The g factor (general intelligence)

Charles Spearman published it in 1904 in the American Journal of Psychology, in a paper titled "General intelligence, objectively determined and measured". He studied schoolchildren's marks in subjects such as classics, French, English and maths and found that they were correlated. He proposed a two-factor theory: a general factor, g, shared by every task, and specific factors, s, unique to each.

The pattern of positive correlations among all cognitive tests is called the positive manifold. It is one of the most replicated results in psychology. Ian Deary's 2012 review in the Annual Review of Psychology treats it as the starting point for understanding intelligence differences.

Later work found that abilities cluster. John Carroll reanalysed more than 460 datasets for his 1993 book Human Cognitive Abilities and arranged the results in three strata: g at the top, about eight broad abilities such as fluid reasoning and crystallised knowledge in the middle, and more than seventy narrow skills at the base. The Cattell-Horn-Carroll model that most modern tests use builds on this map. See fluid vs crystallised intelligence for the two best-known broad abilities.

Two further results matter. First, g does not depend on one particular test. Johnson and colleagues gave three different test batteries to the same people in 2004. The g factors from the batteries correlated .99, .99 and 1.00. Second, g is not limited to western samples. Warne and Burningham analysed 97 datasets from 31 non-western countries in 2019 and found a single dominant factor in most of them, explaining about 46% of the variance on average, close to the figure in western samples.

Tests that measure g well predict many outcomes. The strongest link is with school results. Deary and colleagues followed more than 70,000 English children and found that a general ability factor at age 11 correlated .81 with a general factor of exam results at 16, after correcting for measurement error. For any one test and one exam the figure is lower.

For work and life the links are smaller. The numbers on income and education have their own page. Schmidt and Hunter put the correlation of general mental ability with job performance at .51 in 1998. Sackett and colleagues argued in 2022 that earlier corrections were too generous and estimated about .31. Health is another area. A 2011 meta-analysis of 16 studies by Calvin and colleagues found that a one standard deviation higher score in youth went with a 24% lower risk of death over follow-ups of 17 to 69 years.

Correlations describe groups. A prediction that works on average can still miss badly for one person, and the unexplained share is large: effort, health, opportunity and luck all play a part. The IQ test validity guide explains what these numbers mean.

The main debate is about what the correlations mean, not whether they exist. Critics raise several points.

• g may be a summary, not a cause. Stephen Jay Gould argued in The Mismeasure of Man (1981) that treating a statistical factor as a single thing in the head is a mistake, which he called reification.
• The positive manifold may arise without a single general ability. In 2006 van der Maas and colleagues showed in a simulation that if cognitive processes help each other develop, positive correlations appear with no g inside the person. This is called mutualism.
• Overlapping processes. Kovacs and Conway proposed in 2016 that different tests draw on overlapping sets of processes, such as executive control. The more tests overlap, the more they correlate, so the correlations need no single general ability.
• It leaves out much. g says little about creativity, practical judgement, social skills or motivation. Broader theories, such as Gardner's multiple intelligences and Sternberg's triarchic theory, aim to cover them. They are still argued over.

Tests differ in how closely they track g. Tests of reasoning on novel problems, such as matrix puzzles, correlate strongly with it, and in many studies fluid reasoning and g are hard to tell apart statistically. Tests of narrow memory or speed correlate with it less. That is one reason that full batteries such as the WAIS use many subtests: each is an imperfect measure of g, and combining them gives a more stable estimate.

A short test built on a single item type, such as the 25 matrix-style questions in the LifeScore test, is a narrower sample. It can show how you handle that kind of reasoning. It will track overall ability less reliably than a full battery, which is one more reason to read its result as an estimate.

g also changes shape across life. Different abilities peak at different ages: Hartshorne and Germine found that processing speed peaks around the end of school, abstract reasoning plateaus in early adulthood, and vocabulary keeps growing into the 40s and beyond. Age-normed scores hide this by comparing you with your own age group.

g is a well-replicated statistical regularity and a useful summary. Whether it reflects one general ability or many interacting ones is still debated. Either way, a single IQ number hides the profile of strengths behind it, and two people with the same score can differ a lot on the parts.

For the scale that expresses g-based scores, see the standard deviation guide. For how much of g-related variation is genetic, see heritability.