Cogniusy
Back to blog

Does IQ Measure Intelligence? What the Research Actually Shows

By Cogniusy Team — Psychometric Review

"Does IQ measure intelligence?" sounds like a simple yes-or-no question. It isn't. IQ tests measure something real, stable, and predictive — but "intelligence" is a much bigger, fuzzier concept than any single number can capture, and the research on the brain biology, genetics, and malleability of IQ is more nuanced than either IQ boosters or IQ skeptics usually let on. This article works through three separate questions using peer-reviewed sources: what an IQ score actually reflects, what the brain evidence says about it, and whether training or education can change it.

What IQ scores are actually built on

Modern intelligence research starts from a well-replicated empirical fact: people who do well on one type of cognitive test — vocabulary, matrix reasoning, working memory, processing speed — tend to also do well on the others. This positive correlation across very different tasks is what statisticians extract as a single underlying factor, usually called g, or general cognitive ability. IQ scores are a way of quantifying a person's standing on that factor relative to others.

As Deary, Cox, and Hill put it in their 2022 review in Molecular Psychiatry, Genetic variation, brain, and intelligence differences:

"Individual differences in human intelligence, as assessed using cognitive test scores, have a well-replicated, hierarchical phenotypic covariance structure. They are substantially stable across the life course, and are predictive of educational, social, and health outcomes."

That's a stronger empirical claim than people usually give IQ credit for. The structure of test scores (the fact that diverse cognitive tasks cluster together into a general factor) is one of the most consistently replicated findings in psychology, and scores correlate with real-world outcomes — school performance, occupational attainment, even longevity — well beyond what chance would predict. That's the case for IQ measuring something meaningful.

The case for caution is that g is a statistical abstraction, not a physical thing you could point to on a brain scan. It's extracted from whichever tests happen to be in the battery, and it says nothing directly about creativity, emotional intelligence, practical know-how, or the kind of context-specific problem-solving that matters most in daily life. IQ measures a form of cognitive ability reliably. Whether that is the whole of "intelligence" as most people use the word is a definitional question, not just an empirical one.

Does the brain evidence back it up?

If g reflects something real about the brain, you'd expect it to correlate with measurable brain properties — and it does, though the effect sizes are more modest than popular science articles often imply.

The foundational review here is Deary, Penke, and Johnson's 2010 Nature Reviews Neuroscience paper, The neuroscience of human intelligence differences, which established that intelligence differences track two things at the brain level: overall brain volume, and the efficiency of white-matter connections between parietal and frontal regions — the basis of what's called the Parieto-Frontal Integration Theory (P-FIT). In plain terms: it's not just how much brain tissue you have, it's how well-connected and efficiently organized specific circuits are.

The follow-up 2022 review updates these effect sizes with more recent, larger-sample data. A meta-analysis of brain volume and intelligence found:

"The association at r = 0.24," rising to "r = 0.31" in healthy adults and "r = 0.39" when using higher-quality intelligence testing.

For white matter specifically:

"Higher FA and lower MD are associated with higher intelligence test scores in studies together covering ages 8–81 years, typically with small effect sizes."

And for cortical thickness and volume, "effect sizes remain relatively low, at an upper limit of ~r = 0.30." (Deary, Cox & Hill, 2022)

An r of 0.24–0.39 is a real, reproducible association — but it also means brain structure alone explains well under 20% of the variance in test scores. The authors are explicit that this rules out using brain scans to size someone up individually:

Polygenic scores and brain measures "do not predict well; neither is of practical use for predicting the intelligence of an individual." (Deary, Cox & Hill, 2022)

In other words: at the population level, IQ tracks something biologically real about brain structure and connectivity. At the individual level, no scan or genetic test can substitute for actually giving someone a test.

Is IQ mostly genetic, then?

This is where the same 2022 review is unusually precise, and where the heritability ≠ fixed distinction really matters.

Twin and family studies put the heritability of intelligence at roughly 50% overall — but that number moves a lot across the lifespan:

"Genetic differences account for about 50% (standard error [SE] about 2%) of the variation in intelligence... Higher heritability estimates are found in samples of adults (where it can be 70% or slightly more) than in children (where estimates as low as 20–30% have been reported)." (Deary, Cox & Hill, 2022)

This counterintuitive pattern — heritability rising with age rather than falling — is known in the literature as the Wilson effect. The likely explanation is that as people age, they increasingly select and shape their own environments in ways that amplify their genetic predispositions, while the shared-family-environment effects that dominate in childhood fade out. DNA-based (SNP) heritability estimates, which capture only common genetic variants directly, come in lower — around 20–30% — reflecting the fact that twin studies also pick up rarer variants and gene-gene interactions that SNP arrays miss.

Crucially, the authors reject the leap from "heritable" to "fixed":

"The associations should not be taken to mean that there are immutable contributions to intelligence." They add: "Probably, intelligence and education probably have a dynamic bi-directional, and possibly causal, association." (Deary, Cox & Hill, 2022)

Heritability describes how much of the variation between people in a given population and environment is attributable to genetic differences — it is not a measure of how fixed or unchangeable a trait is for any one person. A highly heritable trait can still shift substantially if the environment shifts (height is a classic example: highly heritable, yet average height has risen several centimeters per generation in many countries due to nutrition).

Is IQ actually changing over time?

If intelligence test scores were purely a fixed biological ceiling, average scores across a population should be flat over time. They are not. The clearest recent evidence on this comes from the 2023 meta-analysis by Wongupparaj, Wongupparaj, Morris, and Kumari in the journal Intelligence, Seventy years, 1000 samples, and 300,000 SPM scores: A new meta-analysis of Flynn effect patterns.

The study pooled 1,038 independent samples and 299,155 participants across 72 countries, spanning 1948 to 2020, all using Raven's Standard Progressive Matrices — a nonverbal fluid-reasoning test often treated as one of the "purest" measures of g because it minimizes dependence on language and cultural knowledge. Their headline finding confirms the so-called Flynn effect: a long-term, generation-over-generation rise in raw test performance.

The analysis found an average "IQ gain of 0.22 points per year" across the period studied.

That's roughly 2 points per decade — enough to shift an entire population's average test performance by double-digit IQ points over the 20th century, even though nobody's underlying brain biology changed on that timescale. The gains were not uniform, either:

"Stronger [Flynn effects] were evident in middle-income countries and younger generations," and the authors argue that explaining the pattern requires "a multicausal explanatory framework" rather than a single cause. (Wongupparaj et al., 2023)

The leading explanations for the Flynn effect — better nutrition, reduced childhood disease burden, more years of schooling, smaller families, and greater cultural familiarity with abstract, test-like reasoning — are all environmental, not genetic. This is some of the strongest population-level evidence that whatever IQ tests measure is sensitive to the environment people grow up in, even though it also correlates with brain structure and is substantially heritable within any one generation. Heritability and malleability are not opposites; both can be true at once.

So: does training raise your IQ?

Given that population-average scores can rise over decades, it's natural to ask whether an individual can meaningfully raise their own score. The honest answer, based on separating out three different effects, is: it depends what you mean by "training."

Repeating an IQ test. Simple retest familiarity — learning the format, strategies, and sometimes specific items — reliably raises scores on that test. This is not an increase in underlying ability; it's why test publishers norm scores against fresh samples and why score gains from retaking the same test are treated cautiously in clinical and admissions settings.

Narrow cognitive training. Programs that drill working memory, n-back tasks, or specific puzzle types reliably improve performance on the trained task and closely related ("near transfer") tasks. Evidence for transfer to overall intelligence test scores ("far transfer") is much weaker once studies use active control groups and pre-registered designs — a pattern documented in meta-analyses of brain-training interventions that is broadly consistent with what the neuroscience literature above would predict: narrow training changes performance on a narrow circuit, not the broader parieto-frontal network that P-FIT associates with general ability.

Education and sustained cognitively demanding experience. This is the category with the strongest causal evidence. Quasi-experimental studies that exploit natural experiments — changes in compulsory schooling laws, cutoff birth dates for school entry — consistently find that additional years of education raise IQ-test scores, on the order of 1–5 points per year of schooling. Unlike a few weeks of brain games, this involves years of sustained exposure to reading, abstract reasoning, structured problem-solving, and novel information across many domains — plausibly enough to shift the same brain networks that the neuroimaging research associates with g, and it lines up with the Flynn effect data above, where the countries and cohorts showing the largest gains are exactly the ones that saw the largest expansions in schooling, nutrition, and health.

The bottom line

IQ tests measure a real, statistically robust, and predictively useful construct — general cognitive ability, g — that correlates with brain volume and white-matter connectivity at correlations in the 0.2–0.4 range, is roughly 50% heritable in the general population (more so in adults than children), and yet has risen substantially across generations as living conditions and education have improved (Deary, Penke & Johnson, 2010; Deary, Cox & Hill, 2022; Wongupparaj et al., 2023).

None of that means IQ is intelligence in the full, everyday sense of the word — creativity, wisdom, emotional attunement, and practical judgment sit outside what any timed test can capture. But within its own domain, IQ is neither an arbitrary number nor a fixed ceiling stamped in at birth. It's a stable, biologically grounded, and environmentally sensitive estimate of one important kind of cognitive ability — which is also why familiarity with a test's question formats moves the score you get on that test, even when it does nothing to the underlying ability the test is trying to estimate. If you want the mechanics of how raw answers become a number like 100 or 130, see our companion piece on how IQ scores are calculated.

Sources