The intelligence quotient is often presented as a measure of intelligence. This formulation seems simple, almost obvious, yet it rests on a difficulty situated upstream of the test itself. Measuring intelligence as a whole first presupposes a sufficient definition of the phenomenon to which that word refers. That definition remains debated. A 2024 publication in the journal Intelligence notes that several classical definitions function mainly as lists of capacities — reasoning, planning, solving problems, understanding complex ideas or learning quickly. The authors propose a formulation centred on the maximal capacity to attain a novel goal successfully through perceptual and cognitive processes. Their proposal does not create a new consensus. It shows above all that, after more than a century of psychometrics, the general definition of intelligence is still an object of research.
Psychology can perfectly well study a construct that is not directly observable. It does so by relating behaviours, performances and theoretical models. The difficulty appears when three different levels are conflated: the theoretical phenomenon called intelligence, the capacities chosen to approach it, and the performances recorded by an instrument. A test may measure certain manifestations precisely without thereby turning those manifestations into an exhaustive definition of the phenomenon. The quality of the measure and the extent of the concept are two related but distinct questions.
A more satisfactory scientific definition should meet at least two requirements. It should propose a principle general enough to explain why several different capacities belong to the same phenomenon, instead of merely gathering them into a list. It should also produce hypotheses liable to be confronted with observations. Without a common principle, the definition remains descriptive. Without the possibility of being tested, it becomes hard to distinguish from a philosophical proposal.
Other disciplines have attempted to formalise the problem from a more general principle. In the field of theoretical artificial intelligence, Shane Legg and Marcus Hutter proposed defining intelligence by an agent's capacity to attain goals across a wide variety of environments. This approach is not a psychometric theory of human intelligence and does not directly resolve the question of IQ. It does, however, illustrate a useful point: as soon as a definition seeks to become formal, it must make explicit its choices concerning goals, environments and criteria of success. What remained implicit in the vocabulary then becomes a visible part of the model.
What psychometrics actually establishes
More than a century of research has considerably refined the psychometric representation of cognitive abilities. Contemporary models go far beyond the image of a mere addition of exercises designed to produce a single figure. The Cattell–Horn–Carroll framework, often designated by the acronym CHC, organises numerous aptitudes across several levels. It distinguishes notably capacities relating to fluid reasoning, acquired knowledge, visual processing, memory and processing speed. This multidimensional architecture makes it possible to describe differences between individuals with far more finesse than the idea of a general level alone.
The general factor, called g, rests on a particularly robust observation. Performances obtained in different cognitive tasks tend to be positively correlated. A person who performs relatively well in some tests is statistically more likely to obtain good results in several others. This regularity, known as the positive manifold, makes it possible to extract a common variance and explains why a general score can contain important information.
The interpretation of this organisation remains more open than its statistical description. A classical reading regards g as the expression of a latent general capacity contributing to performance across different domains. The mutualist model proposed by Han van der Maas and colleagues offers another possibility. Several cognitive processes could reinforce one another over the course of development. An improvement in one capacity would facilitate certain learnings, which would then favour the development of other capacities. Positive correlations could emerge progressively without a single cognitive cause being necessary to produce them.
This proposal does not make g disappear as a statistical summary. It reminds us that a factor structure does not by itself determine the mechanism that produced it. Network approaches extend this idea by representing cognition as a set of interdependent processes capable of generating part of the observed correlations between tasks themselves. The general factor remains informative for describing the organisation of individual differences, while its causal interpretation belongs to another level of the reasoning.
The correlations observed between cognitive performance and certain cerebral characteristics do not settle this question either. All mental activity depends on neuronal functioning. It would be surprising if relatively stable differences in reasoning, memory or attention bore no relation to the brain. Identifying these correlates informs us about the mechanisms associated with the performances measured. It does not automatically fix the boundaries of intelligence, nor allow us to conclude that the dimensions selected by the tests represent its full extent.
From mobilisable potential to observed score
The relations between psychometric intelligence, working memory and attentional control make it possible to specify what the tests capture. The work of Engle, Kane and colleagues has shown important relations between working-memory capacity, control of attention and fluid intelligence. Working memory does not correspond merely to the temporary holding of information. It also involves the capacity to maintain a goal despite interference, to select relevant information and to resist certain competing responses.
These capacities overlap strongly. They do not, however, make working memory, executive functions, attention and general intelligence synonymous. Two people obtaining a comparable overall score may display different profiles when it comes to sustaining attention over long periods, managing several constraints, changing strategy or resisting interference. The overall score condenses this complex organisation. That synthesis facilitates comparison at the price of an inevitable loss of part of the initial information.
The distinction developed by Keith Stanovich between psychometric intelligence and rationality makes this limit particularly visible. A person may have substantial cognitive resources while producing certain biased reasonings or poor decisions. Resisting biases, seeking out contradictory information or adjusting a belief to new data mobilises processes that are not identical with the capacities assessed by classical tests. Stanovich used the term dysrationalia to designate this possible dissociation between psychometric intelligence and rationality.
This distinction also avoids indefinitely widening the word intelligence until it contains every desirable cognitive quality. Rationality, creativity, critical thinking and social competence can be studied as distinct dimensions that maintain certain relations with psychometric capacities. Their existence suffices to show that a high score does not by itself describe the quality of all the cognitive processing a person produces.
Creativity provides a further example. A meta-analysis published in 2021 finds a positive and modest relation between performance on intelligence tests and creative achievement, around r = 0.16. Another meta-analysis devoted to divergent thinking, assessed through tasks requiring the production of several ideas or solutions, obtains a mean uncorrected relation of about r = 0.25. Depending on the scoring methods, the instructions and the corrections applied, this relation may be higher. The contrast nevertheless remains instructive: creativity and psychometric intelligence maintain measurable relations without being identical.
The difference between maximal performance and typical functioning brings another important qualification. A standardised test generally places the person in a situation where they are trying to succeed at a clearly defined task. Everyday life works otherwise. Goals may be ambiguous, information incomplete, distractions numerous, and effort must sometimes be sustained for weeks or months. Motivation, fatigue, stress, sleep or interest in the problem then influence how the available resources are actually mobilised.
Research on motivation during testing nevertheless invites us to keep realistic proportions. A 2022 publication covering six studies with 4,208 adults found a modest relation between declared effort and cognitive performance. In three experiments using financial incentives, the effects taken separately were not statistically significant and the combined estimate remained small, around 2.5 IQ points in the samples studied. The test situation can influence performance without turning the score into a mere reflection of motivation.
The result of a test thus corresponds to a capacity mobilised under defined conditions. It has a real stability while remaining produced by a person situated in a particular context. This intermediate position describes the instrument better than a reading that would turn the figure into a property isolated from every situation.
What predictive value allows us to assert
Predictive value is one of the strongest arguments in favour of cognitive tests. A meta-analysis covering 240 samples and 105,185 participants estimated at about 0.54 the corrected population correlation between measures of intelligence and school results. This relation varies with subjects, school levels, instruments and periods studied. Its magnitude remains large enough to show that the capacities assessed play a real role in academic success.
The relation with job performance also exists, with contemporary estimates more moderate than certain historical figures still frequently cited. A meta-analysis covering 153 samples and 40,740 twenty-first-century workers finds a mean observed correlation of 0.16 between general cognitive ability and overall job performance, and then a corrected estimate of about 0.22. The authors conclude that the relation is real while obtaining a magnitude appreciably lower than the classical estimates close to 0.50.
This predictivity makes it hard to present IQ as an arbitrary or contentless number. An instrument that maintains reproducible relations with learning and with several external performances manifestly captures important capacities. The reach of this conclusion remains limited to what the data establish. Predicting part of a school or occupational outcome is not the same as describing all the capacities that contribute to that outcome, still less as defining intelligence as a whole.
Incremental validity makes it possible to situate IQ among other information. A meta-analysis covering 267 samples and more than 413,000 participants shows that cognitive abilities and personality traits together explain more variance in school results than cognitive abilities alone. The latter remain the most important predictor in the analysis, while conscientiousness makes a substantial independent contribution. Success appears as the product of several dimensions, none of which suffices in isolation to describe the whole.
The same principle holds in a professional context. Cognitive abilities can contribute to rapid learning and problem-solving, while specific knowledge, experience, motivation, personality, work organisation or social interactions add other information depending on the post. The usefulness of IQ lies in the additional information it brings to a precise question, not in a claim to explain everything.
Ecological validity extends this reasoning. The correlations with school and work show that a standardised test retains a relation with certain behaviours outside the assessment situation. They do not allow that relation to be extrapolated automatically to every form of adaptation, innovation, understanding of a complex system or decision under uncertainty. Each extension requires data corresponding to the behaviour one actually wishes to predict.
The figures issuing from meta-analyses likewise deserve a methodological reading. A meta-analysis gathers studies whose populations, instruments, criteria and quality may vary. Publication bias, range restriction, statistical corrections or the definition of the outcome studied can modify the final estimate. The evolution of the figures concerning job performance illustrates this limit well. A quantitative synthesis represents the best available summary of a determinate set of data and method. It is not a natural constant.
Stability, development and transformation
Cognitive abilities display a substantial longitudinal stability, particularly from late adolescence onward. A meta-analysis published in 2024 gathers 205 longitudinal studies, 87,408 participants and 1,288 test–retest correlations. For a person aged 20 reassessed five years later, the mean stability of relative rank is estimated at around 0.76. This stability is far lower during the first years of childhood, rises rapidly and then remains high through much of adult life.
This value describes an average tendency and not the trajectory of each individual. High rank stability indicates that relative positions tend to be maintained in the population studied. It implies neither that each person keeps exactly the same position, nor that their score remains unchanged. Age, the interval between assessments, the capacities considered and the characteristics of the studies modify this stability. A high average can coexist with appreciably different individual trajectories.
Education clearly shows this coexistence of continuity and change. A meta-analysis based on 42 datasets and more than 600,000 participants estimates that an additional year of schooling produces on average a gain of about 1 to 5 points on various cognitive measures, depending on the methods used. Capacities influence the ease of learning, while education contributes in return to developing certain performances measured by the tests.
Ageing also shows why an overall score can mask internal transformations. Fluid capacities, which draw more on reasoning in novel situations and on the active processing of information, tend to decline earlier in adult life. Acquired knowledge and several crystallised components can remain stable much longer, or even improve during part of life. Average trajectories differ between capacities, to which substantial individual variation is added.
The Flynn effect makes sensitivity to context visible at the level of populations. A meta-analysis published in 2023, based on 1,038 samples, 299,155 participants and 72 years of observation with Raven's Progressive Matrices, finds a mean increase of about 2.2 points per decade. The magnitude varies strongly with countries, generations and periods. The phenomenon shows that standardised performances evolve with the conditions under which individuals grow up and learn. A score always compares a person to a reference population situated in a given era.
Individual stability, the effects of education, age-related transformations and historical trends describe a functioning that is relatively coherent without being fixed. This combination explains why tests can retain a good capacity to rank while recording real changes over the course of development.
Culture, norms and comparability
Language, schooling, perceptual habits, familiarity with certain forms of abstraction and many elements of context can influence the way a task is understood and carried out. Non-verbal tests reduce certain obvious linguistic influences. An abstract task nevertheless remains designed, administered and interpreted within a particular human environment.
Psychometrics has tools for examining part of this problem. Measurement invariance seeks to determine whether an instrument functions comparably enough across several groups to make certain comparisons interpretable. When it is established at the necessary level, it provides evidence for the idea that a difference of score reflects the construct studied rather than a different functioning of the instrument.
This statistical invariance does not by itself guarantee complete conceptual equivalence. The cross-cultural literature distinguishes notably procedural equivalence, which bears on the technical and structural properties of the measure, from interpretive equivalence, which concerns the meaning of the construct in different contexts. An instrument may display a comparable structure across several populations without the phenomenon assessed occupying exactly the same place or being conceptualised in a perfectly identical way in each of them.
Comparability gains from being formulated at the level actually demonstrated. A satisfactory invariance permits certain comparisons. It does not automatically establish the conceptual universality of the construct. Conversely, cultural influence does not make every comparison arbitrary. The degree of confidence depends on the population, the instrument, the use and the forms of equivalence actually verified.
From score to decision
The usefulness of IQ becomes particularly concrete when it enters into an educational or clinical decision. In education, a cognitive assessment can help in understanding certain difficulties, specifying a profile of strengths and weaknesses, or guiding further investigation. In a clinical context, it can contribute to describing a functioning, to a differential diagnosis or to monitoring certain cognitive impairments. Professional guidelines insist on the fit between the instrument and the question posed, on taking account of personal, linguistic and cultural factors, and then on using several sources of information when the consequences of the assessment are significant.
The same principle applies to communicating the result. A sentence such as « your intelligence is 130 » turns a statistical position into a personal quantity. A more precise formulation will describe the performances observed on the battery used, their position relative to the reference population, the uncertainty associated with the score, and the differences between indices where these carry relevant information. It is, for example, more accurate to write that « performance on the fluid-reasoning tasks lies well above that of the reference population » than to turn that observation directly into a general claim about the person's intelligence.
The overall score keeps its interest when the synthesis it produces matches the question posed and when the profile reasonably permits that reading. The indices become more informative when the gaps between domains are needed to understand the situation. Good practice consists less in choosing between using and abolishing IQ than in using the level of precision suited to the decision.
A psychological measure also changes status when it takes part in decisions that will subsequently modify the person's environment. A result may influence the resources offered, the educational path, the expectations, or the way an individual interprets their own capacities. The number then enters the system it was initially charged with observing. This possibility reinforces the importance of a descriptive and contextualised formulation.
Digital and automated uses make this question still more visible. In recruitment, cognitive tests, games or algorithmic systems can integrate a score directly into a pre-selection procedure. Automation facilitates the processing of the result without in itself increasing its validity. The easier a number becomes to apply at scale, the more important it becomes to verify its relevance to the intended purpose, its fairness and its limits.
The exact place of IQ
Criticism of IQ gains from starting out from what psychometrics actually establishes. Cognitive tests capture relatively stable differences between individuals. Their results display a robust statistical organisation, maintain important relations with working memory and attentional control, predict part of school or occupational success, and can supply useful information in educational or clinical contexts.
The difficulty appears when the success of the instrument begins to replace the definition of its object. Certain capacities are selected because they can be measured under standardised conditions. Their relations are studied, common factors are identified, and scores make it possible to compare individuals. Since these scores are reliable, predictive and easy to use, it becomes tempting to consider that the initial question has been resolved.
Research nevertheless continues to debate the mechanisms responsible for the correlations between capacities, the relations between psychometric intelligence and executive functions, the distinction from rationality, the links with creativity, the evolution of the various aptitudes across the lifespan, their dependence on cultural context, and the best way to define the phenomenon itself. This scientific activity does not weaken psychometrics. It specifies the limits of what its instruments allow us to assert.
A conception of intelligence can also go beyond the alternative between a single general faculty and an ever longer collection of independent capacities. Dynamic approaches offer a third possibility. Several distinct components may form an interdependent system whose organisation is built and transformed through development, learning and interactions with the environment. The unity observed in the data might then emerge in part from these relations. This orientation remains a family of hypotheses to be tested and not an established definition.
IQ then recovers a clearer place. It is a standardised synthesis of cognitive performances selected because they make it possible to study certain differences between individuals. That synthesis has real scientific and practical usefulness. It is neither a physical unit of intelligence nor an exhaustive description of a person's intellectual functioning.
We have learned to measure certain cognitive manifestations with great precision, to study their relations, their stability, their evolution and part of their consequences. In parallel we continue to debate the boundaries and the exact nature of the phenomenon we call intelligence.
A measure can be reliable, predictive and useful while remaining a partial representation. When that distinction fades, IQ progressively ceases to be one tool among others for studying certain cognitive performances and begins to determine what we agree to call intelligence.
Sources
1. Gignac, G. E., et al. (2024). Defining intelligence: Bridging the gap between human and artificial perspectives. Intelligence, 104, 101832.
2. Legg, S., & Hutter, M. (2007). Universal Intelligence: A Definition of Machine Intelligence. Minds and Machines, 17, 391–444.
3. 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.
4. Schneider, W. J., & McGrew, K. S. (2018). The Cattell-Horn-Carroll theory of cognitive abilities. In D. P. Flanagan & E. M. McDonough (Eds.), Contemporary Intellectual Assessment.
5. Kane, M. J., & Engle, R. W. (2003). Working-memory capacity and the control of attention: The contributions of goal neglect, response competition, and task set to Stroop interference. Journal of Experimental Psychology: General, 132(1), 47–70.
6. Stanovich, K. E. (1993). Dysrationalia: A new specific learning disability. Journal of Learning Disabilities, 26(8), 501–515.
7. Karwowski, M., Czerwonka, M., Wiśniewska, E., & Forthmann, B. (2021). How Is Intelligence Test Performance Associated with Creative Achievement? A Meta-Analysis. Journal of Intelligence, 9(2), 28.
8. Gerwig, A., Miroshnik, K., Forthmann, B., Benedek, M., Karwowski, M., & Holling, H. (2021). The Relationship between Intelligence and Divergent Thinking: A Meta-Analytic Update. Journal of Intelligence, 9(2), 23.
9. Bates, T. C., & Gignac, G. E. (2022). Effort impacts IQ test scores in a minor way: A multi-study investigation with healthy adult volunteers. Intelligence, 92, 101652.
10. Roth, B., Becker, N., Romeyke, S., Schäfer, S., Domnick, F., & Spinath, F. M. (2015). Intelligence and school grades: A meta-analysis. Intelligence, 53, 118–137.
11. Sackett, P. R., Demeke, S., Bazian, I. M., Griebie, A. M., Priest, R., & Kuncel, N. R. (2024). A Contemporary Look at the Relationship Between General Cognitive Ability and Job Performance. Journal of Applied Psychology, 109(5), 687–713.
12. Poropat, A. E. (2009). A meta-analysis of the five-factor model of personality and academic performance. Psychological Bulletin, 135(2), 322–338.
13. Ritchie, S. J., & Tucker-Drob, E. M. (2018). How Much Does Education Improve Intelligence? A Meta-Analysis. Psychological Science, 29(8), 1358–1369.
14. Breit, M., et al. (2024). The Stability of Cognitive Abilities: A Meta-Analytic Review of Longitudinal Studies. Psychological Bulletin.
15. Scharfen, J., et al. (2023). Seventy years, 1000 samples, and 300,000 SPM scores: A new meta-analysis of Flynn effect patterns. Intelligence, 98, 101750.
16. Wilson, C. J., Bowden, S. C., Byrne, L. K., Joshua, N. R., Marx, W., & Weiss, L. G. (2023). The cross-cultural generalizability of cognitive ability measures: A systematic literature review. Intelligence, 98, 101751.
17. Hedden, T., & Gabrieli, J. D. E. (2004). Insights into the ageing mind: A view from cognitive neuroscience. Nature Reviews Neuroscience, 5, 87–96.
18. American Psychological Association. Standards for Educational and Psychological Testing.
19. American Psychological Association. Guidelines for Psychological Assessment and Evaluation.
20. Equal Employment Opportunity Commission. Artificial Intelligence and Algorithmic Fairness Initiative.