Do memory games make children smarter? What the research actually says
We sell a learning product with memory games in it, so take this as evidence against our own interest: training a child's memory with abstract puzzles does not make them better at reading or arithmetic. The research on this is unusually large, unusually consistent, and unusually negative.
What was tested
Working-memory training — Cogmed, n-back tasks, dual n-back, the "brain training" category generally — has been studied for about twenty years. The question is never whether children get better at the training task. They do. The question is whether that improvement reaches anything else: reading, maths, attention, intelligence.
That is called far transfer, and here is what the meta-analyses found.
| Study | Scope | Near transfer | Far transfer |
|---|---|---|---|
| Melby-Lervåg & Hulme, 2013 | 23 studies | Yes, short-lived | None convincing to decoding, arithmetic or reasoning |
| Melby-Lervåg, Redick & Hulme, 2016 | 87 publications, 145 comparisons | Reliable | Effectively zero across verbal ability, decoding, comprehension and arithmetic |
| Simons et al., 2016 | Full review of the commercial literature | Yes, on the trained task | Little evidence for distantly related tasks or everyday performance |
| Aksayli, Sala & Gobet, 2019 (Cogmed) | 50 studies, 637 effect sizes | Small to medium on memory tasks | "Highly consistent near-zero effects" on attention, intelligence, language and maths |
| Sala & Gobet, 2020 | 41 studies, 2,375 children | g = 0.39 immediately | g = 0.001 against active controls; 0.006 at follow-up |
The pattern across all of them is the signature of a null effect: the better the study design, the smaller the transfer, until it disappears. When researchers compare against an active control — children doing something else equally engaging — the benefit goes to zero.
Why it does not work
The appealing idea is that working memory is a muscle, and a bigger one helps everywhere. The evidence suggests it is closer to a skill: you get better at the specific thing you practise, including the specific way it is presented. A child who gets very good at remembering sequences of coloured squares has got very good at remembering sequences of coloured squares.
This is also a consumer-protection matter, not only an academic one. In 2016 the US Federal Trade Commission fined Lumos Labs, the maker of Lumosity, two million dollars for claiming its games improved performance at school and work. In 2014 around seventy scientists signed a consensus statement from the Stanford Center on Longevity and the Max Planck Institute objecting to exactly those claims. If a product tells you its games will improve your child's memory, focus or IQ, it is making a claim a regulator has already acted on.
What does transfer
Three things, with large and repeatedly replicated effects, all of them specific to the material being learned:
- Retrieval practice. Being asked to produce the answer beats being shown it, by a wide margin.
- Spacing. The same total practice spread over days beats it massed into one session.
- Interleaving. Mixing problem types beats blocking them, even though it feels worse while you do it.
None of these grow a child's memory. They make what the child learns stick.
So why build memory games at all?
Because the design can be inverted. Rather than train memory and hope the gain reaches the curriculum, use memory load as the difficulty setting on curriculum practice.
Holding something in mind while you retrieve it makes the retrieval effortful, and effortful retrieval is better remembered — for that material. So a child who plays a matching game where the pair is a letter and a picture that starts with its sound is practising letter-sound correspondence under load. They get better at letter-sound correspondence. Whatever happens to their memory is a side effect we do not measure and will not advertise.
That is the rule every Ladderbug game follows: every card, tile and path cell carries a real item from the curriculum. None of them generate abstract stimuli. The memory demand sets the difficulty; the content is what is being learned.
What to look for in any product that claims memory benefits
- Does it claim to improve memory, focus, attention or IQ? That claim is not supported. Treat it as marketing.
- Do the games use abstract stimuli — shapes, colours, positions — or curriculum content? Abstract stimuli train the game.
- Is there a timer? Time pressure plus memory load produces anxiety, and anxiety reduces available working memory. It makes the thing worse while looking like effort.
- Is there a visible "memory level"? If there is, parents will read it as a memory score, and the product is back to making the claim it should not make.
One honest carve-out
Working-memory training shows somewhat larger and more durable near effects for children with identified working-memory deficits, and that is a real finding worth knowing. Separately, and much better evidenced: reducing memory load during instruction helps everyone. Shorter instructions, one step at a time, the information left on screen rather than held in mind. Lowering the load is well supported. Raising the capacity is not.
Play games that teach something
Ladderbug has eight. Every one of them practises phonics, number facts, letter formation or listening comprehension while loading memory, and none of them show a timer, a score or a rank.
Sources
Melby-Lervåg & Hulme (2013), Developmental Psychology. Melby-Lervåg, Redick & Hulme (2016), Perspectives on Psychological Science. Simons, Boot, Charness, Gathercole, Chabris, Hambrick & Stine-Morrow (2016), Psychological Science in the Public Interest. Aksayli, Sala & Gobet (2019), Educational Research Review. Sala & Gobet (2020), Psychonomic Bulletin & Review. Gobet & Sala (2023), Perspectives on Psychological Science. FTC v. Lumos Labs (2016).
Written by the Ladderbug team. We build a free reading and math site for ages 4 to 11, so we have an interest here — where that interest might bend an answer, we have tried to say so on the page rather than hide it.