What Is the g Factor in Intelligence?

The positive manifold

Give a few hundred people twenty different mental tests and compute all the correlations. Every one will be positive. Someone who scores above average on vocabulary is more likely than not to score above average on mental rotation, digit span and number series, even though those tasks look nothing alike. Spearman called this the positive manifold, and showed that a single common factor could explain most of it, with each test also having something specific to itself. He named the common factor g.

A century of replication in every population tested has made the positive manifold the most reliable finding in differential psychology. It is why a “general intelligence” score is meaningful at all: if the tests did not correlate, adding them up would be like adding shoe size to blood pressure.

How much g explains

In a well-designed battery such as the WAIS, g accounts for roughly half of the variance in subtest scores; the rest is broad abilities (verbal, spatial, memory, speed) and test-specific skill. Tests differ in their g loading — how strongly they correlate with the common factor. Vocabulary and matrix reasoning are among the highest (around 0.8); simple reaction time and clerical speed are lower (0.4–0.5). A test's g loading predicts how well it predicts real-world outcomes, which is the practical reason g matters: it is the part of IQ that carries the predictive validity.

g, broad abilities and the CHC model

Spearman's two-level picture (g plus specific factors) was refined by Cattell and Horn, who split off fluid and crystallized ability, and then by John Carroll's 1993 reanalysis of 460 datasets into the three-stratum model: g at the top, around ten broad abilities in the middle (fluid reasoning, crystallized knowledge, visual-spatial processing, working memory, processing speed, long-term retrieval and others), and dozens of narrow abilities below. Merged with Cattell and Horn's work, this is the Cattell-Horn-Carroll (CHC) model that modern test batteries are built on. The broad abilities are real and useful — they are what a cognitive profile reports — but they all correlate, and g is the name for what they share.

What g is, physically

This is where the agreement ends. g correlates moderately with brain volume (about 0.25–0.3), with the efficiency of white-matter connections, with working-memory capacity and with processing speed on elementary tasks, but no single mechanism has been found. The mutualism model proposes that g is not a thing at all but emerges because abilities help each other develop: better working memory helps you learn vocabulary, more vocabulary helps you reason, and so on, producing the positive manifold without a common cause. The sampling model proposes that every test draws on overlapping sets of many small processes. Each explains the data; none has won. For scoring purposes it does not matter — g as a statistical factor behaves the same under every account.

What g does not explain

About half of the variation in any specific ability, for a start: two people with the same g can have very different profiles, and the profile is often what matters for a job or a course. Creativity, motivation, conscientiousness and emotional regulation are largely independent of g, and in most studies conscientiousness predicts outcomes about half as strongly as intelligence does, additively. An IQ score tells you about g; the five-domain breakdown on our full test tells you about the broad abilities; neither tells you about the rest of the person.

Frequently asked questions