Editorial material-photography illustration for Theories of Intelligence: Which Are Best Supported?.

Theories of Intelligence: Which Are Best Supported by Evidence?

After more than a century of research, intelligence psychology has no shortage of theories. That can create the impression of a field unable to decide what intelligence is.

After more than a century of research, intelligence psychology has no shortage of theories. That can create the impression of a field unable to decide what intelligence is. The reality is more structured. Many of the theories are not competing over exactly the same claim. Spearman, Thurstone, Carroll and CHC are deeply concerned with the structure of individual differences. Process Overlap Theory asks what cognitive processes could generate the general pattern. Mutualism and network theories emphasize development and interaction. Gardner and Sternberg challenge, in different ways, the boundaries of what psychologists should count as intelligence. Once those levels are separated, the evidence is much easier to read. There is strong agreement about several empirical facts: diverse cognitive tests tend to correlate positively; cognitive abilities are also meaningfully differentiated; and hierarchical psychometric models can represent both general and more specific structure. The deeper disagreement concerns what causal processes generate that structure—and how far the concept of intelligence should extend beyond it.

What is established most securely?

1. Cognitive performances show a positive manifold

This is not a theory-specific result. Spearman, Carroll, Process Overlap Theory, mutualism, and network accounts all begin from the fact that diverse cognitive measures tend to correlate positively.

2. Cognitive ability is differentiated

Reasoning, acquired knowledge, visual-spatial ability, memory-related abilities, speed, and other domains are not interchangeable.

3. General and specific structure coexist

A broad general dimension can summarize covariance across many measures while broad and narrow factors represent additional regularities. That is why the one-versus-many debate is misleading.

4. Psychometric structure is not automatically causal structure

A factor model can organize correlations without identifying the unique mechanism that generated them. This is where theoretical disagreement begins.

The theories are answering different questions

FrameworkPrimary scientific jobCore proposalWhat it explains wellMain unresolved issue
SpearmanStructure and interpretationGeneral and specific components contribute to cognitive performancePositive manifold and general covarianceWhether g corresponds to a single causal psychological capacity
ThurstonePsychometric structureSeveral primary mental abilities organize performanceBroad-ability differentiationPrimary abilities are correlated and do not eliminate higher-order covariance
Cattell–HornStructure and developmentBroad abilities such as fluid reasoning and acquired knowledge are differentiatedNovel reasoning, acquired knowledge and differentiated abilitiesExact developmental mechanisms and higher-order structure
CarrollPsychometric structureNarrow, broad and general abilities form a hierarchyLarge factor-analytic literature and practical inventory of abilitiesExact causal meaning and incremental value of lower levels
CHCTaxonomy and structural synthesisOrganizes broad and narrow cognitive abilities using related Cattell–Horn and Carroll traditionsCommon measurement vocabularyNot a complete mechanistic theory; constituent traditions disagree on some points
Process OverlapProcess theoryOverlapping domain-general processes can generate positive covarianceAlternative causal interpretation of gDiscriminating evidence versus rival causal models
MutualismDevelopmental theoryAbilities become correlated through reciprocal developmental interactionsEmergence of the positive manifold without one common causeDirect longitudinal tests of reciprocal causal dynamics
Network modelsSystems/developmental theoryCognition develops as interacting networks of skills and knowledgePositive manifold and developmental organizationEmpirical identification and comparison with latent-variable accounts
GardnerBroad competence theoryProposes several relatively autonomous intelligencesEmphasizes diversity of human competenceWeak support for the proposed intelligences as psychometrically independent
SternbergContextual/process-oriented theorySuccessful functioning involves analytical, creative and practical abilitiesEmphasizes intelligent behavior beyond conventional testingPsychometric distinctiveness and generality of the proposed abilities

The table simplifies each framework. It is a map, not a verdict by itself.

Why not simply say “g theory won”?

Because g itself can mean different things. If the claim is:

a general factor can be extracted from diverse cognitive measures,

the evidence is extremely strong. If the claim is:

that factor proves one unitary psychological mechanism causes individual differences across all cognitive tasks,

the evidence is not equivalent. A psychometric general factor is not automatically a complete psychological theory. This is why theories such as Process Overlap and mutualism can accept the statistical phenomenon while disputing its causal interpretation.

Why not say “multiple abilities won”?

For the mirror-image reason. Evidence for broad and narrow abilities does not make general covariance disappear. A theory that recognizes only domain-specific differences must still explain why those domains correlate. The data resist both extremes:

not one undifferentiated faculty; not many unrelated faculties.

What would a genuinely better theory have to do?

A mature causal theory of intelligence would need to explain more than a correlation matrix. It should connect:

  • cognitive processes;
  • development;
  • learning;
  • brain systems;
  • genetic and environmental variation;
  • knowledge acquisition;
  • task demands;
  • psychometric structure;
  • lifespan change;
  • external prediction.

And it should make discriminating predictions. If two theories both reproduce the positive manifold after the fact, that is not enough. We need observations that one theory predicts more successfully than the other.

Could the final answer be plural?

Probably. The expectation that one mechanism must explain every level of intelligence may itself be too restrictive. Biological differences can affect processing. Executive processes can overlap across tasks. Abilities can influence one another developmentally. Knowledge networks can amplify earlier differences. Schooling and culture can shape what becomes crystallized. Measurement models can summarize the resulting structure. A future theory may therefore integrate mechanisms operating at different levels rather than identify one winner that makes all others unnecessary. That is a possibility, not an established conclusion.

Where the evidence leaves the field

The strongest current synthesis is:

Human cognitive abilities are differentiated but positively correlated. A hierarchical psychometric structure captures that pattern well. General intelligence is therefore a robust statistical dimension of individual differences, while broad and narrower abilities also matter. What remains unsettled is the causal architecture that makes those abilities covary and develop as they do.

This allows us to be confident where the evidence is strong without pretending that a century-old theoretical problem has been completely solved.

References

  • Spearman, C. (1904). “General intelligence,” objectively determined and measured. American Journal of Psychology, 15, 201–293. DOI
  • Carroll, J. B. (1993). Human Cognitive Abilities. Cambridge University Press. DOI
  • McGrew, K. S. (2009). CHC theory and the human cognitive abilities project. Intelligence, 37, 1–10. DOI
  • van der Maas, H. L. J., et al. (2006). A dynamical model of general intelligence. Psychological Review, 113, 842–861. DOI
  • Kovacs, K., & Conway, A. R. A. (2016). Process Overlap Theory. Psychological Inquiry, 27, 151–177. DOI
  • Savi, A. O., et al. (2019). The wiring of intelligence. Perspectives on Psychological Science, 14, 1034–1061. DOI
  • Wasserman, J. D. (2019). Deconstructing CHC. Applied Measurement in Education, 32, 249–268. DOI
  • McGrew, K. S. (2023). Carroll’s Three-Stratum cognitive ability theory at 30 years. Journal of Intelligence, 11, 32. DOI