Search for board games and the brain and you will meet the same two numbers over and over. Board games cut your risk of dementia by seventy-four percent. Or by fifteen percent. One hour of playing a number game transforms a preschooler’s grasp of numbers.
All three come from real studies published in real journals. None of them says what it is quoted as saying.
This page is the long version of that sentence. It is not an argument that board games are bad for you, and it is not an argument that scientists got it wrong. It is an account of what happens to these findings when you read past the summary at the top of the paper, and of what is left standing afterwards, which turns out to be more interesting than the headline.
Where a study uses a technical term, we explain it the first time it appears. You do not need a statistics background to follow any of this, and you should not have to take our word for any of it either. Every study is linked at the bottom.
No affiliate links appear on this page. It exists to be checked, not to sell you anything.
The Seventy-Four Percent
The source is a 2003 paper in the New England Journal of Medicine by Joe Verghese and colleagues. It was what researchers call a cohort study: you recruit a large group of people, write down what they do with their time, and then follow them for years to see who develops the condition you are interested in. In this case, 469 adults over the age of seventy-five living in the Bronx, followed for a median of five years.
Among the leisure activities they tracked, playing board games frequently came out with a hazard ratio of 0.26.
A hazard ratio is a comparison of risk between two groups over time. A hazard ratio of 1.00 means the two groups fared identically. Below 1.00 means the first group did better; above means worse. So 0.26 means the frequent players developed dementia at roughly a quarter of the rate of the people who rarely played. Subtract 0.26 from 1.00 and you get 0.74, and that is where the seventy-four percent in the headlines comes from.
The arithmetic is fine. Four things about the number rarely survive the trip to a blog post.
The item was not board games alone
The results table in the paper lists the activity as “playing board games.” But the Methods section, the part that describes what the researchers actually asked people, defines the activity that was assessed as “playing board games or cards.” The paper never separates the two, and never reports how much of that 0.26 came from board games and how much from a weekly game of cards. In a group of New Yorkers in their late seventies recruited in the early 1980s, that is not a small ambiguity.
The group was small
Of the 469 people in the study, only 103 were in the frequent-play group. Everything in that hazard ratio rests on those 103 people. Small groups produce unstable numbers, in both directions.
Seventeen activities were tested at once
The researchers looked at seventeen separate leisure activities, and the paper does not report correcting for that.
Here is why that matters, without any mathematics. Imagine flipping a coin twenty times and recording the result. Do that once, and eight heads out of twenty is unremarkable. But run twenty separate sets of twenty flips and one of them will probably throw up something that looks striking, say fifteen heads, purely by chance. If you then report only that one set and ignore the other nineteen, you have made luck look like a finding. Statisticians correct for this by raising the bar when many things are tested together. That correction is not reported here.
You can see the consequence inside the same paper. Crossword puzzles came out at a hazard ratio of 0.59, which sounds like a substantial protective effect, with a confidence interval of 0.34 to 1.01.
A confidence interval is the study’s own statement of how much its number might be off. It is the range the true value plausibly sits in, given the data. And when that range includes 1.00, which is the number that means “no difference at all”, the honest reading is that the study cannot rule out that there was no effect. That is what researchers mean when they call a result not statistically significant. It does not mean the effect is zero. It means the data are too weak to say.
The authors ran the obvious check, and it did not fully hold
To their credit, Verghese and colleagues tested the most worrying alternative explanation themselves: that people already sliding into undiagnosed dementia were the ones who had stopped playing. If that were happening, the games would look protective without doing anything.
The way to test it is to throw out the people diagnosed early in the study, on the logic that those are the ones most likely to have been already ill at the start, and see whether the association survives.
Excluding the 94 people diagnosed in the first seven years, the overall cognitive activity score held up: a hazard ratio of 0.94, still statistically significant. Excluding the 105 diagnosed within nine years, it did not: the hazard ratio moved to 0.96, with a confidence interval running from 0.89 to 1.04. That range includes 1.00.
In fairness to the study, only nineteen dementia cases were left by that point. With nineteen cases you cannot detect much of anything, so that test was very weak. It does not show the effect was an illusion. It does mean the data cannot rule that out.
The authors themselves were direct about the limits of the whole thing: “despite the magnitude and consistency of the associations, our findings do not establish a causal relation between participation in leisure activities and dementia, and controlled trials are therefore needed.”
The Fifteen Percent, and What Happened To It
The second number comes from a 2013 study in BMJ Open led by Jean-François Dartigues, and it is a considerably better study: 3,675 French adults over sixty-five, followed for twenty years, during which 840 of them developed dementia.
The researchers reported a hazard ratio of 0.85 for regular board game players, a fifteen percent lower rate of dementia, after adjusting for age, sex, education, marital status, diabetes, stroke and sensory impairment.
Adjusting means using statistics to strip out the influence of things that could produce a fake association. Better-educated people are both more likely to play games and less likely to be diagnosed with dementia, for instance, so unless you account for education you might simply be measuring education twice. Adjustment is standard and good practice. The question is always which things you adjust for.
Then the authors did something most write-ups of this study skip entirely. They ran the model again with two more adjustments: the participant’s cognitive test score at the start, and depression.
The hazard ratio moved to 0.96, with a confidence interval from 0.82 to 1.12, and a p-value of 0.61.
A p-value is a rough measure of how easily plain chance could have produced a result this large, if there were really nothing going on. The smaller it is, the harder the result is to explain away as luck. By long-standing convention, anything above 0.05 is treated as “chance could comfortably account for this.” A p-value of 0.61 is not near the line. It is not in the same postcode.
So the famous fifteen percent is not a smaller effect after the fuller adjustment. It is an association the study can no longer distinguish from nothing at all. That does not prove the true effect is exactly zero, since no study can prove that, but this study, in its own more complete model, provides no evidence of one.
The paper says as much. The authors write that they “cannot exclude that an unmeasured cognitive decline before baseline could precede the discontinuation of board game playing,” that “the relationship could be bidirectional,” and that their own supplementary analysis “is more in favour of a reverse causation from outcome to exposure.”
Two more details worth carrying with you. The measure of how much people played was a single question, asked once at the very start, and never updated again across twenty years. So somebody who played weekly in 1990 and stopped in 1995 is still counted as a player in 2010. And the question counted cards, bingo, chess and draughts alongside board games.
Why the Association Keeps Showing Up Anyway
None of this means the pattern is imaginary. Dozens of studies do find that people who play games in later life go on to develop dementia less often. That association is real and it is consistent.
The question is which way the arrow points. Does playing protect the brain, or does an already-changing brain stop playing?
Researchers call the second possibility reverse causation, where the thing you assumed was the cause is actually the effect. It is the single hardest problem in this entire field, and two large studies get closer to settling it than anything else available.
Whitehall II: the association depends on when you measure
Whitehall II is a long-running British study that has followed 8,280 civil servants for decades. Researchers used it to ask a clever question: does it matter when in someone’s life you record their leisure activities?
It matters enormously. Leisure activity measured when participants were around fifty-six showed no significant association with dementia later on. The same measure taken at sixty-six showed a hazard ratio of 0.82, an apparent protective effect.
Think about what that means. If games protected the brain, the measurement taken decades before diagnosis should be the informative one, because that is when the protecting would be happening. Instead the signal only appears when you measure close to the diagnosis. The researchers also found that a drop in someone’s activity between one survey and the next predicted dementia afterwards, which is what you would expect if withdrawal from activities is an early symptom rather than a cause.
33,263 people: the effect fades the longer you wait
A 2024 study in the journal Age and Ageing pooled five separate groups of participants, 33,263 people in total, and sorted the results by how long researchers had waited between recording activity and counting dementia cases.
With less than ten years of follow-up, the numbers looked spectacular: mentally stimulating leisure activity came out at a hazard ratio of 0.52, which is to say roughly half the risk.
With ten years or more, in the authors’ own phrase, “all associations attenuated toward the null”. In plain words, the apparent benefit shrank away toward nothing.
A genuine protective effect should get clearer with a longer follow-up, not weaker. An effect that only exists in the years right before diagnosis is exactly the shape you would expect from reverse causation.
The reading that fits both studies is uncomfortable but simple. Dementia begins altering the brain years before anyone notices, and one of the earliest things to go is the appetite for complicated, sociable, mentally demanding activities. People in the earliest stage of the disease quietly stop playing. Measure a few years later, and it looks as though the people who kept playing were protected.
The strongest counterargument
This deserves airing, because it cuts against everything above. A 2025 study in the same journal deliberately built in a seven-year gap between measuring activity and counting cases, precisely to sidestep the reverse causation problem — and still found hazard ratios around 0.50.
That is a serious result and it is not easily dismissed. But note which activities were carrying it: crosswords and artistic activities showed the strongest independent associations. Board games did not.
What About Children?
The claim that a short course of number board games transforms a preschooler’s grasp of numbers traces to a series of studies by Geetha Ramani and Robert Siegler, starting in 2008.
The game itself is simple and rather charming. Children play on a linear board with squares numbered one to ten, spin a spinner, and move their piece, counting the numbers out loud as they go rather than just counting squares. Four sessions of fifteen to twenty minutes, about an hour of play in total.
The reported effects were enormous. The papers use a measure called Cohen’s d, which expresses the size of a difference between two groups in a standardised way so that results from different tests can be compared. The rough convention in education research is that 0.2 is a small effect, 0.5 is medium, and 0.8 is large. The original studies reported numbers above 1.0.
Here is what happened to those numbers as the studies got bigger and the comparisons got fairer.
| Study | Children | Effect |
|---|---|---|
| Siegler & Ramani, 2008 | 36 | d = 1.62 |
| Ramani & Siegler, 2008 | 124 | d = 1.08, falling to 0.55 after nine weeks |
| Siegler & Ramani, 2009 (tighter comparison) | 85 | d = 0.65 to 0.71 |
| Hawes et al., 2019 (independent team) | 43 | No effect on four of five measures |
| James-Brabham et al., 2024 | 249 | No significant effect |
| Nelson et al., 2025 (18 studies pooled) | — | g = 0.21 |
Two things explain that slide.
The test looked like the game
During the intervention, children spent their time counting aloud from one to ten and naming the numerals one to ten. They were then tested on counting and naming the numerals one to ten. One of the headline measures asked children to place numbers on a line running from zero to ten — the same layout as the board they had just been playing on.
Children got better at the exact thing they had been practising, measured on a test that looked like the thing they had been practising. That is a real result, and it is also the least surprising result available.
The comparison group changed everything
Every study like this needs a comparison group, and there are two kinds. A passive control group carries on with its normal school day while the other group gets a novel activity with an enthusiastic adult. An active control group gets an equally novel, equally enjoyable activity with the same enthusiastic adult — just a different one.
The difference between those two designs is the difference between measuring the game and measuring the attention.
The 2024 trial by James-Brabham and colleagues is the largest and cleanest test anyone has run: 249 children, and a control group that played an alphabet game. Same adult, same small-group time, same novelty, same structure of play. Only the content differed. Eight sessions over five weeks, twice the dose of the original studies.
Both groups improved. The difference between them was not statistically significant.
The Pattern Behind All of It
Once you have seen it, you cannot unsee it: the size of the reported effect tracks what the comparison group was doing. Passive control, big effect. Active control, small or no effect.
This is not a board game problem. It is the central finding of thirty years of research into cognitive training in general, and it has a name: the difference between near transfer and far transfer.
Near transfer means getting better at the thing you practised, and at things closely resembling it. Practise counting, get better at counting. That is real, well established and not in dispute.
Far transfer means the practice makes you generally sharper: better at unrelated tasks, better at everyday thinking. That is what everybody actually wants when they buy a brain game, and it is the part that keeps failing to hold up.
In 2019 Giovanni Sala, Fernand Gobet and colleagues published what is called a second-order meta-analysis. A meta-analysis pools the results of many separate studies to get a more reliable average than any single study gives. A second-order one pools the meta-analyses. They combined ten of them: 332 separate results, 21,968 participants, covering memory training, video games, music lessons, chess and exercise games.
They then did two things to the raw numbers. They corrected for publication bias, the well-documented tendency for studies that find an effect to get published while studies that find nothing sit in a drawer, which inflates every average in the literature. And they looked separately at the studies that used active control groups.
| Raw | Corrected for publication bias | With active control groups | |
|---|---|---|---|
| Near transfer | g = 0.30 | 0.21 | 0.19 |
| Far transfer | g = 0.12 | 0.04 | 0.00 |
Hedges’ g in that table is the same kind of measure as Cohen’s d: the size of a difference, expressed so results from different studies can be compared. Read the bottom right cell again. Once you account for the drawer full of unpublished null results and compare against a group doing something equally engaging, the general-improvement effect across all of cognitive training is zero.
Their conclusion: “The lack of generalization of skills acquired by training is thus an invariant of human cognition.”
Now notice where the board game meta-analysis landed. Pooled across eighteen studies, number board games came out at g = 0.21 — small, but statistically significant, and genuinely not zero. That figure is almost exactly the corrected near-transfer estimate for cognitive training as a whole. Board games are not an exception to the pattern. They are a clean example of it: a real but modest gain on the skills the game rehearses, and nothing reliably detectable beyond them.
A 2026 review put the same question to executive function, the umbrella term for planning, resisting impulses, holding several things in mind and switching between tasks, across all age groups. It did find some positive results, including visual-spatial short-term memory in kindergarten children and processing speed in older adults. Its own conclusion was that “the current empirical evidence is not yet sufficient to support a strong recommendation for the use of board games to improve executive functions.” That review came out of a laboratory that has itself published several of the positive trials in this field, which makes the verdict harder to wave away.
For a sense of how much practice it takes to produce nothing at all, consider the same researchers’ work on chess. They gave children twenty-five hours of chess instruction and then tested their mathematical problem-solving against a comparison group that spent the same time playing draughts. The statistical test for a difference between the groups returned a p-value of 0.679, meaning chance alone would easily produce a gap that size. Twenty-five hours. No board game study in this article comes anywhere near that amount of practice.
Why This Matters When Someone Is Selling You Something
In January 2016 the United States Federal Trade Commission, the agency that polices deceptive advertising, settled a case against Lumos Labs, makers of the brain-training app Lumosity. The company paid two million dollars in refunds to customers, against a fifty-million-dollar judgment that was suspended because the company could not pay it.
The claims the Commission deemed deceptive included that the games improved performance at school, at work and in daily tasks, and that they delayed age-related cognitive decline and protected against mild cognitive impairment, dementia and Alzheimer’s disease.
Jessica Rich, then the director of the Commission’s Bureau of Consumer Protection, put it in one sentence: “Lumosity simply did not have the science to back up its ads.”
That same year, a major review in the journal Psychological Science in the Public Interest went through the entire brain-training literature against a defined checklist of what a trustworthy study should do. It found plenty of evidence that training improves performance on the trained task, less on closely related tasks, and “little evidence that training enhances performance on distantly related tasks or that training improves everyday cognitive performance.” Not one study it examined met the full checklist.
What Actually Holds Up
Here is where the story turns, and it is the part almost nobody writes about.
The research group that published the most favourable review of board games and cognition in 2023 ran their own randomised trial and published an updated meta-analysis alongside it in 2025. Their finding: a significant effect on depression, at a standardised mean difference of −0.48. On the same scale as the effect sizes above that is a moderate effect, and the minus sign means depressive symptoms went down. But there was no significant effect on cognition.
That is the enthusiasts’ own current position. Mood, not memory.
And it fits everything else. In the twenty-year French study, the dementia result dissolved under adjustment, but the effect on depression survived it, at a hazard ratio of 0.84. Randomised trials in care homes report gains in quality of life, emotional wellbeing and social inclusion far more consistently than they report gains on cognitive tests. And in the one properly blinded trial that pitted modern board games against matched pencil-and-paper exercises in older adults, the games did not outperform the worksheets on anything.
The study that changes how you should read all of this
A 2019 randomised trial in Japan taught the game Go to 72 older adults, split into three groups.
The first group attended classes and played face to face with other people. The second group ran the same programme, alone, on a tablet. The third group attended health lectures instead. Twelve weekly sessions each.
Both Go groups improved on a memory measure. But only the face-to-face group significantly outperformed the comparison group. The tablet group was indistinguishable from it. The face-to-face group was also far more likely to still be playing six months later.
Same game. Same amount of it. The one thing that differed was whether there were other people in the room.
One trial with 72 people does not settle anything on its own, and moving an activity from a tablet into a classroom changes more than just the company. The structure, the encouragement and the obligation to turn up all change too. But it is a rare design that isolates the social ingredient at all, and it points somewhere worth noticing.
Escape Rooms, Specifically
Escape rooms have a research literature of their own, mostly from nursing and medical education, where instructors build them to teach a syllabus. It shows the same split as everything above, with unusual clarity.
A 2026 review of twenty such studies sorted them using the Kirkpatrick model, a standard framework for judging how deeply any training programme has actually been evaluated. It has four levels, in ascending order of how much they tell you: did participants enjoy it, did they learn anything, did their behaviour change afterwards, and did it improve real-world outcomes.
Here is how the twenty studies distributed across those four levels. All twenty measured enjoyment. Nine measured learning. One measured whether behaviour changed, and it found no change. None measured real-world outcomes at all.
The meta-analyses do find genuine knowledge gains, with three independent teams landing around g = 0.86, which is a large effect. It would be dishonest to wave that away. But look at what the comparison usually is. In most of these studies the escape room is an extra revision session, measured against students who got no extra session. One review found ninety percent of the activities were used for reviewing material students had already been taught. Compare an extra hour of structured practice against no extra hour, and the extra hour wins. That is a finding about practice, not about escape rooms.
Where the comparison group gets a genuine alternative, the gap closes. A 2025 study put 134 nursing students through either an escape room or a conventional assessment, and measured two things: how much they enjoyed it, and how much they still knew a month later. Enjoyment was much higher for the escape room. Knowledge retention showed no statistically significant difference between the two groups.
Teamwork is the benefit escape rooms are sold on more than any other, and we could not find a single study demonstrating it on an objective measure against a comparison group. One 2024 meta-analysis abandoned its teamwork analysis entirely for lack of well-designed studies to pool. In one systematic review, twenty-one of thirty-nine studies listed teamwork as a goal, and nineteen of those assessed it by informal observation or by asking participants how they felt afterwards. That review’s own conclusion is deflating: teamwork is a requirement for finishing an escape room in time, not a product of having finished one.
And the boxed games on the shelf
This is the part that matters most if you are standing in a shop. We could not find a single peer-reviewed study testing the cognitive, educational or social effects of a boxed escape room game. Not EXIT: The Game, not Unlock!, not Deckscape. The only academic paper that engages with them seriously is a classification of puzzle types, which measures no outcomes whatsoever.
Claims that those boxes build teamwork or teach problem-solving are extrapolated from studies of purpose-built educational escape rooms, run by trained instructors, in classrooms, with a structured discussion afterwards in which students are walked through what they just did and why.
Four people opening a cardboard box at a kitchen table is a different activity. There is no instructor, no learning objective, no discussion afterwards and no assessment. The research may well transfer. Nobody has checked.
So What Should You Do With All This?
Buy the game because you want to spend an evening with people you like. That is the claim the evidence actually supports, and it is not a small one.
What the research keeps pointing at is not the cardboard. It is the table. Sitting down with other people, regularly, doing something that holds your attention and gives you pleasure, is associated with better mood and better quality of life. And the one trial that varied whether the game was played alongside other people found that only the version with company beat the comparison group.
Board games happen to be an excellent excuse for exactly that. They give people a reason to sit down together for two hours without a screen, and a structure that carries the conversation when nobody has anything in particular to say. That is worth the price of a box.
It is just not neuroscience. If someone is selling you a game on the promise that it will protect your memory, rescue your child’s mathematics or keep dementia away, they are making a claim that the Federal Trade Commission has already fined a company two million dollars for making.
Choose a game the way you would choose a restaurant to take friends to. That is the honest version, and it is a better reason anyway.
If you came here while deciding what to actually buy, our top ten entry level games is the place to start, and the best escape room games to start with covers the boxed escape games discussed above.
Every Study Mentioned Here
- Verghese J, et al. Leisure activities and the risk of dementia in the elderly. New England Journal of Medicine, 2003;348(25):2508–16. The source of the seventy-four percent figure.
- Dartigues JF, et al. Playing board games, cognitive decline and dementia: a French population-based cohort study. BMJ Open, 2013;3(8):e002998. The source of the fifteen percent figure, and of its disappearance.
- Sommerlad A, et al. Leisure activity participation and risk of dementia: An 18-year follow-up of the Whitehall II Study. Neurology, 2020;95(21):e2803–15. Activity at fifty-six versus at sixty-six.
- Heikkilä K, et al. Stimulating leisure-time activities and the risk of dementia: a multi-cohort study. Age and Ageing, 2024;53(7):afae141. The 33,263-person study where the effect fades past ten years.
- Ramani GB, Siegler RS. Promoting broad and stable improvements in low-income children’s numerical knowledge through playing number board games. Child Development, 2008;79(2):375–94. The original number board game study.
- James-Brabham E, Jay T, Sella F. No evidence that playing a linear number board game improves numerical skills beyond teaching as usual. Journal of Educational Psychology, 2024;116(3):411–25. 249 children, alphabet game control, free to read in full.
- Sala G, et al. Near and Far Transfer in Cognitive Training: A Second-Order Meta-Analysis. Collabra: Psychology, 2019;5(1):18. The 21,968-participant analysis where far transfer reaches zero. Free to read.
- Sala G, Gobet F. Does chess instruction improve mathematical problem-solving ability? Learning & Behavior, 2017;45(4):414–21. Twenty-five hours of chess.
- Estrada-Plana V, et al. Cognitive training with modern board and card games in healthy older adults: two randomized controlled trials. International Journal of Geriatric Psychiatry, 2021;36(6):839–50. Board games against pencil and paper.
- Estrada-Plana V, et al. Can executive functions be improved by playing board games across the lifespan? A systematic review and meta-analysis. Applied Psychology: Health and Well-Being, 2026;18(4).
- Iizuka A, et al. Does social interaction influence the effect of cognitive intervention program? A randomized controlled trial using Go game. International Journal of Geriatric Psychiatry, 2019;34(2):324–32. Face to face versus the same game on a tablet.
- Simons DJ, et al. Do “Brain-Training” Programs Work? Psychological Science in the Public Interest, 2016;17(3):103–86.
- Federal Trade Commission. Lumosity to Pay $2 Million to Settle FTC Deceptive Advertising Charges for Its “Brain Training” Program. January 2016.
- Sturm H, Brown G, Gantwerker EA. Evaluating Medical Education Escape Rooms: A Scoping Review Using the Kirkpatrick Model. MedEdPublish, 2026. Twenty studies, one behaviour measure, no real-world outcomes.
- Veldkamp A, et al. Escape education: A systematic review on escape rooms in education. Educational Research Review, 2020;31:100364. Twenty-one of thirty-nine studies claimed teamwork.
- Eriksson M, et al. The behavioral effects of cooperative and competitive board games in preschoolers. Scandinavian Journal of Psychology, 2021;62:355–64.
Last reviewed: September 2026. This page is updated when new evidence warrants it rather than rewritten for freshness.
