General Intelligence (g)

What it is

Spearman published General Intelligence, Objectively Determined and Measured in the American Journal of Psychology in 1904, and the attribution is confirmed against the primary paper. His two-factor theory says that every test score is a mixture of the general factor and a factor specific to that test, which is why diverse tests correlate but do not correlate perfectly.

Lewis Terman later amalgamated the general factor with William Stern's quotient at Stanford, and the combination is what most people mean when they say an IQ test.

The case for the construct is strong at the level of the observation. The positive manifold is among the most reliably replicated findings in psychology. Vocabulary, spatial rotation, working memory and reasoning tests correlate across samples and across decades, a general factor can be extracted from almost any battery, and the resulting score predicts schooling outcomes and some occupational performance better than most single measures the field has.

In effect

The case against is not about the data. It is that a factor which can be extracted is not thereby a thing. Stephen Jay Gould pressed this most forcefully in The Mismeasure of Man: factor analysis produces a mathematical summary, and treating that summary as a single underlying quantity of mental energy is reification.

The strongest modern version of the objection is not rhetorical but formal. Han van der Maas and colleagues proposed a mutualism model in which the positive manifold emerges from mutually beneficial interactions between cognitive processes during development rather than from any single common cause, and a body of network-psychometric work has pursued that alternative since. This matters because mutualism fits the same correlations without a unitary general ability behind them. A manifold that could be produced either by one underlying quantity or by many processes helping one another along is not evidence for the first reading over the second. Louis Thurstone's primary mental abilities and the later hierarchical models are the historical alternatives.

The other thing to keep in view is that g is a latent variable. It is inferred from a pattern of correlations and cannot be measured directly. Test scores are the working proxy and not the thing, and popular writing on intelligence almost never draws that distinction.

So anyone writing from this concept should say which of the three claims they mean, because the evidence for each is different.

What it does not say

It does not say that a general factor cannot be extracted. It can, from almost any battery, and that is not in dispute.

It does not say that g is measured. It is inferred, and the instruments that carry it produce scores that stand in for it.

It does not establish a single cause. The mutualism model accounts for the same correlations without one, which is why the causal reading is recorded here as unresolved rather than as settled in either direction.

It does not carry any implication about heredity. That is a separate argument with its own evidence, and the histories of intelligence testing this publication has read take no technical position on it. A history of how intelligence was defined, measured and used is never the authority for what intelligence is.

---