General intelligence
Also known as: g factor, General cognitive ability
The common factor underlying performance across diverse mental tasks.
What it means
General intelligence, denoted g, is the broad latent factor that explains why performance on diverse cognitive tasks tends to be positively correlated — people who do well on one type of mental test tend to do well on others (the 'positive manifold'). Extracted statistically via factor analysis, g sits atop hierarchical models of cognitive ability above more specific factors. It is substantially heritable, fairly stable across the lifespan, and among the best single predictors of educational attainment, job performance, and health and longevity. Its interpretation is contested: whether g reflects a single underlying capacity (such as processing efficiency or working-memory capacity) or emerges from many interacting processes remains debated, and measured scores are influenced by environment, motivation, and test fairness. It matters because cognitive ability is a powerful, pervasive individual difference that moderates learning, decision quality, and susceptibility to error.
The original demonstration
The idea traces to Charles Spearman, who in 1904 noticed that children's marks across unrelated school subjects, and their scores on sensory discrimination tasks, were all positively correlated. To account for this pattern he proposed that every cognitive test taps two things: a factor specific to that test and a factor common to all of them, which he labelled g for general intelligence. Spearman's two-factor theory was a statistical inference, not a claim about a brain organ. What made it durable was the robustness of the underlying observation. When a diverse battery of mental tests is administered to a large sample, the correlations among them are almost always positive, a regularity later named the positive manifold. Factor analysis extracts a first, dominant dimension that typically explains a large share of the variance shared across tests. A century of subsequent work has confirmed the manifold repeatedly, while the interpretation of what g actually represents has remained the contested part of Spearman's legacy.
What the evidence shows
Modern factor models are hierarchical. John Carroll's 1993 survey re-analysed hundreds of historical datasets and arrived at a three-stratum structure with g at the apex, broad abilities such as fluid reasoning and processing speed below it, and narrow skills at the base. This layered picture, blended with earlier work into the Cattell-Horn-Carroll framework, underlies most current test batteries. Twin and family studies show that individual differences in g are substantially heritable, with estimates rising from roughly a fifth to a quarter in childhood to well over half in adulthood; Plomin and Deary's review summarised this and related findings, while also noting that molecular studies attribute the heritability to very many genes of tiny individual effect rather than a few large ones. Rank-order stability is high: people's relative standing changes only modestly across decades, one of the larger longitudinal correlations in psychology. Predictive validity is real but has been recalibrated. Schmidt and Hunter's much-cited estimate that general mental ability predicts job performance at around 0.5 was revised sharply downward by Sackett and colleagues in 2022, who showed that the standard corrections for range restriction had been systematically too large.
Why it happens
A statistical factor is not automatically a thing in the world, and the mechanism behind the positive manifold is the crux of the interpretation debate. Spearman read his common factor as a single general capacity, a unitary resource every demanding task draws on; candidates offered since include neural processing efficiency, the speed and fidelity of information transmission, and working-memory capacity. Godfrey Thomson objected as early as 1916. His sampling, or bonds, theory showed that a positive manifold, and even a clean hierarchy, can arise when each test recruits an overlapping subset of many independent elementary processes, so tests correlate because they share components rather than because any one capacity underlies them all. Thomson's point was mathematical: the very correlations Spearman took as evidence for g could be generated with no g at all. Van der Maas and colleagues revived this line in 2006 as mutualism, a developmental account in which abilities begin largely uncorrelated but reinforce one another over time, so strength in one domain fosters growth in others and the manifold emerges without any common cause. On this reading g is a statistical summary, the first principal component of a battery, rather than a discovered entity, and treating that summary as a biological quantity is the reification error pressed by Stephen Jay Gould and others. The critique is fair as far as it goes, but it cuts less than it appears to: g's practical value, its stable prediction of schooling, job performance, and health, holds whatever its ultimate nature, because a summary can be predictive without being a single thing. What the correlational data cannot do is adjudicate between these accounts, since single-factor and non-g models are often indistinguishable when fitted to one cross-section.
Where it shows up
Because g captures variance shared across tasks, it surfaces wherever sustained reasoning matters. In education, performance across subjects hangs together enough that a general factor predicts attainment and years of schooling. In work, measures of general mental ability retain incremental predictive value across a wide range of jobs, though, as noted, the size of that value is now estimated more modestly than the classic literature claimed, and it is one predictor among several rather than a uniquely dominant one. The most striking extension is cognitive epidemiology, the study of test scores as predictors of physical health. Deary, Weiss and Batty documented associations in which higher measured intelligence in youth predicts lower later illness and mortality, even after adjusting for social background. The associations are consistent but observational, and the causal pathways, whether through education, occupation, health behaviours, safer environments, or shared bodily-system integrity, are only partly understood and cannot be read off the correlations themselves.
Limits and caveats
Several cautions temper any strong reading of g. Measured performance is not fixed: average raw scores on many tests rose markedly across the twentieth century, the Flynn effect, at a pace far too fast to reflect genetic change and concentrated in some abilities more than others. The gains show that whatever the tests capture responds to environment, schooling, and test familiarity, but they do not by themselves overturn the heritability estimates, because heritability describes the sources of differences within a population at a given time while the Flynn effect describes a shift in the population mean over time. The two are not in contradiction, and reconciling them cleanly remains an open problem. Test content itself can be a source of bias: items drawn from a particular language, schooling tradition, or body of cultural knowledge can disadvantage people outside that tradition, and scores are further moved by motivation, coaching, and familiarity with the format, so a gap between groups on a given instrument does not license a claim about underlying capacity. Heritability estimates, likewise, describe variance within a studied population under its particular conditions and say nothing directly about the causes of differences between groups or about immutability. And predictive validity, while genuine, explains a minority of the variance in real outcomes; the honest summary is that g is a powerful and unusually general individual difference whose interpretation, magnitude, and portability across contexts all still warrant care.
Examples
A candidate who scores high on verbal reasoning also tends to score high on spatial and numerical tests — the shared variance is g.
A staffing team adopts a short reasoning test to screen applicants for roles of varying complexity. It does add predictive signal above unstructured interviews, but after the 2022 recalibration of range-restriction corrections the team plans around a modest correlation with later performance, not the near-0.5 figure quoted in older manuals.
A secondary school notices that pupils who top the class in mathematics also tend to lead in history and languages, and that the weakest performers cluster across subjects too. The pattern is the positive manifold in miniature: a shared factor pulling otherwise distinct subjects into alignment.
In a long-running cohort, people who scored higher on a group cognitive test at age eleven show lower rates of later cardiovascular death, even among those from similar homes. The link is robust but observational, and researchers cannot say from it alone whether ability, the schooling and jobs it opens, or healthier behaviour is doing the work.
A test publisher re-standardises a widely used battery and finds that the old norms now flatter test-takers: a person of average ability today outscores the average from a generation earlier. Rather than any sudden rise in a fixed capacity, the drift is the Flynn effect, and the outdated cut-scores are quietly retired.
First described in Charles Spearman (1904).
Key references
- Spearman, C. (1904). "General intelligence," objectively determined and measured. The American Journal of Psychology, 15(2), 201-292. doi.org/10.2307/1412107
- Deary, I. J. (2012). Intelligence. Annual Review of Psychology, 63, 453-482. doi.org/10.1146/annurev-psych-120710-100353
- Thomson, G. H. (1916). A hierarchy without a general factor. British Journal of Psychology, 8(3), 271-281. doi.org/10.1111/j.2044-8295.1916.tb00133.x
- Van der Maas, H. L. J., Dolan, C. V., Grasman, R. P. P. P., Wicherts, J. M., Huizenga, H. M., & Raijmakers, M. E. J. (2006). A dynamical model of general intelligence: The positive manifold of intelligence by mutualism. Psychological Review, 113(4), 842-861. doi.org/10.1037/0033-295X.113.4.842
- Plomin, R., & Deary, I. J. (2015). Genetics and intelligence differences: Five special findings. Molecular Psychiatry, 20(1), 98-108. doi.org/10.1038/mp.2014.105
- Deary, I. J., Weiss, A., & Batty, G. D. (2010). Intelligence and personality as predictors of illness and death: How researchers in differential psychology and chronic disease epidemiology are collaborating to understand and address health inequalities. Psychological Science in the Public Interest, 11(2), 53-79. doi.org/10.1177/1529100610387081
- Schmidt, F. L., & Hunter, J. E. (1998). The validity and utility of selection methods in personnel psychology: Practical and theoretical implications of 85 years of research findings. Psychological Bulletin, 124(2), 262-274. doi.org/10.1037/0033-2909.124.2.262
- Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology, 107(11), 2040-2068. doi.org/10.1037/apl0000994