Arthur R. Jensen (2000) The Ubiquity of Mental Speed and the Centrality of Working Memory. Psycoloquy: 11(038) Intelligence g Factor (37)

Volume: 11 (next, prev) Issue: 038 (next, prev) Article: 37 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 11(038): The Ubiquity of Mental Speed and the Centrality of Working Memory

Reply to Conway et al. on Jensen on Intelligence-g-Factor

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


Speed of information-processing, or "mental speed" (MS) and working memory (WM) capacity are now the two most important and fundamental constructs in theorizing about the causal nature of psychometric g. It is doubtful that either construct alone explains all the phenomena associated with g, and even both in combination are proabably inadequate. The true substrate of g may be a common cause of individual differences in both MS and WM capacity.


behavior genetics, cognitive modelling, evoked potentials, evolutionary psychology, factor analysis, g factor, heritability, individual differences, intelligence, IQ, neurometrics, psychometrics, psychophyiology, skills, Spearman, statistics
1. Conway, Kane, & Engle (1999) [henceforth Conway et al.], have presented a most thoughtful and theoretically provocative argument for the primacy of working memory (WM) capacity as the causal factor in g. Some of the key references in support of their thesis postdate by two years or more the completion of my book, "The g Factor" (Jensen, 1998; 1999). Hence further light has since been shed on my own discussion (Chapter 8) of these issues, in which I explicated the nature of (1) the relationship between mental speed (MS), as measured by reaction time (RT) to an elementary cognitive task (ECT), and measures of WM capacity, and (2) the relationship of both MS and WM to psychometric g, in terms of a hypothesis originally formulated by Lehrl and Fischer (1988).

2. This hypothesis explains individual differences in WM capacity in terms of two, presumably more basic, constructs: the speed (S) of information processing (in bits per second) and the duration (D) of neural traces (in seconds), such that the capacity (C) of WM is expressed as: (C bits) = (S bits/sec) x (D sec). We do know, at the physiological level, that the brain is a time machine, that is, neural processes take place in real time, measurable in milliseconds, and neural activity persists for some time after the termination of its eliciting stimulus. This is probably the appeal of the Lehrl & Fischer formulation of WM capacity. In turn, it is the resultant WM capacity, C, that is most highly correlated with g. According to this theory, variance in C thus has two sources: speed of information-processing and duration of short-term neural activity in WM.

3. However, explaining g as caused by WM, though a valuable step in its own right, and one with which neither my book (Jensen 1998) nor I myself at the present time disagree, it only postpones the reductionist's next question, viz., what is the cause of individual differences in WM capacity? Lehrl & Fischer (1988) have hypothesized an answer, which may be either right or wrong, and it is their formulation to which Conway et al.'s critical scrutiny should be specifically addressed. They have cited recent experimental evidence that appears inconsistent with a mental speed interpretation of WM, and this gives pause to the line of theorizing I, following Lehrl & Fischer (1988), adopted in "The g Factor." The points made by Conway et al. are so very much at the cutting edge of this field that I think it would be premature to wholly agree or disagree with their theory. I can say that eventually WM itself will have to be explained in more fundamental terms, and that explanation will perhaps itself be the explanation of g. There is much to study in Conway et al.'s paper and their many references, and I am eagerly looking forward to studying their "in press" references. It would seem foolhardy for anyone to go on theorizing about the relationship of WM to g and measures of MS before thoroughly digesting all of Conway et al.'s work on this topic. It is experimentally ingenious, theoretically probing, and most worthy of careful attention.

4. Not that Conway et al. need to be told, but readers should keep it in mind that at present WM is not a thing, but a cognitive construct "a black box (as is g itself). This is alright as far as it goes, so long as it is defined such that experimental and differential psychologists agree on objective means of measuring individual differences in WM capacity. After all, between Mendel's discovery and Watson's and Crick's discovery the gene, too, was a purely hypothetical construct, a black box, and breeding experiments elucidated a great deal about the actions of genes for almost a hundred years before anyone knew what a gene consists of. The reductionist path is often long and arduous, and this may also be true in investigating the physical basis of g. Meanwhile we must hope that experimental cognitive psychology will be as successful as were Gregor Mendel's and R.A. Fisher's models of gene action and all the still-valid genetic research they spawned prior to the advent of molecular biology and the Watson-Crick breakthrough.

5. But today, measures of WM capacity are essentially just another psychometric variable, albeit with a high g loading and the virtue of experimental manipulations permitting highly refined specification of its properties and relationship to other psychometric variables, including measures of MS. It's an advance over what Spearman was able to do, of course, but I don't see it as different in principle from what Spearman did in trying to infer the nature of g, by comparing various tests that consistently differ in their g loadings and inferring the differences in their cognitive demands. This analysis of similarities and dissimilarities between high and low g-loaded tests led to Spearman's famous characterization of g as "the eduction of relations and correlates." Yet even variables that seem to make little or no eductive demands have some degree of g loading. Galton's and Spearman's discoveries that sensory discrimination have some g loading, and a battery of various tests of sensory discrimination in different modalities yield a general factor that is correlated around +.40 with psychometric g based on conventional tests (Acton & Schroeder, 1999; Li et al., 1998). It is hard to see how this is explainable in terms of either WM or MS.

6. And what about tasks such as visual and auditory inspection time, which has a remarkably high correlation with IQ, yet seems to make no demands on WM capacity? The big question we are trying to answer is: What is the essential difference between persons with a high level of g and persons with a low level of g? We can show that groups of persons who differ in g differ on this, that, or the other measure of cognitive ability, and we can describe these measuring instruments in remarkable detail, and try to infer their common features as clues to the hypothetical constructs called "cognitive processes" that they presumably call upon. Part of the problem is the incredible ubiquity of g across the varied spectrum of cognitive tasks, whether conventional tests or specially contrived experimental tasks. And measures of MS taken on almost any kind of cognitive task above the level of complexity of simple reaction time shows some degree of g loading. It is the ubiquity of the MS-g correlations in so many different types of cognitive performance that maintains interest in MS as a key variable in this whole puzzle. Can it be dismissed from g theorizing simply because measures of WM are more highly correlated with g than are measures of MS? Even that is doubtful; because when various measures of MS are aggregated, they correlate almost as highly with g as conventional psychometric tests of g correlate with each other (Vernon, 1989).

7. MS, of course, is also a hypothetical construct, used to account for measured individual differences in RT and IT. These measures have the virtue of being a ratio scale and the repeatability of the testing paradigms yielding RT and IT make it possible to obtain any desired level of precision or reliability of the measurements. A prime question about individual differences in speed measures, such as RT and IT, is whether they are a secondary effect of some other brain processes which are not themselves related to the speed of brain processes, such as nerve conduction velocity per se. The nonsignificant correlation found between the visual evoked potential and measures of RT in the Hick paradigm raises a puzzling question that only further research can answer (Reed & Jensen, 1993). Attention is another construct and can be defined so as to plausibly explain individual differences in RT or IT, and particularly the intraindividual variability of RT across test trials, which has a higher correlation with g than does median RT itself (Jensen, 1992). But then how does saying that attention correlates with RT and the intraindividual variability in RT, on the one hand, differ from saying that g shows these same correlations, on the other? High g persons evidently have better "attentional resources" and vice versa. Is this unidirectional causality or a common cause? We are still in the correlational loop. The only way out that I can imagine at present is by identifying the physical conditions that determine correlations between individual differences in cognitive tasks, however diverse, that is, the physical conditions that determine g.

8. Now for some quite minor points, just to preclude any misunderstandings about my present position on these matters. Conway et al. write that g "probably has some physiological and genetic basis." There's no "probably" about it; we just don't know the details. Conway et al. also refer to fluid general ability (Gf), which I consider to be the same as Spearman's g, that is Gf = g. Crystallized ability (Gc) is the stored memory of knowledge and skills acquired through the agency of Gf = g and its interaction with opportunities, interests, and personality traits such as Ackerman's Typical Intellectual Engagement (TIE) (Ackerman, 1996). In children and young adults from similar backgrounds (especially siblings reared together), tests of Gf and Gc are so highly correlated as to be hardly distinguishable as separate factors in a factor analysis.

9. Notice Figure 8.6 (p. 237) in "The g Factor," showing a highly systematic relationship between RT on a number of elementary cognitive tasks and the g factor of the Armed Services Vocational Aptitude Battery (ASVAB), even though the ASVAB would be characterized as almost entirely a measure of Gc, comprising subtests of arithmetic, reading comprehension, mechanical knowledge, and the like. The fact that skills that become automatized through extensive practice and over-learning are less g loaded than are tasks involving controlled processing (because novel elements are constantly being introduced into the task) is undoubtedly related to the fact that controlled processing makes demands on attention and WM capacity, whereas automatic processing makes minimal, if any, demands on WM. The basis for this marked difference in g-loading could be that strongly conditioned responses and over-learned material change their site in the brain from early learning to over-learning, and the neural networks that retain memory may be less instrumental in reflecting the conditions that cause g, whatever these might be.

10. I recall hearing a talk some ten years ago by the physiological psychologist Richard Thompson in which he explained how the cortical site of a conditioned eye blink response during its conditioning phase later shifts to a different site in the brain after it has become well-established. The speed of knowledge or skill acquisition is g loaded, whereas apparently the retention of knowledge or skill shows little if any g loading. Yet it appears that the efficiency of later retrieval is quite g loaded. The RT for retrieval of past over-learned verbal codes (the Posner paradigm) is correlated with g. And both g and speed of retrieving information from long-term memory show a gradual decline with increasing age beyond about age 30, with an accelerating decrement after age 60 or so. No one who lives very long entirely escapes this trend, although there are vast individual differences in the toll it takes.

11. Conway et al. (1999, par. #13) refer to working memory capacity as not a "capacity" per se, but rather the ability to control activation. I like this formulation of WM capacity, but I believe it describes an involuntary physiological state or condition of the nervous system and it is therefore inappropriate to call it an "ability," at least not in the sense in which I have defined "ability" in my book (Jensen, 1998, pp. 51-53): a form of conscious, voluntary behavior (in addition to other crucial criteria).

12. Individual differences in the degree of neural arousal or activation may well be a causal process in g. It can hardly be a conscious, voluntary, or self-willed variable, because people have no subjective experience of their own level of g. And even the amount of incidental learning of things that a person has no interest in and pays little attention to is related to g. The observed difference in g levels at the extremes of its normal distribution in the population (say, below the 10th and above the 90th percentiles, which are roughly equivalent to IQs 80 and 120) are behaviorally so pronounced, and their real-life advantages and disadvantages so profound, that it would be hard to attribute them to voluntary or self-willed aspects of behavior.

13. Finally, I congratulate Conway et al. for pursuing a line of research that cannot help but advance our understanding of g. Their admirably analytic research deserves close attention.


Acton, G.S., & Schroeder, D.H. (1999). Color discrimination as related to other aptitudes: An analysis of the Farnsworth-Munsell 100-hue test. Paper presented at the 9th Biennial Convention of the International Society for the Study of Individual Differences, Vancouver, B.C., Canada, July, 1999.

Conway, A.R.A., Kane, M.J., & Engle, M.J. (1999). Is Spearman's g determined by speed or working memory? PSYCOLOQUY 10(074) psyc.99.10.074.intelligence-g-factor.16.conway

Jensen, A.R. (1992). The importance of intraindividual variability in reaction time. Personality and Individual Differences, 13, 869-882.

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). psyc.99.10.023.intelligence-g-factor.1.jensen

Lehrl, S., & Fischer, B. (1988). The basic parameters of human information processing: Their role in the determination of intelligence. Personality and Individual Differences, 9, 883-896. Reed, T.E., & Jensen, A.R. (1993). Choice reaction time and visual pathway nerve conduction velocity both correlate with intelligence but appear not to correlate with each other: Implications for information processing. Intelligence, 17, 191-203.

Vernon, P.A. (1989). The generality of g. Personality and Individual Differences, 10, 803-804.

Li, S-C, Jordanova, M., & Lindenberger, U. (1998). From good senses to good sense: A link between tactile information processing and intelligence. Intelligence, 26, 99-122.

Volume: 11 (next, prev) Issue: 038 (next, prev) Article: 37 (next prev first) Alternate versions: ASCII Summary