad practical predictive validity, IQ is quite a good estimator of g factor scores, and this is true whether or not the construction of the IQ tests were based on factor analysis. But other factors, particularly verbal, numerical, and memory factors, also constitute some part of the total variance of most IQ tests.

6. Now back to the question of the biological basis for g, or more loosely speaking, individual differences in IQ. The well established fact that IQ (and g) has very substantial heritability, which increases from about .40 in early childhood to about .80 in late adulthood, averaging about .60 after the late teens, can only mean that a large part of the variance in IQ (specifically its broad heritability) is biologically based. This is certain, even though we don't know the chain of biological, biochemical, anatomical, and physiological processes by which the genotypic variance gets transformed into the phenotypic variance in IQ. But we don't yet have this knowledge either for other heritable traits, such as stature, blood pressure, longevity, or the probability of giving birth to dizygotic twins. However, a number of specific differences in the DNA between groups with average IQ and very high IQ have been identified (studies cited in Jensen, 1998). Further, myself and others have shown that the degree to which the various subtests that compose the IQ scores are g loaded predicts (with correlation coefficients in the .60 to .80 range) the size of the heritability coefficients of those subtests tests. This means that g per se is more highly related to the tests' heritability coefficients than are the other sources of variance in all the subtest scores. IQ therefore also has a slightly lesser heritability than g factor scores. The differing magnitudes of inbreeding depression (a wholly genetic effect) on various mental test scores are predicted by the tests' g loadings, with correlations around .80.

7. The evidence on IQ heritability and inbreeding depression necessarily implies an evolutionary origin of the biological basis of IQ or g. The mechanism of heritability consists of individual differences in gene (or allele) frequencies, and such differences have come about mainly through the chief mechanisms of evolution -- spontaneous genetic mutations and natural selection. Both genetics and evolution are absolutely fundamental and essential for understanding g. It might even be hypothesised that psychometric g is merely a lower-order factor in an even more over-arching biological super-G factor something like Darwinian fitness. Let's see what the evolutionary psychologists will make of this!

8. There is no "essence" of g, which is only the first factor in a common factor analysis or the highest-order common factor in a hierarchical analysis. Both methods yield highly similar (i.e., correlated) g factors, and highly similar g factors appear in most test batteries, provided the tests that compose them are numerous and highly diverse in the kinds of knowledge, skills, and mental operations called for. The obtained g factors vary only slightly across the different methods of extraction, and also with the breadth of psychometric sampling (i.e., number and diversity of tests included in the analysis). These variations in tests all give highly intercorrelated estimates of some "true" g, in the same psychometric sense that an obtained measurement is an estimate of the true measurement (which can't be known) but conceptually is the mean of an infinite number of measurements of the thing being measured. The g factor obeys the same logic, in this respect, as all other kinds of measurements, which are merely estimates of some "true" but necessarily unknowable value. At the present time, there are few ideal test batteries for measuring g and I have had to use the evidence that is available, which, it should be noted, is in some cases merely attenuated estimates of g, so that more accurate measurements would have yielded even stronger correlations with other, non-psychometric variables. With one exception, I have never used a g estimate based on any test battery that didn't include a variety of tests, as the first factor of a test battery test with rather homogeneous subtests would reflect some first-order or second-order group factor as much or more as it reflects g. A vocabulary test, for example, measures g and a verbal factor. The one exception I referred to is Raven's progressive matrices test (or other tests like it), which, in most factor analyses, typically loads only on g but not on any group factors.

9. Buckhalt is right that the frontier between psychometric g and its undoubted biological and neurological underpinnings is necessarily uncertain as to the specific mechanisms that mediate between physiology and behaviour. But it has been only recently in the history of psychology that the necessary tools have been available for research into this subject -- DNA analysis, electrophysiology, and imaging techniques like fMRI and PET. I have simply looked at the best evidence presently at hand; I haven't gone beyond it, except to express my hope that scientists will continue researching in this Galton-Spearman research paradigm that continually increases our knowledge and seems more promising than anything else on the scene in differential psychology. It represents, I believe, what the philosopher of science Imre Lakatos refers to as a "progressive" research program, in contrast to a "degenerating" program. The contrast between the empiricism and theoretical coherence of the Galton-Spearman paradigm and the alternative behaviouristic, humanistic, mentalistic, and egalitarian philosophies that claim to oppose it is brilliantly illustrated by Peter Urbach (1974) in terms of Lakotos's philosophy of science. I would urge everyone to read Urbach's article. The rapidly growing systematic body of knowledge based on empirical studies and reported in contemporary journals such as Behaviour Genetics, Intelligence, and Personality and Individual Differences clearly exemplifies the Galton- Spearman tradition. Other approaches have had nothing at all comparable or competitive to show.

10. Overall, I find little if anything to argue with in Buckhalt's well-written commentary, since for every critical point he makes, he is usually able to anticipate what my own response to it would be and then spells it out himself. Overall I find a high congruence between his and my own thinking on all these issues and feel glad to realise that someone else has so well understood the message of my book.


Buckhalt, J. A. (1999). Defending the science of mental ability and its central dogma. Review of Jensen on Intelligence-g-factor. PSYCOLOQUY 10(47). psyc.99.10.047.intelligence-g-factor.4.buckhalt

Jensen, A. R. (1998). The g factor: The science of mental ability. Westport, CT: Praeger.

Jensen, A. R. (1999). Precis of: "The g Factor: The Science of Mental Ability" PSYCOLOQUY 10(23). psyc.99.10.023.intelligence-g-factor.1.jensen

Urbach, P. (1974). Progress and degeneration in the 'IQ debate' British Journal of the Philosophy of Science, 25, 99-135 and 235-259.

Volume: 10 (next, prev) Issue: 083 (next, prev) Article: 20 (next prev first) Alternate versions: ASCII Summary