Mackintosh advocates experimental and componential analysis of the cognitive components involved in IQ tests items, rather rthan correlations between psychometric g and brain processes, as the more promising strategy for discovering the causes of individual differences in the abilities we refer to as intelligence. It can be argued that this componential approach is a blind alley because, at the behavioural level, the components themselves are correlated and yield a g factor much like that derived from a battery of psychometric tests. The fact that g has many nonpsychometric correlates, including various brain measurements -- which are not behavioural and are not abilities in any usual sense of the word -- suggests that a focus on the brain correlates of g will get us to a physical explanation of g more efficiently than research that is confined solely to behavioural variables and psychological constructs.
2. It is a fact that, among all psychometric factors identified so far, g is common to ("saturates") all cognitive tasks to varying degrees. And g has a wider range of nonpsychometric correlates than any other factor, or perhaps even any combination of factors, independent of g. If we statistically partial g out of the validity coefficients of various aptitude tests, remarkably little of practical predictive value is left. So it seems worth the trouble of trying to find out the causal source of this ubiquitous g factor, which can be measured in an almost unlimited number of ways. This source of variance, g, need not be attributed to some "unitary" process (whatever that might mean), but it is conceivably something simpler than the neural mechanisms or processes -- the hard wiring, cell assemblies, modules, and the like -- that make possible all the cognitive functions characteristic of Homo sapiens.
3. Elementary cognitive tasks (ECTs) are studied; these are typically so elementary and simple that individual differences in performance on them must be measured chronometrically, i.e., in terms of response time (RT) to the target stimulus or in terms of the duration of stimulus input needed for a correct but untimed) response (such as inspection time, or IT). These performances are usually measured in milliseconds. The most interesting ECTs in terms of their correlations with conventional psychometric measures of g are those that elicit an RT or have a stimulus duration of less than one second for adults or two seconds for elementary school children. These measures may serve as good analytical tools, because they can be experimentally manipulated in various ways that can reduce or increase their correlations with psychometric g. And this provides some handles by which we can gain some insight into the conditions that best elicit g (or other possibly correlated psychometric factors); and we can identify the particular features of ECTs that differ in their g loadings more precisely than these can be inferred from conventional psychometric tests.
4. I have devoted some 20 years to doing these kinds of studies, and many others have done so, too. (The journal INTELLIGENCE publishes much of the research in this vein.) Two things that have emerged from this research about which there is now little doubt or controversy are: (1) "mental speed," whatever its underpinnings, is positively related to g in all types of ECTs; and (2) it is entirely possible to measure the same g factor that emerges from conventional IQ and scholastic achievement tests by means of ECTs on which performance depends neither on (a) any previously acquired informational content or particular intellectual skills, nor on (b) individual differences in the knowledge content when tested under nonspeeded conditions. Yet chronometric measures from these brief ECTs (with individual differences ranging from about 250 to 1000 msec) are significantly correlated with scores on highly g-loaded, verbal and nonverbal tests of complex reasoning and general knowledge (e.g., Raven's Advanced Progressive Matrices, Terman's Concept Mastery Test, Wechsler Adult Intelligence Scale) administered to university students without time limits.
5. The problem with the ECT and cognitive components approach to understanding g is the very fact that questions about the source of g are simply pushed down to simpler tasks. This may help in describing the characteristics of g, but it does not disclose the causal basis of g. And the simpler the ECTs are made, the more they have one of the psychometric characteristics of single test items in conventional IQ tests, viz., more and more of the total variance becomes task specific relative to the g variance. Single test items in IQ tests typically correlate around .10 with each other and seldom more than .20 to .30 with the total score on the whole test. It can be shown that any test's specificity, after g and the significant group factors are partialled out, no longer correlates with real-life criteria, being in this respect the opposite of g, which has many such correlates.
6. Sternberg's (1977) well-received book on the componential analysis of intelligence laid out what, at the time, looked like a most promising program of research, highly similar to that which Mackintosh (1999) advocates in his final paragraph (para 11). But this program, despite the early enthusiasm and admirable ingenuity and effort expended on it, seems to have come to naught. The problem was essentially twofold: (1) the various identified cognitive components were themselves intercorrelated, leading Sternberg to conclude that Spearman's g was rediscovered through componential analysis, and (2) what remained of the components independently of g was just task specificity, which fails to transfer across different experimental tasks in which the particular components were hypothesised to operate. So Sternberg has since turned his theorising about human abilities in a quite different direction, far more molar and comprehensive than the cognitive components analysis of abilities.
7. The fact that g is more highly heritable than the smaller group factors, and certainly more so than any narrow components or task specificity, suggests that the locus of g is in some genetically influenced, hence physiological, processes, which are undoubtedly located in the brain. The main value of working with ECTs, as I see it, is their possible use in connection with MRI, PET, and other means of measuring actual brain processes. ECTs afford simpler, more sharply focused cognitive probes for stimulating the brain activities to be observed by these recently developed techniques. ECTs lend themselves to experimental manipulations that are impossible for conventional psychometric tests and might be more precisely reflected in measurable brain functions. We should not be satisfied with merely psychological descriptions of g; we must work toward a physical explanation as was urged by the discover of g (Spearman, 1927, p. 403).
8. Granted that in the literature referred to by Mackintosh there are almost as many variations in the ways that the average evoked potential (AEP) has been measured as there are experimenters working with AEP and IQ. Some of these methods show significant correlations with IQ while others don't. It reminds one of Edison's several hundred trials with electric light bulbs that didn't work because he had not yet found the material for a filament that could withstand incandescent temperatures for more than a few minutes. There are evoked potential procedures that replicate when followed carefully, and two of them show a remarkable relationship specifically to g, not just IQ (Jensen, 1998, pp. 152-157). This I consider something we should be excited about, rather than burying it beneath citations of numerous studies based on other procedures that failed to show significant correlations with psychometric tests. Let's have many more exact replications of these studies that have come close to the bull's eye. The easiest, technically, is probably Schaefer's procedure for measuring the habituation of the amplitude of the evoked potential over repeated trials (see Jensen, 1998, p. 155 and cited references). We have to start somewhere, and AEP, PET, and fMRI seem to be the best tools presently available.
9. Mackintosh states that performances on different ECTs correlate with different IQ factors. This is generally true when the group factors are not measured independently of g. There is some degree of stimulus content that is similar between some ECTs and some psychometric tests with a loading on some group factors. But it is the g loadings of these tests, whatever group factors they also measure, that is largely responsible for these tests' correlations with the ECT mental speed measures. For instance, when g is partialled out of the Wechsler IQ subtests (in which the remaining non-g common factor variance forms the group factors, such as verbal, spatial, and memory factors), the residualised subtest scores have near-zero correlations with the ECT measures. As for crystallised and fluid intelligence (Gc and Gf), each correlates independently with the mental speed measures derived from ECTs. Mackintosh mentions specifically the ECT known as the "sentence verification task" (SVT) as being correlated with Gc. True, but SVT is also just as substantially correlated with Raven's matrices, an exemplar among tests of Gf.
10. In a sample of university students, with only a quarter of the population variance in IQ, response times on the SAT correlated -.45 with the untimed Raven's Advanced Progressive Matrices and intraindividual differences (i.e., a subject's trial-to-trial variation in RT) correlated -.50. It is hard to imagine two more different tests than the Raven and the SVT, but they are both substantially g loaded and hence correlated with each other. (See Jensen, 1998, pp. 202-269 for references to this and a number of similar studies.) Inspection time (IT), is correlated with overall IQ about -.50, and among the independent factors in IQ, IT is correlated with g as well as with a spatial or perceptual speed factor. Although the g extracted from a number of diverse psychometric tests is highly correlated with the g extracted from diverse chronometric ECTs (Vernon, 1989), little if any of that correlation is attributable to features that are common to the features of the psychometric and chronometric tests.
11. The causal nature of the g factor may seem elusive only because psychologists have been looking for it in the wrong places. I believe, as did Spearman, that its nature will be discovered, not in the tests (whether psychometric or chronometric) but in certain processes in the brain that modify the speed and efficiency of the neural processes in the many of behavioural functions, common to all humans, that we view as instances of "intelligence." I hope most of us will be around to see Spearman's prophecy realised.
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(023). 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.J. (1999). Exploring component correlations rather than g. PSYCOLOQUY 10(075) ftp://ftp.princeton.edu/pub/harnad/Psycoloquy/1999.volume.10/ psyc.99.10.075.intelligence-g-factor.17.mackintosh http://www.cogsci.soton.ac.uk/cgi/psyc/newpsy?10.075
Spearman, C. (1927). The abilities of man. London: Macmillan.
Sternberg, R.J. (1977). Intelligence, information processing, and analogical reasoning: The componential analysis of human abilities. Hillsdale, NJ: Erlbaum.
Vernon, P. A. (1989). The generality of g. Personality and Individual Differences, 10, 803-804.