|
"When I use a word, it means exactly what I mean it to mean, no more and no less." -- Humpty Dumpty "Intelligence is what intelligence tests measure." -- Edward Boring I'm actually surprised by how few comments have been made regarding the actual nature of intelligence in this thread. yummyfajitas' parent comment is one of those few, and he equates intelligence as we understand it with g, the hypothesized general intelligence posited by psychometricians. y.f. makes the same two arguments that psychometricians make for the validity of g as a representation of human intelligence,[1] which are supposed to demonstrate g's "internal validity" and "external validity". The former refers to the positive correlations between various (supposed) intelligence tests (described in the jargon as the "positive manifold"), and the latter to the positive correlations between g and supposed success in life. Broadly, the internal validity argument proves the "general" in "general intelligence", whereas the external validity argument proves the "intelligence". The first wrinkle in these arguments is the claim that all human intelligence tests positively correlate with each other to give a universal intelligence factor. This is false; there exist demographics where g is _not_ the predominant factor explaining inter-demographic test score differences.[2] Here is another problem. Psychometricians tend to decide whether some test metric measures intelligence by how well it correlates with existing "intelligence tests" that're g-associated. y.f. does this in his comment: "It is relatively independent of cultural knowledge: i.e., french is nearly independent of g (1), while physics, plumbing and loading irregularly sized boxes into a truck are highly correlated to g". (I.e., French tests can't be real tests of intelligence, since they don't correlate with g, but physics, plumbing, and bin packing must be, because they do.) But this renders the internal validity argument circular: its two prongs now become: (1) g must exist because intelligence tests correlate positively, and (2) and intelligence tests are those tests that correlate positively with g. The circle closes. The external validity argument supposedly rescues the g concept from this trap, by demonstrating that g correlates vaguely with real-life behaviours. Unfortunately, the correlations are nowhere near perfect, that g correlates with success doesn't suffice to show causality, and pointing out the correlations does not eliminate the underlying circularity in g's definition. All in all, the evidence is inadequate to reject the null hypothesis that there is no causal link between g and life success. [1] Well, I say "human intelligence", but if I recall correctly, at least one psychometrician even tried to argue that g might well be a cross-species phenomenon! [2] Dolan, C.V. Roorda, W., and Wicherts, J. M. (2004). "Two failures of Spearman's hypothesis: The GATB in Holland and the JAT in South Africa", Intelligence, vol. 32, p. 155-173. |
Take as your statistic the result of many tests: physics, box loading, french, basketball, etc, and then do a PCA or similar test. One of the principal components will be g, provided you have enough data. This is what defines g.
Intelligence tests are simply tests designed to be more highly correlated with g. If we discarded intelligence tests, we could recreate them (or equivalent tests) based on statistical analysis of the other test data. They are certainly not arbitrary measures.
The external argument doesn't rescue g; g is quite safe. The external argument merely claims that 'g' and 'intelligence' are the same thing, or very close. g exists regardless of what you want to call it.
As for the paper you cite, I skimmed it. Unless I misunderstood it horribly, it merely claims that particular IQ tests don't effectively measure g across different groups. That doesn't mean g doesn't exist as a hidden variable, merely that a particular test is poorly correlated with it for some population.