IQ and Video Games: Small Association

0 The Short Answer

Gamers do tend to score a little higher on some cognitive tasks, and there is still no convincing evidence that playing games caused it. Those two sentences have to sit next to each other, because almost every popular article about this topic keeps only the first half. The gap between them is where the entire scientific argument of the last twenty years has been fought, and the people fighting it are not cranks on either side: they are the labs that produced the original findings and the labs that tried to reproduce them.

What the evidence actually supports is narrow. Practice at an action game reliably improves performance at that game, and it often improves performance at laboratory tasks that resemble the game, tracking several moving dots, spotting a target in visual clutter, judging where something appeared in the periphery. What it does not reliably do is raise the underlying capacity those tasks are supposed to sample. When researchers correct for publication bias, for participants who knew what result was expected of them, and for the fact that people who already have quick visual attention are more likely to enjoy fast games in the first place, the effect on general ability shrinks toward nothing.

This page walks through both sides in order: what Daphne Bavelier and Shawn Green found and why it was genuinely exciting, what the methodological critiques established and why they were serious rather than pedantic, how the near versus far transfer distinction resolves most of the confusion, and what the Federal Trade Commission concluded when it looked at the commercial version of the same promise. It ends with the part that matters most to anyone reading an IQ score: why getting better at a test is not the same as getting better.

1 What Green and Bavelier Actually Reported

The modern literature starts with a single short paper: C. Shawn Green and Daphne Bavelier, "Action video game modifies visual selective attention," published in Nature in 2003, volume 423, pages 534 to 537. It contained five experiments. The first four compared habitual action game players against non-players on tasks measuring visual attention, including the capacity to enumerate briefly presented items, the spatial extent over which attention can be distributed, and the ability to handle attentionally demanding displays. Players outperformed non-players on all of them. A measure that narrow describes one visual attention skill rather than general ability, which is why an online IQ test that samples ability broadly reports several domain scores before any composite.

Cross-sectional comparisons like that prove nothing about causation on their own, and the authors knew it, which is why the fifth experiment mattered so much. They trained non-players on an action game and tested them again. The trained group improved relative to their own pre-training performance. That design turned an observation about who plays games into a claim about what games do, and it is the reason a two page paper became one of the most cited findings in applied cognitive psychology.

The finding was exciting for a specific reason that is easy to miss now. Perceptual learning had a long reputation for being stubbornly narrow. Train someone to detect a particular orientation at a particular retinal location and they get better at exactly that, often not even transferring to the other eye. Green and Bavelier reported improvements across several different visual attention measures at once, which suggested that something more general had shifted. If a commercial entertainment product could produce broad perceptual gains in ten days, that was worth understanding.

Bavelier's group spent the following two decades extending the work, and in 2018 published a meta-analysis of the field with Benoit Bediou as first author in Psychological Bulletin, volume 144, pages 77 to 110. It reported a cross-sectional advantage for habitual action gamers of about half a standard deviation (g = 0.55) and a smaller intervention effect of about a third of a standard deviation (g = 0.34), concentrated in top-down attention and spatial cognition. Notably, the authors themselves estimated that publication bias inflated the reported effects by roughly 30 percent and called for larger studies with more than 30 hours of training.

2 The Objections, and Why They Are Not Nitpicking

Three problems dominate the critical literature, and none of them is a general complaint about sloppiness. Each one is a specific mechanism that can manufacture a positive result out of nothing.

Recruitment tells participants what to expect. Walter Boot, Daniel Blakely and Daniel Simons laid this out in Frontiers in Psychology in 2011. If you advertise a study for "expert video game players" and compare them against people recruited as non-gamers, both groups arrive with a theory about what the experimenter wants. Cross-sectional gamer advantages are then partly an artifact of who answered which flyer.

Active control groups do not automatically fix it. The standard defense is to give the control group a different game rather than nothing at all. Boot, Simons, Christopher Stothart and Cary Stutts argued in Perspectives on Psychological Science in 2013 that this is insufficient. An active control only removes placebo effects if the control participants expect to improve as much as the treatment participants do, and almost no study in this literature measured expectations to check. Someone assigned to a fast shooter and someone assigned to a puzzle game do not hold the same beliefs about what is happening to their brain.

The published record is filtered. Joseph Hilgard, Giovanni Sala, Boot and Simons re-analyzed the exact study set from the 2018 Bediou meta-analysis and published the result in Collabra: Psychology in 2019. With stronger corrections for publication bias, they found minimal evidence that action game training transfers to cognitive ability measures. They also documented that several studies from the original lab distributed different outcomes from the same or heavily overlapping participant samples across separate publications without flagging the overlap, which inflates the apparent number of independent replications. Bavelier's group has continued the exchange, publishing a further meta-analysis with Bediou as first author in Technology, Mind, and Behavior in 2023. The disagreement is live, which is itself informative: a robust effect does not usually require this much statistical archaeology to see.

3 Near Transfer and Far Transfer: the Distinction That Decides Everything

Almost every confused claim about games and intelligence dissolves once you separate two questions. Did the training improve performance on tasks that resemble the training? That is near transfer. Did it improve the underlying capacity, such that unrelated tasks drawing on the same capacity also improve? That is far transfer.

Susan Barnett and Stephen Ceci gave the field its working framework in Psychological Bulletin in 2002, arguing that transfer varies along nine separate dimensions, including knowledge domain, physical context, temporal context, social context and modality. Their conclusion is worth holding onto: because transfer distance is multidimensional, estimating one effect size for "far transfer" as a single quantity is misleading. Two studies that both call themselves transfer research may be measuring completely different distances.

Applied to gaming, the distances are easy to line up. A first person shooter trains rapid detection of peripheral onsets under time pressure. A multiple object tracking task asks you to follow several moving dots among distractors under time pressure. That is close, and it is where the positive results cluster. A matrix reasoning item asks you to induce an abstract rule from a pattern with no time pressure on perception at all. That is far, and it is where the results collapse.

This is not a special weakness of video games. It is the general shape of every cognitive training result ever produced, and it applies just as much to working memory drills, chess, music lessons and commercial brain games. Monica Melby-Lervåg, Thomas Redick and Charles Hulme titled their 2016 review in Perspectives on Psychological Science, volume 11, pages 512 to 534, with the finding itself: working memory training does not improve performance on measures of intelligence or other measures of far transfer. If you want to understand why your fluid reasoning score barely moves while your score on a trained task climbs steadily, the near and far distinction is the whole explanation.

4 Sala and Gobet: Three Meta-Analyses, Three Small Numbers

The most direct test of the gaming claim came from Giovanni Sala, Kingsley Tatlidil and Fernand Gobet, then at the University of Liverpool, published in Psychological Bulletin in 2018, volume 144, pages 111 to 139, under a title that leaves little room for interpretation: "Video game training does not enhance cognitive ability."

They ran three separate random-effects models rather than one, which is the methodological choice that makes the paper hard to wave away. The first model, with 310 effect sizes, examined the correlation between video game skill and cognitive ability. The second, with 315 effect sizes, compared players against non-players on cognitive ability. The third, with 359 effect sizes, tested the effect of game training on cognitive ability. All three returned small or null overall effects.

The structure matters because each model addresses a different objection. If only the training model were null, you could argue the interventions were too short. If only the correlational model were null, you could argue the skill measures were crude. Getting near zero in all three, across correlation, group comparison and intervention, is much harder to explain away as a measurement failure. Their summary was that overall cognitive ability and video game skill are only weakly related, and that they found no evidence of a causal relationship.

One detail in their first model deserves attention because it is quietly damaging to the popular version of the story. If gaming built general capacity, then people who are excellent at games should be measurably stronger in general, and across 310 estimates that link came back weak. Skill at a demanding competitive title is real expertise, built through thousands of hours of deliberate practice, pattern learning and tuned motor timing. It just behaves like every other expertise: deep inside its own domain and shallow outside it, the same way a chess grandmaster's positional judgment stops helping the moment the pieces come off the board.

Their framing of the result is the part worth remembering. They did not treat gaming as a uniquely disappointing case. They treated it as one more instance of a pattern already documented in working memory training, music training, chess training and commercial brain training: skills stay where they were built. Gaming, in their words, represents no exception to the general difficulty of obtaining far transfer.

5 What Happens When You Meta-Analyze the Meta-Analyses

In 2019 the same research group went one level up. Sala, together with Deniz Aksayli, Tatlidil, Yasuyuki Tatsumi, Yasuyuki Gondo and Gobet, published a second-order meta-analysis in Collabra: Psychology, pooling the results of existing meta-analyses rather than individual studies. It is one of the cleanest summaries available of what cognitive training does. One adjacent literature does come out of the same machinery with a positive number, and its boundary is the instructive part: the 2024 Pediatrics meta-analysis of 14 randomized exercise trials found a small gain in children, while no randomized study has shown a rise in measured IQ in healthy adults.

Their first model, covering 99 effect sizes, looked at working memory training and near transfer, meaning memory outcomes. Near transfer was real, and its size depended on the population, with children, adults and older adults responding differently. Nobody in this debate denies that training works on the trained thing.

The second model, covering 119 effect sizes, looked at far transfer from working memory training to reasoning, processing speed and language. The third model, covering 233 effect sizes, widened the net to include six more meta-analyses of other training programs: video games, music, chess and exergames. In both, far transfer effects were small or null. Then comes the sentence that should be printed on the box of every brain training product: when placebo effects and publication bias were controlled for, the overall effect size and the true variance both equaled zero.

True variance equaling zero is the technically important half. It means the remaining scatter across studies was consistent with pure sampling noise, so there was no hidden subgroup where the training secretly worked. You cannot rescue the result by proposing that it only helps children, or only helps older adults, or only works with the right genre. The authors concluded that the lack of generalization is an invariant of human cognition, which is a strong statement, and one earned by a lot of data.

6 The Evidence in Six Numbers

Arguments in this field are usually conducted in adjectives. Here are the actual quantities, each from a named study, so you can see the shape of the disagreement instead of reading about it.

Read the grid as a single sentence and it says this: the optimistic estimates come from comparisons that cannot separate cause from selection, and the pessimistic estimates come from analyses that tried to. That asymmetry is the reason the skeptical position has been gaining ground rather than losing it.

7 The Lumosity Case: What Happens When Marketing Outruns Data

On January 5, 2016, the Federal Trade Commission announced that Lumos Labs, the company behind Lumosity, had agreed to settle charges of deceptive advertising. The company paid $2 million in redress. A $50 million judgment was entered and then suspended because of the company's financial condition. The complaint was filed in the U.S. District Court for the Northern District of California, and the Commission vote authorizing it was 4 to 0. Regulatory action does not kill a brain claim, it only prices it: the left brain versus right brain personality story outlived an imaging analysis of more than a thousand brains that found no individual with an overall dominant hemisphere, and it is still sold as a training premise.

The specifics are more instructive than the headline. According to the FTC's complaint, the program consisted of 40 games marketed as training specific areas of the brain, with the promise that 10 to 15 minutes of play three or four times a week could help users reach their full potential in every aspect of life. Subscriptions ranged from $14.95 monthly to $299.95 for a lifetime membership. The company bought hundreds of Google search keywords related to memory, cognition, dementia and Alzheimer's disease.

The FTC alleged three families of claim: that training would improve performance at school, at work and in athletics; that it would delay age-related cognitive decline and protect against mild cognitive impairment, dementia and Alzheimer's disease; and that it would reduce impairment associated with conditions including stroke, traumatic brain injury, PTSD, ADHD and chemotherapy side effects. The complaint further charged that some consumer testimonials on the site had been solicited through contests offering prizes, a detail the advertising did not disclose. Jessica Rich, then director of the Bureau of Consumer Protection, put the finding plainly: the company did not have the science to back up its ads.

The order's remedy is the part worth internalizing. It did not ban the product. It required competent and reliable scientific evidence before making future claims about real-world performance or health conditions. That is the standard this entire topic runs into. Nobody has to prove that games are useless. The burden sits on whoever wants to sell you the improvement, and in 2016 a regulator found the burden unmet by the largest company in the category.

8 The Scientific Consensus Behind That Ruling

The FTC did not arrive at its conclusion in a vacuum. In 2014 two groups of scientists published duelling open letters. The first, signed by more than 70 researchers, stated that brain games do not offer a scientifically grounded route to improving cognition or slowing decline. The second, signed by 133 scientists and practitioners, replied that the literature is full of demonstrated benefits. Two expert groups, one literature, opposite conclusions.

Daniel Simons, Boot, Neil Charness, Susan Gathercole, Christopher Chabris, David Hambrick and Elizabeth Stine-Morrow set out to resolve the contradiction in Psychological Science in the Public Interest in 2016, volume 17, pages 103 to 186. Their method was pointed: rather than assemble their own study list, they evaluated the studies cited on the websites of the leading brain training companies themselves, on the reasoning that these represent the best evidence the industry can offer.

The verdict was graded rather than flat. They found extensive evidence that training improves performance on the trained tasks, less evidence for improvement on closely related tasks, and little evidence for distantly related tasks or for everyday cognitive performance. They also reported that many of the studies had design or analysis problems severe enough to preclude firm conclusions, and that not one of the cited studies met all of the best practices they identified as necessary.

Owen and colleagues had already provided the largest single data point in Nature in 2010, volume 465, pages 775 to 778. Over six weeks, 11,430 participants trained online on tasks designed to improve reasoning, memory, planning, visuospatial skills and attention. Every trained task improved. No transfer was found to untrained tasks, and the authors noted that this held even for tasks that were cognitively closely related. That last clause is the one that matters: the failure was not at the outer edge of transfer distance. It was near the starting line.

9 Children, Screen Time and What Big Observational Studies Can Show

Recent large cohorts have reopened the question with better data and a familiar limitation. Bruno Sauce, Magnus Liebherr, Nicholas Judd and Torkel Klingberg analyzed 9,855 American children from the Adolescent Brain Cognitive Development study in Scientific Reports in 2022. Their design was unusually careful: they controlled for socioeconomic status and for genetic differences in cognition, which are the two confounds that wreck most screen time research. Observational designs carry the same limits wherever they appear, including the crime and IQ literature, where detection bias does the work that selection does here.

At baseline, ages 9 and 10, time spent watching correlated negatively with intelligence (r = -0.12), socializing correlated negatively (r = -0.10), and gaming did not correlate at all. Two years later, gaming showed a positive standardized effect of +0.17 on intelligence and socializing showed none. Watching videos also showed a positive effect of +0.12, which the authors flagged as contrary to prior research, and which lost significance in a posthoc analysis controlling for parental education instead of full socioeconomic status.

That result deserves to be reported honestly in both directions. It is one of the better-controlled studies in the area and it points toward a small positive association for gaming specifically. It is also observational. Controlling for polygenic scores and family background narrows the space for confounding but does not close it, and a longitudinal association across two years cannot rule out reverse causation, where children who are developing faster gravitate toward more demanding games.

The cautionary tale sits right next to it. In October 2022, Bader Chaarani and colleagues published a comparison of 2,217 ABCD children in JAMA Network Open, contrasting non-gamers at zero hours per week against gamers at 21 or more hours per week on stop signal and n-back tasks, and reported better performance in gamers. In April 2023 the journal issued a notice of retraction and replacement after a reader raised concerns. The corrected notice acknowledges that the study had been misdescribed as case-control when it was cross-sectional, that Table 1 contained inaccurate means, standard errors, percentages and p values because different authors used inconsistent participant lists and covariates, that race and ethnicity had been wrongly reported as unavailable in the data, and that a stated absence of mental health differences between groups changed after correction for multiple testing. The main conclusion survived reanalysis by two independent statisticians, but several key findings changed. When a well-resourced team using a flagship dataset needs a reanalysis to get its own table right, treat any single headline study in this field as provisional.

10 Why Gamers Score Higher Anyway: Selection, Not Causation

If training effects are close to zero, the observed gamer advantage needs another explanation, and the leading candidate is that fast games recruit the people who are already fast at the things fast games demand. Selection dressed as evidence is precisely why gaming lands among the weak but real correlates of high intelligence rather than among the markers that describe ability directly, such as a large vocabulary or how quickly someone learns. Taste in other media behaves the same way once the confounds are taken out, and the classical music finding turns out to be a preference for instrumental over vocal rather than for a genre.

Boot, Arthur Kramer, Simons, Monica Fabiani and Gabriele Gratton ran the decisive design in Acta Psychologica in 2008. They did two things at once: compared expert gamers against non-gamers, and trained non-gamers for more than 20 hours on an action game, a puzzle game or a real-time strategy game. The expert versus non-gamer comparison reproduced the usual pattern, with experts tracking objects at higher speeds, detecting changes in visual short-term memory better, switching tasks faster and rotating objects mentally more efficiently. The training arm did not. Twenty-plus hours of play produced no substantial gains on most tasks, with mental rotation the one partial exception. The gap between groups was real. Playing did not create it.

Nash Unsworth, Redick, Brittany McMillan, Hambrick, Michael Kane and Randall Engle sharpened the point in Psychological Science in 2015 with a statistical demonstration that explains why this literature keeps producing contradictions. In two experiments they measured working memory, fluid intelligence and attention control alongside gaming experience. An extreme-groups analysis, comparing heavy players against non-players, showed the familiar gamer advantage. Analyzing the full range of participants, at both the task level and the latent construct level, nearly all of the relations were near zero.

That is not a subtle statistical footnote, it is a recipe for manufacturing findings. Comparing the top and bottom slices of a distribution while discarding the middle exaggerates whatever weak relationship exists, and extreme-groups designs are common in this field precisely because they are cheaper to recruit. If you want to understand why one study finds a large gamer advantage and another finds nothing, check which participants each one kept before checking anything else. The same question settles a neighboring controversy, since the headline deficit in the cannabis and IQ literature rests on 23 heavy users in one New Zealand birth cohort, and the twin studies that compared siblings against each other mostly found nothing.

11 Practice Effects: Why Getting Better at a Test Is Not Getting Smarter

Everything above has a direct consequence for anyone who takes an IQ test twice, and it is the single most useful idea on this page. Test scores rise on retest for reasons that have nothing to do with ability. This is called the practice effect, and it is large enough to be measured with precision. Retest gains are one of the reasons a second number is not automatically a better one, a limit discussed alongside the honest accuracy ceiling of online and professional testing.

John Hausknecht, Jane Halpert, Nicole Di Paolo and Meghan Moriarty Gerrard meta-analyzed 50 studies of retesting on cognitive ability measures in the Journal of Applied Psychology in 2007, drawing on 107 samples and 134,436 participants. The adjusted overall effect size was 0.26 standard deviations. Effects were larger when practice came with coaching and larger again when the same form was reused. Matthew Calamia, Kristian Markon and Daniel Tranel found the same pattern across nearly 1,600 effect sizes in The Clinical Neuropsychologist in 2012, with alternate forms, participant age, clinical status and the length of the retest interval all moderating the size of the gain.

Translate 0.26 standard deviations onto the scale where the mean is 100 and the standard deviation is 15 and you get roughly 4 points, from nothing but having sat the test before. That is enough to move someone across a percentile band and produce a genuine feeling of improvement. It reflects familiarity with instructions, remembered items, a settled strategy and lower anxiety.

The expectation half is bigger still. Cyrus Foroughi, Samuel Monfort, Martin Paczynski, Patrick McKnight and Pamela Greenwood recruited two groups in PNAS in 2016 using different flyers. One flyer suggested that the training would boost cognition. Participants who self-selected into that group improved after a single one hour session by an amount the authors describe as equivalent to a 5 to 10 point increase on a standard IQ test. Participants recruited by a neutral flyer showed no improvement. One hour, no real intervention, a shift the size of most published training effects. Any honest reading of a retest score has to subtract both of these before crediting anything to the activity in between. If you want the fuller picture of what these instruments do and do not capture, what IQ measures and how accurate online IQ tests are cover the mechanics.

12 What This Actually Means for You

Strip the debate to its practical residue and a few statements survive intact. Playing action games will make you better at action games, and probably a little better at closely related perceptual tasks while you are playing regularly. Nothing in the evidence supports the idea that a genre choice will move your general reasoning capacity, and the best-controlled analyses put that transfer at or near zero. Games remain a legitimate way to spend time; they are simply not a cognitive intervention. Marketing in this category almost always leans on an older promise, that most of the brain sits unused and waiting to be switched on, yet the ten percent figure behind that promise has never been traced to a credible experiment, and its five, ten and twenty percent variants never say what was being counted. Unsourced percentage claims are the same species of promise that the red flags that disqualify an IQ test warn about, along with missing norm information and silence about limits.

The corollary for test taking is stricter than most people expect. If you take the same battery three times, your score will climb, and the climb tells you about your familiarity with the battery rather than about you. This is why serious assessment uses alternate forms, records the date of administration, reports confidence intervals rather than a single point, and treats a retest inside a short window with suspicion. A score is a measurement of performance on specific tasks on a specific day, not a permanent property being read off a dial.

It is also why a domain profile beats a single number. A composite hides the structure that matters: someone can have strong verbal comprehension and ordinary processing speed, or the reverse, and the two people share a headline score while differing in every practical way. Gaming research runs into exactly this problem, because the abilities games plausibly touch, visual spatial processing and elements of attention, are only part of what a full battery samples. Understanding what counts as a good score and reading it against an IQ percentile chart is more useful than chasing point gains.

There is one more habit worth adopting, and it costs nothing. When you meet a headline claiming that some activity boosts intelligence, ask three questions before reading further: did the study train people or merely compare existing groups, did the control group expect to improve as much as the treatment group, and did the outcome measure resemble the training. Almost every inflated claim in this literature fails at least one of them, and you can usually tell which within a paragraph of the abstract. The three questions generalize well beyond gaming, to supplements, apps, music lessons and any product that promises a general gain from a specific drill.

ACIS is an online self-assessment rather than a clinical evaluation, and it reports results the way this page argues they should be reported: a profile across cognitive domains with intervals around each estimate, not a single figure presented as a fixed trait. If you would rather see your own structure than read about the average of thousands of strangers, that is a thing you can do today.

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.

• Dekker, S., Lee, N.C., Howard-Jones, P. & Jolles, J. (2012). Neuromyths in education. Frontiers in Psychology. Open access. Measured how widely brain myths are believed, even among teachers who work with learning every day.
• Nisbett, R.E. et al. (2012). Intelligence: new findings and theoretical developments. American Psychologist, 67(2). The broad APA review of what moves measured intelligence and what does not.
• 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).