Exercise and IQ: Promising, Not Guaranteed

0 The Short Answer

In children, structured exercise programs do appear to raise scores on intelligence tests by a small amount. In healthy adults, nobody has shown a gain in measured IQ at all. The strongest single piece of evidence is a 2024 meta-analysis in Pediatrics by Morales, Valenzuela, Martínez-de-Quel and colleagues, pooling 14 randomized controlled trials and 3,203 participants aged 5 to 14. It found a standardized mean difference of 0.54 favoring the exercise groups, with a 95 percent confidence interval running from 0.11 to 0.97.

That interval is the whole story, and almost every page you will read about exercise and IQ hides it. A standardized difference of 0.54 sounds like roughly 8 points on a scale with a standard deviation of 15. The bottom of the interval is worth under 2 points. The top is nearly 15. The same analysis, when the authors expressed the result on the IQ metric itself, came out at about 4 points. So the honest summary is not a number. It is a direction plus an admission: the effect is probably positive, it is probably modest, and the data are not precise enough to tell you how modest.

Anyone who tells you that jogging buys you 8 IQ points has taken the midpoint of a wide interval, multiplied it by 15, and quietly deleted the error bars. Anyone who tells you exercise does nothing for cognition has ignored a statistically significant pooled result from randomized trials. Both readings are wrong in the same way, by treating an uncertain estimate as a settled fact. This page walks through what the trials did, what a standardized effect size actually converts to, which cognitive domains moved and which did not, why the pediatric result cannot be transplanted onto adults, and what any of it means for you on the morning of a test.

1 What a Trial Like This Actually Measures

Before any number means anything, you need a clear picture of the experiment producing it. A typical trial in this literature takes a few hundred schoolchildren, administers a standardized cognitive battery, randomly assigns each child to an exercise program or to a control condition, runs the program for somewhere between eight weeks and a school year, and then administers a battery again. The published effect is the difference between how much the exercise group's scores changed and how much the control group's scores changed.

Notice what that difference contains. It contains any genuine change in cognitive capacity, which is what everyone wants it to mean. It also contains practice effects, because a child who has seen a matrix reasoning task before is faster on the second sitting. It contains motivational and rapport differences, because a child who has spent three months with enthusiastic coaches may sit down with a different attitude than one who has not. It contains regression to the mean when groups start unevenly. And in a design where nobody can be blinded, because you always know whether you have been jumping around a gymnasium three times a week, it contains expectancy on the part of children, teachers and sometimes the examiners themselves.

The control condition is the pivot on which all of this turns. Comparing an exercise program against a group that received nothing at all measures exercise plus attention plus novelty plus adult contact. Comparing it against an equally engaging non-physical activity isolates the physical component much better and, predictably, tends to shrink the estimate. Most trials in this field use passive or usual-activity controls, which is one structural reason to expect the pooled figure to sit above the true causal effect rather than below it.

None of this makes the research worthless. Randomization still rules out the biggest confounders that plague observational work, and that is a real achievement. It does mean that the phrase "exercise raises IQ" is doing more work than the design can support. What these trials license is narrower and more careful: children assigned to structured physical activity programs score somewhat higher on intelligence batteries afterward than children who were not. Whether that reflects durable ability, better test-taking conditions, or some blend of the two is a question the designs cannot fully separate. If you want the fuller picture of what these batteries are built to capture in the first place, what IQ tests actually measure is the necessary companion to this page.

2 The Pediatrics Meta-Analysis in Detail

The 2024 Pediatrics paper (volume 154, issue 6, article e2023064771) is the current reference point, so it is worth unpacking rather than merely citing. The authors searched PubMed, Web of Science, PsycINFO and Scopus from inception through February 2024 for randomized controlled trials measuring intelligence outcomes in participants aged 19 or under. Fourteen trials met the criteria, together enrolling 3,203 children and adolescents aged 5 to 14. The pooled random-effects estimate for general intelligence was a standardized mean difference of 0.54 relative to controls, with a 95 percent confidence interval of 0.11 to 0.97 and a p value of .01.

Fourteen is not a large number of trials. That matters because a random-effects pool of 14 studies with variable designs, ages, instruments and program lengths produces exactly the kind of wide interval reported here. The interval is not an editorial flourish; it is the honest output of pooling heterogeneous evidence. When a meta-analysis reports a confidence interval nine times wider at the top than at the bottom, the correct interpretation is that the studies disagree with each other substantially and the pool is doing its best to average that disagreement.

The paper reports two further findings that deserve more attention than they usually get. First, fluid intelligence measures showed a benefit as well, which fits the broader pattern in this literature where reasoning-on-the-spot tasks respond more readily than stored knowledge. Second, crystallized intelligence was assessed in only one of the included trials, the ActiveBrains study led by Ortega with 109 children who had overweight or obesity, which found a beneficial effect. One trial is a starting point, not a body of evidence, and it would be a misreading to present crystallized ability as an established beneficiary. The difference between those two families of ability is worth understanding on its own terms, and fluid versus crystallized intelligence covers it properly.

The authors also report that the association held across subgroups: children with different baseline ability levels, different ages within the sampled range, and interventions of different lengths all showed the pattern. Consistency across subgroups is reassuring about the direction of the effect. It says nothing about its size, and it is precisely the size that people want and that the data decline to deliver.

3 Turning 0.54 Standard Deviations Into Points

Standardized mean difference is a unit designed to let researchers pool results from instruments that do not share a scale. It expresses a group difference in standard deviations of the outcome measure. To convert it into IQ points, you multiply by the standard deviation of the IQ scale, which on modern batteries is 15. Multiply 0.54 by 15 and you get 8.1 points. That single multiplication is the origin of nearly every inflated headline in this topic area.

Here is why the multiplication misleads. The standardizing denominator in each trial is the standard deviation observed within that trial's sample, not the standard deviation of the general population. Trial samples are frequently narrower than the population, because they are recruited from a few schools, or a single clinical category such as children with overweight, or one narrow age band. A narrower sample deflates the denominator, which inflates the standardized effect without any change in the underlying points gained. Convert that inflated figure back using the population's standard deviation of 15 and you have laundered a modest gain into a dramatic one.

The check on this is available in the same paper. When the results are expressed on the IQ metric directly, the improvement over control comes out at roughly 4 points, about half the naive conversion. Four points is a real quantity and a meaningful one at a population level. It is also small enough to sit comfortably inside the measurement error of a single administration of a good test, which is typically around 4 to 5 points at the 95 percent level for a full-scale score.

A confidence interval is a statement about which values the data fail to rule out. This one fails to rule out an effect so small it is clinically invisible, and it fails to rule out an effect large enough to reshape education policy. Both remain live. That is not a defect of the researchers, who reported the interval plainly; it is a defect of the evidence base, which is still thin. The intellectually honest sentence is: exercise programs in childhood are associated with better intelligence test performance, best estimated at a few points, with genuine uncertainty about whether the true figure is closer to one or to twelve.

4 What Kind of Exercise Was Actually Studied

"Exercise" in this literature is not one thing, and the differences between programs matter more than most summaries admit. The trials pooled in the pediatric evidence include aerobic training in the familiar sense (running games, circuit sessions, cycling), coordinative and motor-skill training that emphasizes balance and precise movement, multicomponent school programs that mix aerobic work with games and skill practice, and in some cases yoga or perceptual motor training. These are cognitively different activities that happen to share a physical component.

A 2025 Bayesian network meta-analysis in Frontiers in Physiology attempted to rank those modalities across 15 randomized trials and about 3,400 participants. Dual-task balance training, meaning balance work performed while a cognitive demand runs in parallel, ranked highest across general, fluid and crystallized measures, with standardized differences reported around 0.93, 0.97 and 1.05. The same analysis suggested that programs delivering roughly 220 minutes per week across three sessions and running at least eleven weeks were the most effective.

Treat those rankings with real caution. Network meta-analysis with 15 trials spread across six modalities means some comparisons rest on one or two studies each, and rank orders derived from such sparse networks are notoriously unstable. What survives the caution is a qualitative pattern that shows up repeatedly: programs that demand cognitive engagement alongside physical effort tend to outperform programs that are physically demanding but cognitively empty. Sprinting on a treadmill and playing a fast tactical game are not equivalent stimuli even at identical heart rates.

Duration follows a similar pattern. Most included programs ran on the order of two to six months, which is long enough for physiological adaptation and short enough for a school calendar. There is no credible evidence for a one-week protocol producing measurable ability change, and there is no evidence at all supporting the idea that a single hard session before an assessment does anything useful for a reasoning score. The interventions that produced these findings were sustained habits, delivered by instructors, over months. Any claim that compresses that into a quick trick has left the evidence behind entirely.

5 Which Cognitive Domains Move and Which Stay Put

The unevenness of the response across domains is one of the most informative features of this literature, and it is routinely flattened into a single claim about "brain health." A 2023 multilevel meta-analysis in Educational Psychology Review, covering 92 randomized trials and 25,334 typically developing children aged 5 to 12, broke outcomes down by sub-domain rather than pooling them. The largest effects landed on on-task behaviour, with a reported effect size around g equals 1.04, and on creativity measures at around g equals 0.70, while executive function components and academic achievement sub-domains were analyzed separately and showed a more varied picture.

On-task behaviour deserves a flag rather than a celebration. It is a classroom observation of whether a child is attending to assigned work, not a measure of cognitive capacity. A very large effect on it is entirely consistent with the mundane and useful conclusion that children who have moved recently sit still better, which improves the conditions for learning without changing the underlying ability. Confusing that with an intelligence gain is one of the commonest errors in the popular coverage.

That gradient tells you something about mechanism. Domains that depend on your state right now respond most. Domains that depend on what you have accumulated over a decade respond least. In the language of the CHC framework that organizes modern batteries, you would expect the clearest movement in working memory and processing speed, some movement in fluid reasoning, and very little in verbal comprehension, which is exactly the shape the trials produce. If a study reports that a two-month exercise program raised vocabulary knowledge substantially, the most likely explanation is something other than the exercise.

6 Why the Children's Result Does Not Transfer to Adults

The pediatric finding is regularly recycled into advice for adults, and the leap has no support behind it. The trials establishing it enrolled participants aged 5 to 14, an age range in the middle of rapid neural maturation, motor skill acquisition and formal schooling. Interventions delivered during a period of high plasticity can plausibly do things they cannot do later. Assuming otherwise is the same reasoning error as concluding that because a nutritional intervention affects growth in children, it will make adults taller. The analogy is closer than it looks, since height and measured ability really do travel together: 6,815 adults in the Generation Scotland cohort put the correlation between height and cognitive ability at 0.16, a figure produced by shared upstream causes such as childhood nutrition and family circumstances rather than by either trait acting on the other.

For adults, the closest comparable evidence targets cognitive function rather than intelligence as such. Northey, Cherbuin, Pumpa and colleagues, in the British Journal of Sports Medicine in 2018 (volume 52, pages 154 to 160), pooled 333 effect sizes from 36 randomized trials in community-dwelling adults over 50 and found an overall improvement in cognitive function of 0.29, with a 95 percent confidence interval of 0.17 to 0.41. Resistance-based programs came in around 0.13 to 0.22.

Read that carefully, because it is often misreported. Those are cognitive function tasks in older adults, many of whom are on a declining trajectory, and the outcome is not an IQ score. Slowing or partially offsetting age-related decline in processing speed or memory is a genuinely valuable result and a different claim from raising a healthy adult's measured intelligence. The literature contains no randomized demonstration of the latter. When someone cites the older-adult exercise findings as evidence that going to the gym will raise your IQ at 30, they are substituting one population, one outcome and one direction of change for another.

There is a related asymmetry worth stating plainly. Almost every intervention that appears to raise cognitive scores works most visibly where there is a deficit to correct: poor sleep, low fitness, nutritional shortfall, an environment low on stimulation. Repairing a deficit moves a score. Adding more of a good thing to someone who already has enough of it typically does not. If you already train several times a week and sleep adequately, the trial evidence gives you no reason to expect a cognitive dividend from training harder, and it never claimed to.

7 Plausible Biology, Stated Without Overclaiming

There are real mechanistic candidates here, and they are worth describing precisely, because vague gestures at neuroplasticity are how honest research gets turned into supplement marketing. Whole neuromyths live in that slippage, and the left brain right brain story is the standing example: language genuinely is handled mostly on the left in the great majority of people, and nothing in that fact licenses the claim that a person is governed by a dominant hemisphere.

• Brain-derived neurotrophic factor. Exercise raises circulating levels of BDNF, a protein involved in neuronal survival and synaptic plasticity. In animal work this connects fairly directly to hippocampal neurogenesis. In humans, what is measured is usually serum concentration, which is a peripheral proxy of uncertain relationship to what happens inside the brain.
• Structural change in the hippocampus. Erickson and colleagues, in PNAS in 2011, randomized 120 older adults to a year of aerobic training or stretching and reported roughly a 2 percent increase in anterior hippocampal volume in the exercise group, associated with higher serum BDNF. A published letter by Coen and colleagues in the same journal disputed whether the memory improvement could be attributed to either the aerobic training or the volume change, which is the kind of exchange that rarely survives into secondary coverage.
• Cerebral blood flow and arousal. Exercise acutely increases cerebral perfusion and catecholamine availability, which is the most likely route by which a single session shifts attention and reaction time for a while afterward. This is a state effect that decays, not a change in capacity.

Each of these is defensible as a mechanism candidate. None of them has been shown to be the pathway that produced the 4-point difference in the children's trials, and no study has traced that chain end to end from a training program through a biological marker to a psychometric outcome in the same participants. A mechanism that is plausible is not a mechanism that is demonstrated, and the gap between the two is where most overclaiming lives.

The animal-to-human translation deserves its own warning. Rodent studies showing running-wheel access driving neurogenesis and maze performance are frequently cited as though they settle the human question. They involve species with different brains, enclosure conditions that make the sedentary comparison group profoundly deprived rather than merely inactive, and outcomes with no psychometric equivalent. They are excellent for generating hypotheses. They are not evidence about your score.

8 Reverse Causation, Confounding, and the Skeptical Case

Outside the randomized trials sits a much larger observational literature reporting that fitter people score higher, and that literature is far weaker than its volume suggests. The direction of causation is genuinely ambiguous. Higher measured ability predicts better health behaviour, better adherence to routines, higher educational attainment and higher income, all of which make regular exercise easier to sustain. A correlation between fitness and cognition is compatible with fitness improving cognition, cognition improving fitness, or a third variable such as household resources or chronic illness driving both. Measuring ability first closes off one of those three readings, which is why the cognitive epidemiology literature is worth knowing here: it starts from a test sat at eleven, finds that higher scorers later smoked less and moved more, and still cannot say how much of that is the score rather than the circumstances that came with it.

The most informative observational dataset in this area is Åberg and colleagues' 2009 PNAS analysis of 1,221,727 Swedish men who underwent conscription examinations at age 18, including 268,496 full-sibling pairs and 1,432 monozygotic twin pairs. Cardiovascular fitness, but not muscle strength, was associated with cognitive performance. The sibling and twin comparisons are what make it valuable, because they hold shared family background constant. They still cannot randomize, and a fitter twin at 18 may have been the more energetic, healthier or better-coordinated one at 8.

The sharpest skeptical challenge comes from Ciria, Román-Caballero and colleagues, whose umbrella review appeared in Nature Human Behaviour in 2023. They compiled 24 meta-analyses covering 109 randomized trials and 11,266 healthy participants, and found low statistical power in the primary trials, selective study inclusion, publication bias and wide variation in analytic choices. Their own pooled estimate was 0.22, dropping to 0.13 after accounting for key moderators and to 0.05 after correcting for publication bias. Their conclusion was that the evidence does not support a causal effect of exercise on cognitive performance in healthy populations.

That paper drew a formal reply in the same journal in 2024, titled as a warning against underestimating the cognitive benefits of exercise, from researchers who disputed the methodological and theoretical choices behind the null conclusion. The dispute is unresolved, and pretending otherwise in either direction would be dishonest. What it establishes for a reader is the correct posture: this is an active scientific disagreement between competent groups, not a settled fact with a fringe of doubters. Any page that presents the cognitive benefits of exercise as established beyond argument has not read the last three years of the literature.

9 What the Evidence Cannot Promise You

Here is the list of things that get claimed on this topic and cannot be supported by anything published, stated flatly so you can use it as a filter.

• A specific number of points for you. Group averages do not distribute evenly across individuals. A pooled 4-point difference between two groups of children is consistent with most participants changing very little and a minority changing more, and nothing in the data identifies which participant you would be.
• Any gain for a healthy adult's IQ. The randomized intelligence evidence concerns ages 5 to 14. The adult evidence concerns cognitive function tasks, mostly in people over 50, with a pooled effect of 0.29 that is not an IQ score.
• Permanence. The trials measured outcomes at or near the end of the intervention. Whether any difference persists after a program ends is largely untested, and the default expectation for behaviourally-induced score changes is decay.
• A best exercise. Modality rankings exist but rest on thin networks. The recurring signal favors cognitively engaging movement over mechanical repetition, which is a hypothesis worth holding loosely, not a prescription.
• A same-day boost to a reasoning score. Acute sessions can shift attention and speed briefly. Chang, Labban, Gapin and Etnier documented small acute effects on cognitive performance in Brain Research in 2012. Small, brief and attention-weighted is not the same as reasoning ability improved.

There is a broader point behind the list. Measured intelligence is a comparatively stable construct, which is why it predicts outcomes across decades. Stability and easy modifiability are in tension: a quantity that could be moved several points by any twelve-week program would not have the longitudinal track record that makes IQ interesting in the first place. When you see a large, cheap, fast cognitive gain advertised, the prior should be skepticism, and that applies to exercise claims exactly as much as to brain-training apps. If you want to know what score ranges actually correspond to in the population before evaluating any promised gain, the percentile chart and what average IQ means are the right calibration.

10 What to Actually Do Before a Cognitive Test

Given all of the above, the practical guidance for the day of an assessment is short and slightly boring, which is a sign it is honest. Sleep is the one input on test day with a clear direction, and it works by subtraction rather than addition: what a short night actually costs is sustained attention, reaction speed and error monitoring, while vocabulary and accumulated knowledge sit roughly where they were yesterday.

Keep your normal routine. If you usually train in the morning, train in the morning. If you never do, test day is a poor time to start. The value of routine here is not mystical; it is that an unfamiliar physiological state adds noise to your performance, and noise on a test day is pure downside. There is no upside large enough in the literature to justify introducing an experiment into your own morning.

Do not attempt a hard session shortly before sitting down. The acute effects documented in this area are small, they concentrate on attention and reaction speed rather than reasoning, and they compete directly against fatigue, dehydration and elevated arousal, which reliably degrade performance on timed working memory and processing speed tasks. A hard workout an hour before an assessment is a gamble with a small possible gain and a substantial possible loss. Light movement, a short walk, is a reasonable way to reduce restlessness without incurring that cost.

Prioritize the things with larger, better-documented effects. Being adequately rested, not hungry, not sick, in a quiet room, and unhurried will do more for your score than any exercise decision you can make that morning. Test conditions are the single most controllable source of variance in a self-administered assessment, and they are routinely ignored by people who then spend an hour optimizing their pre-test caffeine.

Over the long run, exercise because of what it reliably does. Cardiovascular health, mood, sleep quality and metabolic function are all supported by evidence far sturdier than anything in this article. If a cognitive dividend arrives on top, it is a bonus rather than the justification. Framing exercise as an IQ tactic sets you up to feel cheated by a program that was doing you enormous good on every measure that is actually well established.

11 How to Read the Next Headline About Exercise and the Brain

This topic generates a steady stream of coverage, and four questions will let you assess almost any of it in under two minutes. The brain training and gaming literature went through exactly this cycle, ending in a scientific consensus statement and a federal settlement.

Was it randomized, and against what control? Observational fitness-and-cognition correlations are nearly uninformative about causation. Among randomized studies, an active control that matches the social and attentional conditions of the exercise arm is worth far more than a group that received nothing.

What was the outcome, exactly? "Improved brain function" can mean a full intelligence battery, a single executive function task, a classroom behaviour rating, a reaction time measure, or a brain imaging index with no behavioural counterpart. These are not interchangeable, and the substitution is where most exaggeration enters.

Where is the confidence interval? An article that reports a point estimate without its interval has removed the information you need. If the interval spans from trivial to large, as it does in the pediatric intelligence pool, then the headline number is a midpoint and not a finding.

Who was studied? Children, older adults, clinical populations and healthy young adults respond differently, and results are constantly transplanted across those boundaries. A finding in 70-year-olds with mild cognitive impairment tells you very little about a healthy 25-year-old, and the reverse is equally true.

Apply those four to this very page as a test of good faith. It reports a randomized pool with a passive-control problem, an outcome that is a genuine intelligence battery, an interval that is uncomfortably wide and stated in full, and a sample of 5 to 14 year olds that is explicitly not generalized to adults. That is what the evidence supports. Anything more confident than that, from any source including this one, would require better data than currently exists.

12 Measuring a Profile Instead of Chasing a Number

One reason the exercise question stays confusing is that it is almost always posed about a single composite score. "Does exercise raise IQ" collapses six distinguishable domains into one figure and then asks whether that figure moves. The trial evidence answers in a much more textured way: attention and executive components respond most, fluid reasoning responds somewhat, accumulated verbal knowledge barely responds at all. That texture is invisible if the only thing you ever look at is a full-scale number.

This is the practical case for domain-level reporting in any assessment worth taking. A profile that separates verbal comprehension, fluid reasoning, visual spatial processing, working memory, processing speed and quantitative reasoning tells you where your own performance is stronger and weaker. A composite tells you one thing about where you sit overall, and averages away everything that would let you interpret a change.

ACIS is an online self-assessment rather than a clinical or diagnostic instrument, and it reports results as a domain profile with confidence intervals rather than as a bare number, precisely because a single figure invites the kind of overinterpretation this page has spent several thousand words resisting. It cannot tell you whether your training program is helping your cognition, and no self-administered test can, because a difference of a few points falls inside normal measurement variation between sittings. What it can give you is a structured starting picture across domains, reported with its own uncertainty made visible. If you are weighing options, how to evaluate an IQ test and what counts as a good score are the sensible next reads.

13 Frequently Asked Questions

14 Sources Behind This Page

The relationship discussed here comes from published research, and the honest reading includes its limits. These are the primary sources behind the numbers on this page.

• MedlinePlus, U.S. National Library of Medicine. Exercise and Physical Fitness. The NIH consumer reference on what physical activity is documented to do.
• Harada, C.N., Natelson Love, M.C. & Triebel, K. (2013). Normal cognitive aging. Clinics in Geriatric Medicine, 29(4), 737-752. Open access. Fluid abilities peak in the third decade while vocabulary and knowledge hold or improve into the seventies.
• Schmidt, F.L. & Hunter, J.E. (1998). The validity and utility of selection methods in personnel psychology. Psychological Bulletin, 124(2). The meta-analysis behind general ability as the strongest single predictor of job performance.
• Strenze, T. (2007). Intelligence and socioeconomic success: a meta-analytic review of longitudinal research. Intelligence, 35(5).
• Deary, I.J., Strand, S., Smith, P. & Fernandes, C. (2007). Intelligence and educational achievement. Intelligence, 35(1). A five year cohort of over 70,000 English schoolchildren.
• Kuncel, N.R., Hezlett, S.A. & Ones, D.S. (2004). Academic performance, career potential, creativity, and job performance: can one construct predict them all? Journal of Personality and Social Psychology, 86(1).
• Whalley, L.J. & Deary, I.J. (2001). Longitudinal cohort study of childhood IQ and survival up to age 76. BMJ, 322(7280). Open access.
• Batty, G.D. et al. (2007). IQ and mortality: a population based cohort study. International Journal of Epidemiology, 36(3), 538-553.
• Zagorsky, J.L. (2007). Do you have to be smart to be rich? The impact of IQ on wealth, income and financial distress. Intelligence, 35(5).
• Judge, T.A., Colbert, A.E. & Ilies, R. (2004). Intelligence and leadership: a quantitative review and test of theoretical propositions. Journal of Applied Psychology, 89(3).