N. J. Mackintosh (1999) Exploring Component Correlations Rather Than g. Psycoloquy: 10(075) Intelligence g Factor (17)

Volume: 10 (next, prev) Issue: 075 (next, prev) Article: 17 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 10(075): Exploring Component Correlations Rather Than g

Book Review of Jensen on Intelligence-g-Factor

N. J. Mackintosh
Department of Experimental Psychology
University of Cambridge, Downing Street
Cambridge CB2 3EB
United Kingdom
phone: 44-1223-333551



One of Jensen's central arguments is that g is a biological phenomenon, to be elucidated by the study of the brain and of performance on "elementary cognitive tasks," rather than by an experimental cognitive analysis of IQ test items. This argument is found unconvincing.


behavior genetics, cognitive modelling, evoked potentials, evolutionary psychology, factor analysis, g factor, heritability, individual differences, intelligence, IQ, neurometrics, psychometrics, psychophyiology, skills, Spearman, statistics
1. "The g Factor" (Jensen 1998, 1999) ranges over most of the topics that have been traditionally associated with IQ testing: factor analysis, the heritability of IQ, its social correlates, and group differences in IQ. Most readers will be able to predict much of what Arthur Jensen has to say on these topics. But his central argument is that it is g, the general factor extracted by factor analysis of any IQ test battery, that is primarily responsible for these attributes of IQ: it is g that is heritable, g that predicts educational and occupational success, a lack of g that predicts social failure, and g that differs between different ethnic groups - particularly blacks and whites. Although I believe that Jensen overstates his case, it is useful to have a forceful, well argued and well documented exposition of this strong position. Those who disagree will need to address the evidence Jensen brings to bear on these issues, and will need to develop an equally well documented argument to show that this evidence does not support his position.

2. It would be impossible, in a short review, to do justice to all Jensen's arguments in the detail they deserve. Instead, I shall consider a rather different aspect of his general thesis, best summarised in his words:

    "g cannot be described in terms of the knowledge content of
    cognitive test items, or in terms of skills, or even in terms of
    theoretical cognitive processes. It is not essentially a
    psychological or behavioral variable, but a biological one, a
    property of the brain."

3. Now, there is no question but that g requires explanation at another level, and that factor analysis alone cannot provide that explanation. We surely want to know why performance on one IQ test correlates with performance on all others. Although critics such as Gould (1981/1996) have denounced this endeavour as an exercise in the reification of g, that is a silly argument. We need a theoretical explanation of g, and undoubtedly one possible explanation is that performance on all IQ tests depends, to a greater or lesser extent, on a single underlying process. But what could that process be?

4. Jensen's answer, implied above, is that the sheer diversity of content of existing IQ tests rules out the possibility of appealing to a single psychological or cognitive process. But rather than accept the implication that g must therefore be dependent on a variety of different cognitive processes rather than one, Jensen wants to argue that there must be a single basis for g; and if it is not cognitive then it must be biological - something like the general efficiency of the brain as an information processing device.

5. In Chapter 6, Jensen reviews evidence of the correlation between IQ scores and various measures of the brain. Although starting with a careful and suitably cautious warning about the difficulty of interpreting such correlations, as he progresses so he becomes somewhat less cautious, and in due course a couple of experiments are said to argue strongly that psychometric g is closely related to the electrophysiological information - processing activity of the brain and that g is the main, or even the only, cognitive factor represented in the correlation between IQ and AEP (averaged evoked potentials).

6. This is a remarkable reading of the evidence. Since Hendrickson and Hendrickson (see Robinson 1997) first reported a correlation of .72 between Wechsler IQ scores and their "string length" measure of AEP in 219 schoolchildren, there have been 13 studies examining the relationship between IQ and various measures of the complexity of AEPs to simple, repeated stimuli. Five of these studies have found modest to moderately strong positive correlations - but even here there is little or no consistent replication: different IQ tests have yielded quite different results, and most of these studies have used different measures of AEPs. Seven of the 13, including the largest with an N of 236, have found correlations hovering round zero, and one reported a significant negative correlation.

7. Twenty-eight studies have looked for correlations between the amplitude of AEPs and IQ. Of these, 17 have reported small to modest positive correlations, and 11 small to modest negative correlations. The picture painted by Jensen is surely quite misleading.

8. In all these studies, participants have been instructed to sit quietly while brief, repetitive stimuli are presented. There has been no further task demand. The only moderately consistent evidence of a relationship between brain activity and IQ has come from studies that require some mental work of the participants - performing mental arithmetic, answering questions from an IQ test, making fine sensory discriminations. But here the correlations are uniformly negative, higher IQ being associated with smaller changes in activity. The plausible interpretation is that changes in activity of the brain reflect the amount of effort people need to expend in solving the problem set them. In other words, the direction of causation is not from brain activity to IQ, but from IQ to changes in brain activity.

9. It would surely be surprising if there were no correlation between IQ scores and some measures of the brain. But the discovery of such correlations will do little to elucidate the cause of individual differences in IQ scores (let alone the nature of g) unless informed by theorising rather more sophisticated than that which talks of the "efficiency" of the brain.

10. In Chapter 8, Jensen pursues his reductionist argument by reviewing evidence that IQ scores correlate with performance on various "elementary cognitive tasks" (ECTs), such as reaction time, inspection time, memory scanning and so on. There are indeed significant correlations to be found. Even the reaction time to press a button when a light comes on, or the correct button when one or other of two or more lights come on, correlates about -.20 to -.30 with IQ (i.e. shorter RTS are associated with higher IQ). That is hardly overwhelming - but it still calls for explanation. Moreover, as Jensen documents, these correlations can be increased by changing the nature of the RT task, for example, by making the discrimination between the alternatives more complex, by requiring responses to verbal material, or a decision about the truth or falsehood of a simple sentence, or by imposing a load on working memory, requiring participants to perform the RT task at the same time as another task. As Jensen points out, the multiple correlation between IQ and performance on a variety of these ECTs may be as high as .60, and this correlation is indeed with g.

11. To me at least, it seems unsurprising that what is common to a variety of ECTs, some making demands on verbal, others on visuo-spatial abilities, others on yet other skills, should correlate with what is common to a variety of different IQ tests, some making demands on verbal abilities etc. What Jensen fails to acknowledge is that performance on different ECTs correlates with different IQ factors; sentence verification correlates with crystallized ability, some working memory tests with fluid ability, simple mental rotation with spatial ability, inspection time with perceptual speed. And so on. It is these component correlations that are surely responsible for the overall correlation between average ECT performance and g. And I strongly suspect that our understanding of the causes of individual differences in overall IQ will be advanced more rapidly by further exploration of the causes of these component correlations than by further pursuit of the elusive nature of g (see Mackintosh 1998).


Gould, S. J. (1981/1996). The Mismeasure of Man. New York: Norton.

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

Jensen, A. (1998) The G Factor: The Science of Mental Ability. Praeger

Mackintosh, N.J. (1998). IQ and human intelligence. Oxford: Oxford University Press.

Robinson, D.L. (1997). A test of the Hendrickson postulate that reduced EEG response variance causes increased AEP contour length: Implications for the neural transmission errors theory of intelligence. Personality and Individual Differences, 22, 173-182.

Volume: 10 (next, prev) Issue: 075 (next, prev) Article: 17 (next prev first) Alternate versions: ASCII Summary