The g Factor Explained: Spearman's General Intelligence

The g factor, short for general intelligence, is the common factor behind one of psychology's most replicated findings: people who do well on one kind of mental test tend to do well on all the others. Charles Spearman identified it in 1904. Today g is estimated statistically from a battery of tests, and the full-scale score on an IQ test is essentially a practical estimate of it.

The positive manifold: why g exists

Give a large group of people a vocabulary test, a matrix-reasoning puzzle, a mental-arithmetic task, a block-design task, and a test of how many digits they can repeat backward. The tests look nothing alike, yet the scores on every pair of them correlate positively. People who are above average on one tend, on average, to be above average on the rest. Psychologists call this pattern the positive manifold.

The correlations are moderate rather than perfect, which is important. Plenty of people are strong verbally and average spatially, or the reverse. But the tendency is consistent enough across thousands of studies that it calls for an explanation, and g is the name for whatever the tests share.

What Spearman found

Spearman, an English psychologist, studied schoolchildren's ranks in subjects such as classics, French, English, and mathematics, along with simple sensory tasks like discriminating musical pitch. He found that the measures all correlated positively and that the pattern of correlations could be explained by a single common factor plus a specific factor unique to each task (Spearman, 1904). He later called the common one g and the specific ones s.

His two-factor model turned out to be too simple: some tests share more with each other than g alone accounts for. But the core observation, a general factor running through all cognitive tests, has held up for more than a century. For where this fits in the wider story of testing, see our history of IQ testing.

How g is measured

g is not measured by any one question. It is extracted by factor analysis, a statistical method that finds the smallest number of underlying dimensions that can account for the correlations among many tests. Each test gets a g loading: how strongly it correlates with the general factor.

• Higher loadings tend to go to tasks that require complex reasoning, inferring rules, or drawing on broad knowledge, such as matrix reasoning, verbal analogies, and vocabulary.
• Lower loadings tend to go to simple, highly specific tasks, such as quick symbol matching or basic perceptual speed.

In a typical battery, g accounts for around 40 percent or more of the variation in people's scores (Deary et al., 2010). The rest is shared among narrower abilities or unique to each test.

A striking result is that g does not depend much on which tests you choose. When researchers extracted g separately from three different large batteries given to the same 436 adults, the three g factors correlated at .99 to 1.00 (Johnson et al., 2004). Different sets of tests were measuring the same general factor.

Hierarchical models and CHC theory

For much of the 20th century researchers argued over whether intelligence was one thing or many. Louis Thurstone proposed seven "primary mental abilities" in 1938 as an alternative to g, but his primary abilities turned out to correlate with each other, and a general factor reappeared above them. Raymond Cattell and John Horn split general ability into fluid intelligence (reasoning with new problems) and crystallized intelligence (knowledge and skills built through experience).

The debate was largely settled by John Carroll, who reanalyzed more than 460 datasets from the factor-analytic literature and found the same three-level structure again and again: dozens of narrow abilities at the bottom, about eight to ten broad abilities in the middle, and a single general factor at the top (Carroll, 1993). His model and the Cattell–Horn model were merged into the Cattell–Horn–Carroll (CHC) theory, now the most widely used framework for designing and interpreting intelligence tests (Schneider & McGrew, 2018). Some of its broad abilities:

CHC also includes auditory processing, reading and writing, and two long-term memory abilities (learning efficiency and retrieval fluency), among others. Researchers still disagree on some details, including whether g should sit at the top of the hierarchy as a real ability or be treated as a statistical summary, but the hierarchical picture itself is broadly accepted. Our own test samples six of these areas and weights each by its typical loading on general ability, with fluid reasoning weighted most and processing speed least; the methodology page gives the details.

Is IQ the same as intelligence?

Not quite, though they are closely related. Three ideas get blurred together:

• Intelligence is the broad capacity to reason, solve problems, learn, and understand complex ideas. In everyday speech it can stretch to include wisdom, creativity, and social skill.
• g is the general factor that runs through all cognitive tests. It is the best-supported scientific core of the concept.
• IQ is a score from a specific test. A full-scale IQ from a good, broad battery is a strong estimate of g, but it also contains some test-specific variance and measurement error.

So an IQ score is a measurement of an important part of intelligence, the part that cognitive tests capture well. It is not the whole of what people mean by the word. Researchers who emphasize other abilities, such as Howard Gardner and Robert Sternberg, have proposed broader frameworks; our articles on types of intelligence and multiple intelligences theory cover them and the evidence for each. For a plain-language overview of the score itself, see what IQ is and what it measures.

Where g comes from in the brain

Neuroscience has not found a single "g center." Brain-imaging studies link general intelligence to a distributed network, with frontal and parietal regions especially prominent, and to properties of the whole brain such as overall size and the integrity of the white-matter connections between regions (Deary et al., 2010). Each of these associations is modest on its own.

Cognitive psychologists have looked for simpler building blocks. Working memory capacity and processing speed both correlate with g, and some researchers argue that one of them is the foundation the rest is built on. Neither explanation has won out, and the honest summary is that g is well described statistically but not yet well explained biologically.

What g does imply

• It predicts real outcomes. Most of the predictive power of IQ tests for school achievement and job performance comes from g rather than from narrow abilities.
• It is reasonably stable. People's rank relative to peers tends to be similar across adulthood, though scores can and do shift.
• It is partly heritable. Within the populations studied, genetic differences account for a substantial share of variation in g, and that share rises from childhood to adulthood (Plomin & Deary, 2015).

What g does not imply

• It is not proven to be a single thing in the brain. g is a statistical pattern. One account holds that it reflects a real general capacity. Others explain the positive manifold without one: the mutualism model proposes that separate abilities reinforce each other during development (van der Maas et al., 2006), and process overlap theory proposes that every test draws on several overlapping executive processes (Kovacs & Conway, 2016). The data do not yet decide between these.
• It does not make narrow abilities irrelevant. For some outcomes, such as specific academic skills or particular jobs, spatial, verbal, or mathematical strengths add useful information beyond g.
• It is not fixed. Heritability describes variation within a population at a given time. It does not mean a trait cannot be changed by schooling or environment, and it says nothing about why groups differ. Our piece on nature vs. nurture in intelligence covers this in detail.
• It is not a measure of a person's worth or of qualities such as kindness, honesty, or judgment, which research treats as separate from g.