Give a large group of people a sufficiently varied set of cognitive tests: reasoning, vocabulary, memory, spatial problems, processing speed. The tests are different, and people can show striking strengths and weaknesses across them. Yet a persistent pattern appears when the scores are compared across the group. People who score relatively well on one cognitive test tend, on average, to score relatively well on others. Psychologists call this pattern the positive manifold. It is one of the most consequential empirical findings in the study of intelligence because it creates the problem that almost every major theory must confront. Why should performance on tasks as different as vocabulary, spatial reasoning and memory be positively related at all? There is no single established answer. A general factor can summarize the shared statistical variation, but the existence of that factor does not by itself identify the mechanism that produced the correlations. The positive manifold is therefore the observation; g is one way of representing its common structure; theories of intelligence propose different accounts of why that structure exists.
A general factor can summarize the common statistical variation. But extracting that factor does not, by itself, tell us whether the correlations arise from one broad causal capacity, overlapping processes used by many tasks, reciprocal developmental interactions among abilities, networks of acquired knowledge and skills, or some combination of mechanisms. That distinction is the problem around which much of intelligence theory is organized.
What does “correlated” actually mean here?
Imagine 1,000 people take four fictional tests. Individual profiles can differ substantially even when the group as a whole shows a positive manifold. A simplified illustration might look like this:
| Person | Novel reasoning | Vocabulary | Spatial reasoning | Memory |
|---|---|---|---|---|
| A | High | High | High | Above average |
| B | Above average | Average | High | Above average |
| C | Average | Average | Average | Average |
| D | Low | Below average | Low | Low |
Real data are much messier, and people can show striking relative strengths and weaknesses. The positive manifold appears at the group level when many people are considered together. An illustrative correlation matrix might look like this:
| Reasoning | Vocabulary | Spatial | Memory | |
|---|---|---|---|---|
| Reasoning | 1.00 | .42 | .61 | .48 |
| Vocabulary | .42 | 1.00 | .34 | .38 |
| Spatial | .61 | .34 | 1.00 | .40 |
| Memory | .48 | .38 | .40 | 1.00 |
These numbers are fictional. A correlation of .61 does not mean that reasoning and spatial ability are “61 percent the same.” It means that, across people in this illustrative sample, higher standing on one measure tends to accompany higher standing on the other with a moderately strong linear association. The exact correlations depend on the tests, sample, reliability, age range and other design choices. The important pattern is that correlations among diverse cognitive measures tend to be positive rather than randomly positive, zero and negative.
Why is that scientifically surprising?
If every cognitive ability were completely independent, knowing how someone performs in one domain would provide no information about how they tend to perform in another. But it does. At the opposite extreme, if all cognitive tests measured one identical ability, the correlations would approach perfect relationships and meaningful domain differences would largely disappear. They do not. Human cognitive data therefore show two facts simultaneously:
- commonality — different abilities tend to move together across people;
- differentiation — they remain distinguishable.
Any serious theory of intelligence has to deal with both.
Does the positive manifold prove there is one intelligence?
No. It establishes a covariance pattern. A common-factor model can represent that pattern by proposing a latent dimension related to performance across many tests. That general dimension is commonly called g. But there is a crucial inferential step between:
many cognitive tests correlate
and
one psychological mechanism causes all those correlations.
The first is an empirical observation. The second is a causal hypothesis. Spearman moved toward the second interpretation. Later theories have offered different mechanisms while retaining the first observation.
Four ways the same pattern might arise
Consider four simplified explanations.
1. Common-cause account
Imagine a broad cognitive resource contributes to many tasks. If people differ in that resource, then the same underlying difference influences reasoning, memory, vocabulary learning, and spatial problem solving. The tasks correlate because they share a cause. This is the intuitive logic behind a reflective interpretation of general intelligence.
2. Process-overlap account
Now imagine there is no single process used by every task. Instead,
difficult cognitive tasks repeatedly recruit overlapping domain-general
processes—attention control, maintaining goals, managing
interference—alongside task-specific processes. Reasoning task A uses processes 1, 2, and 5.
Memory task B uses 2, 3, and 5.
Spatial task C uses 1, 4, and 5. No single process has to be “intelligence itself.” Correlations can
arise because tasks overlap in what they require.
3. Mutualistic-development account
Imagine several abilities begin relatively differentiated in childhood but help one another develop. Better reasoning helps a child learn. Greater knowledge makes later reasoning problems easier. Improved memory supports learning; learning supplies more structured material for memory and reasoning. Positive reciprocal interactions can make abilities increasingly correlated over development even without one initial common cause.
4. Network-learning account
Imagine intelligence partly as an evolving network of facts, procedures, strategies, and skills. New knowledge becomes connected to existing knowledge. People with denser or more efficiently growing networks may succeed across many tasks because learning in one area changes what can be learned in another. Again, the resulting test scores may show a positive manifold without requiring a single hidden mental substance.
Can several explanations be true at once?
Yes. Theories are often presented as competitors because they make different claims about the source of the positive manifold. But human cognition could contain several relevant mechanisms. Shared biological influences, domain-general cognitive processes, developmental feedback, schooling, knowledge acquisition, and task overlap need not be mutually exclusive. The empirical challenge is not simply to invent a mechanism capable of generating positive correlations. Many mechanisms can do that. A stronger theory must explain why the correlations have the pattern and developmental behavior they do, make predictions that differ from rival accounts, and survive tests designed to discriminate among them.
Why factor analysis cannot settle the causal question
Factor analysis is extremely useful for describing covariance. Suppose ten tests all correlate positively. A factor model may show that a general latent dimension represents a substantial portion of their shared variation. That is evidence about structure. But different causal systems can sometimes generate similar covariance structures. A common cause can produce correlations. Interacting processes can produce correlations. Developmental feedback can produce correlations. Therefore:
A good-fitting factor model does not uniquely identify the causal architecture that generated the data.
This is one of the most important principles in the entire theories-of-intelligence literature.
The observation that theories must explain
The positive manifold is the phenomenon. g is one statistical representation of its common structure. Theories of intelligence try to explain why that structure exists. The fact that cognitive tests correlate is much better established than any single causal explanation of those correlations. That is why researchers can strongly agree that general cognitive covariance is real while disagreeing about what produces it.
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: A Survey of Factor-Analytic Studies. Cambridge University Press. DOI
- van der Maas, H. L. J., et al. (2006). A dynamical model of general intelligence: The positive manifold of intelligence by mutualism. Psychological Review, 113, 842–861. DOI
- Kovacs, K., & Conway, A. R. A. (2016). Process Overlap Theory: A unified account of the general factor of intelligence. Psychological Inquiry, 27, 151–177. DOI
- Savi, A. O., Marsman, M., van der Maas, H. L. J., & Maris, G. K. J. (2019). The wiring of intelligence. Perspectives on Psychological Science, 14, 1034–1061. —. DOI
