Human Cognitive Abilities : A Survey of Factor-Analytic Studies

Rating:
Summary: The importance of the g factor
Review: Claims for the importance of social class are muted today. Surveyed punters say they are all middle class; children's educational outcomes by their twenties are only slightly related to their fathers' occupations; and even the New Statesman and the Spectator have largely given up the class war. Yet might class differences merely be temporarily obscured - perhaps by mass higher education (or miseducation)? Might class differences always tend to be re-created by differences between people in intelligence? Might 'the classless society' prove a chimera?

Reassurance for egalitarians has long been on offer in the form of multifactorial accounts of intelligence. From 1935, American psychometrician-psychologists could be found who professed - contrary to the view obtaining in Charles Spearman's 'London School' - that there were many human mental abilities of equal importance and that these cognitive variations were largely uncorrelated with each other. From Louis Thurstone's 7 'primary abilities' and J. P. Guilford's 150 'functions' to Howard Gardner's 7 'multiple intelligences' and Robert Sternberg's 676 'components' (including a few hundred 'interaction effects'), the anti-elitist implication of multifactorialism was clear: given such statistical independence of abilities, no-one could be greatly superior to anyone else in all-round, general intelligence (g).

The 819-page scholarly verdict of Thurstone's one-time graduate, Emeritus Professor John Carroll (University of North Carolina), must therefore attract wide interest. Can the modern egalitarians of Harvard and Yale be allowed to bask in their multidimensional proclamations? Or is the London School - with the assistance of its far-flung outposts, especially at Berkeley - vindicated by results? (Oxford and Cambridge psychologists have deemed it impolitic to participate in a debate of such glaring educational relevance.)

Thankfully, few serious readers will miss the main thrust of Carroll's re-analyses of 461 well-found studies of individual differences in mental abilities. Although only a quarter of the studies involved normal population samples, and although most of them had some special focus (and so involved a less-than-catholic selection of mental tests), the positivity of inter-correlation between tests is sufficient to produce a clear g factor in 90% of them. Carroll, it turns out, has no quibble with the idea - introduced in an engaging historical chapter - that around 55% of the variance occurring on anything called a mental test will actually be common, g variance that is shared with most other mental tests. Time and again, tests named as 'language comprehension', 'reasoning', 'auditory discrimination', 'social intelligence', 'spelling' and 'creativity' turn out to correlated well with each other and thus with g. Carroll gladly repeats the handy dictum of the London School: 'g is to psychology what carbon is to chemistry.'

Regrettably, the methods involved in Carroll's factor-analytic opus are a painfully familiar mixture of the flat-footed and the mildly self-indulgent. Correctly, Carroll employs Schmid-Leiman procedures: these extract objective, simple-structure factors (usually themselves inter-correlated) before looking of any higher-order g factor; attention then turns to what mutually independent 'stub' factors remain at work once g's influence is statistically removed. The problem is with what happens next. Some studies will show and independent 'verbal' stub factor; others might add 'visualization', 'memory' and so on. Yet if one study fails to show a visualization factor, perhaps it simply did not involve enough tests of that type? ('Nothing in, nothing out' is a more important principle of factor analysis than is the better known 'Garbage in, garbage out.') Still worse, how is it to be shown that the factors appearing in different studies truly represent the same 'visualization' factor? For that matter, how does Carroll know that any tow studies really do yield the same g factor? Faced with these problems, Carroll honestly admits that his operation is essentially one of 'cataloguing' factors. None will doubt that Carroll has the common sense and scholarly knowledge that make a good cataloguer; but cataloguing by appearance still involves a certain subjectivity and cannot constitute a proof of anything unless the fit of one cataloguing system is somehow compared with the fit of another. Carroll would have done better to have estimated the median correlations between the more popular tests from the better studies, and then to have submitted the resulting notional (but far from fictional) matrix to factor analysis. Keen factorists will buy the data disk that the publishers offer and hope it will enable them to attempt this exercise.

Still more regrettable, intent as he is on cataloguing his factors as belonging under the same heading (or not), Carroll misses the chance to discuss most of the more interesting questions of psychometric psychology. What proportion of test variance is actually accounted for by g in each study? Which tests have the highest g-loadings? Which are the best tests of numerical facility, clerical speed etc? Is there real variance in 'creativity' once the influence of g is set aside? What of Mike Anderson's claim that 'differentiation' of independent, non-g abilities is more easily seen in testees of higher levels of g? Repeatedly, Carroll prefers to evade such questions - together with any possibility of interpreting g's nature in the light of its strong correlations with modern measures of speed of apprehension, such as 'inspection time.'

There has to be a place for this book in any library covering differential psychology. Carroll takes a firm stand against critics of 'reification', maintaining that general intelligence is indeed "a tangible and well-defined entity" - quite as much as electricity, heat or gravity; and his scholarship can probably be trusted - despite 'cognitive' being mis-spelled in the Table of Contents. What doesn't need taking on trust is this: only 110 of Carroll's 461 surveyed studies showed 2 or more distinct higher-order factors; and these factors usually involved a g factor plus one other that varied greatly between studies. That human intellectual differences arise principally because of g will be plain to any factor-analytically sophisticated reader; and Carroll's more discursive concluding chapters outline the modern case for recognizing the causal relevance of g to education, to employment and to the final social class positions and cultural levels that people attain.

Rating:
Summary: An extraordinary review of a century of factorial studies
Review: What would you think of an analysis of more than 400 datasets? John B. Carroll has given us an impressive review of the factorial analytic studies covered most of the human cognitive abilities. But, he has gone beyond the re-analysis level and has made up a new model of intelligence: "The three strata model" which will help us to avoid hopeless discussions about the nature of the most important cognitive abilities and to advance in the actually interested topics about such abilities. Specially, he finishes with the argument about the existence of the g factor but, he also recognizes the importance of the Thurstone's primary mental abilities and the Cattell and Horn's distinction between fluid and crystallized intelligence. I think everyone concerned in the cognitive abilities field should read this book and keep "The three strata model" in mind in the analysis of his/her research.