Do music lessons make kids smarter? The studies are in.

Few claims in education have travelled as far on as little as this one. Music lessons, parents are told, do not just teach music. They raise IQ, sharpen focus, lift maths scores — a general upgrade to the developing brain. The claim sells pianos, apps and after-school programs. It also shapes school budgets, because a subject that boosts everything else is easier to defend than a subject that is merely loved.

So it matters that the claim has now been tested properly. Over three decades, the question moved through the standard machinery of evidence: a striking first finding, a wave of enthusiasm, randomized trials, replications, and finally meta-analyses — studies that pool many trials into one estimate. This article walks that record in order, because the order is the story. The short version: the more carefully the question is asked, the smaller the answer gets.

One promise before the evidence. This is not an article against music education. It is an article against a bad argument for music education — and, at the end, an argument for a better one.

The birth of a beautiful claim

The modern claim began with a one-page letter to Nature. In 1993, Frances Rauscher and colleagues had 36 college students listen to ten minutes of a Mozart sonata. Afterwards, the students did better on spatial-reasoning puzzles drawn from an IQ battery. The gain looked striking — the equivalent of roughly 8 to 9 IQ points. It also faded within about a quarter of an hour (Rauscher, Shaw & Ky, 1993).

What happened next had little to do with the paper. The subjects were adults. The task was narrow. The effect was brief. In the retelling, all three details fell away, and the finding became something else: classical music makes children smarter. By 1998, the governor of Georgia was budgeting to send a classical CD home with every newborn in the state. A tightly bounded lab result had become parenting advice — and then policy.

Two claims had fused in public, and they need separating. One says listening to music produces a short cognitive lift — the so-called Mozart effect. The other says music lessons — years of instruction and practice — make children lastingly smarter. The first is easy to test, and it has been tested to destruction. The second is the one schools and parents care about. It took another decade to get its real trial.

The Mozart effect meets its meta-analyses

The listening claim fell first. In 1999, Christopher Chabris pooled the replication attempts — 16 studies by then — and found a small residue: an advantage on the order of 1.4 IQ points, confined to one narrow class of spatial task (Chabris, 1999). His reading was deflationary. Pleasant stimulation briefly lifts arousal and mood, and people in a good mood do slightly better on speeded puzzles. Nothing in the data required Mozart.

A decade later, a larger meta-analysis buried the residue. Pietschnig, Voracek and Formann pooled roughly 40 studies with over 3,000 participants (Pietschnig, Voracek & Formann, 2010). The pooled effect was small against silence and smaller still against other enjoyable stimuli. The literature also showed clear signs of publication bias, and labs affiliated with the original team reported systematically larger effects than independent labs did. Their title gave the verdict its tone: “Mozart effect–Shmozart effect”.

The Mozart effect matters here as a cautionary tale, not a target. It shows how fast a small, fragile lab finding can become policy — and how long the correction takes. The CDs shipped years before the meta-analyses reported. The music-lesson claim would run the same arc, on a slower clock and with more money attached.

The trial everyone cites

The lessons claim got its landmark test in Toronto. Glenn Schellenberg advertised free arts lessons in a newspaper, enrolled 144 six-year-olds, and randomly assigned them to four groups: keyboard lessons, singing lessons, drama lessons, or a year on the waiting list (Schellenberg, 2004). The programs ran about 36 weeks. IQ was measured before and after with a full Wechsler battery — the standard childhood IQ test.

Every group gained IQ, which is normal: schooling and retesting lift scores for everyone. The music groups gained more — roughly 7.0 full-scale points, against roughly 4.3 in the drama and waitlist groups. The difference, about 2.7 points with an effect size near d ≈ 0.35, was statistically reliable and spread across subtests. Drama, for its part, improved rated social behaviour instead (Schellenberg, 2004).

The paper’s title, “Music lessons enhance IQ”, did heavy lifting in the years that followed. The result itself is more modest. Just under three IQ points is real but small — inside the noise a single child can show between two test sittings. No parent would see it across a kitchen table. The study proved the claim could survive one fair test. What it could not do, on its own, is carry an industry.

The author becomes the sceptic

What happened next is rare in psychology: the author spent the following decade stress-testing his own headline. In 2011, Schellenberg examined 106 nine- to twelve-year-olds and confirmed the familiar pattern — musically trained children scored higher on IQ tests (Schellenberg, 2011). Then came the interesting part. Training showed no association with executive function — the attention-control machinery that was supposed to explain the link.

That absence mattered, because executive function was the favoured causal story. If lessons sharpen attention and self-control, and attention and self-control lift everything else, the correlation would have a mechanism. Without it, a duller explanation fits the data better. Children who take music lessons differ before the first lesson: wealthier and more educated parents, more books at home, higher test scores to begin with.

Schellenberg has since become one of the field’s most persistent internal critics, arguing that music-cognition research keeps dressing correlational findings in causal language (Schellenberg, 2011). The man who wrote the famous study is now its most careful caution label. That is not a scandal. It is what science looks like when it is working.

The Harvard trials

The strongest early-childhood test came from Elizabeth Spelke’s lab at Harvard. Samuel Mehr and colleagues ran two randomized trials with four-year-olds, assigning families to six weeks of parent-child music classes, visual-arts classes, or no classes (Mehr, Schachner, Katz & Spelke, 2013). The outcomes were four specific abilities: spatial-navigational reasoning, visual form analysis, numerical discrimination, and receptive vocabulary.

The first experiment produced a teaser. Music children edged ahead on one measure, arts children on another — each difference small, and the pattern exactly what chance produces when many outcomes are tested. So the team did what the field had rarely done: they ran the replication themselves, with a larger sample and a no-treatment control. The teaser did not repeat. Across both trials there was no consistent cognitive advantage from music enrichment in any domain (Mehr, Schachner, Katz & Spelke, 2013).

The trials were short, and the authors said so plainly — six weeks of enrichment classes is not a conservatory education. But the design standard is the point. Tested the way medicine tests treatments — randomized, actively controlled, then replicated — the preschool music-enrichment claim produced nothing a school could bank.

The meta-analytic verdict

Giovanni Sala and Fernand Gobet then did for music lessons what Chabris had done for Mozart. Their 2017 meta-analysis pooled the training studies on children and young adolescents and found an average effect around d ≈ 0.16 — small, and unevenly earned (Sala & Gobet, 2017). The pattern inside the pool was the tell: the better a study’s design, the smaller its effect. Studies with active control groups — children doing some other engaging activity — showed little or nothing.

Their 2020 update tightened the screws with multilevel modelling across dozens of randomized designs and several thousand children (Sala & Gobet, 2020). The overall effect fell to roughly g ≈ 0.06 — for practical purposes, indistinguishable from zero. In studies with both random assignment and an active control, the estimate was approximately 0.00. No moderator rescued it: not age, not the length of training, not the type of outcome.

An effect that shrinks as rigor rises is the signature of an artifact. Real effects survive better instruments; artifacts are produced by weaker ones. On far transfer, the music-training literature now reads the way the Mozart-effect literature read a decade earlier — a large claim upstream, and approximately nothing where the designs are strongest.

Why far transfer keeps failing

The null is not a quirk of music. It is what the wider transfer literature predicts. Skills transfer near — to tasks that share content with the training — and rarely far. Music training makes children measurably better at music and its close neighbours: pitch discrimination, rhythm, reading notation, fine-grained listening (Sala & Gobet, 2017). Those gains are real. They are simply not IQ.

The research program behind these meta-analyses reports the same shape in the neighbouring literatures — chess instruction and working-memory training likewise show solid near transfer, with far-transfer estimates that collapse toward zero in well-controlled trials (Sala & Gobet, 2020). The brain does not treat cognition as one muscle that any demanding exercise strengthens. It builds specific skills out of specific practice.

And the stubborn correlation? It never went away — trained children really do score higher. But experimental and correlational answers diverge here, and when they diverge, selection is the usual culprit (Schellenberg, 2011). Lessons are a marker of advantage, not a cause of it. A school that funds music to raise maths scores has bought the marker while expecting the cause.

What the evidence doesn’t show

A null on far transfer is a narrow verdict, and it should be held narrowly. Here is what this literature does not say.

• Music education has no value. Nothing in these trials measures musicianship, ensemble skill, confidence on a stage, or a life with music in it. The nulls concern side effects, never the thing itself.
• Near transfer is in doubt. Training reliably improves musical and close auditory skills (Sala & Gobet, 2017). The boundary runs between near and far transfer — not between music and nothing.
• The correlation is fake. Musically trained children genuinely score higher on IQ tests (Schellenberg, 2011). What is missing is evidence that the lessons caused the difference.
• Long training has been trialled. The randomized studies run weeks to months (Mehr, Schachner, Katz & Spelke, 2013). A decade of serious practice has never been randomly assigned — and, practically, never will be.
• Every outcome has been tested. Wellbeing, belonging, persistence and school attachment are thinly measured in this literature. That is absence of evidence, in both directions.
• Adult brains are covered. Claims about musicianship and healthy aging come from a different literature with different designs. This article makes no claim about it.

Where the evidence stops

1. 1Music education has no value
2. 2Near transfer is in doubt
3. 3The correlation is fake
4. 4Long training has been trialled
5. 5Every outcome has been tested
6. 6Adult brains are covered

© 2026 FUTURE PROOF™

The case for music, without the myth

Strip the myth away and the practical case for school music gets easier to make, not harder. Music is a domain of human achievement — a literacy, a performance discipline, a shared inheritance. Schools teach it for the same reason they teach painting and poetry. None of those subjects is asked to raise maths scores to keep its room.

The advocacy lesson is about risk. A program sold on IQ gains is mortgaged to a claim the evidence no longer supports — and when the correction arrives, the funding falls with the claim. A program sold on musical outcomes — children who can sing, play, read and hear more than they could — stands on ground no future meta-analysis can pull away.

For school leaders and ministries, the operating rules are short. Fund music as music, and measure musical outcomes: repertoire, notation, ensemble performance, listening skill. Treat any vendor promising cognitive side effects as making a marketing claim, and ask for the randomized trial. Protect instrument access for poorer children — the correlational record shows who currently gets lessons, and it is not everyone.

And when a budget debate forces the question — what does music do for test scores? — the honest answer is now available: roughly nothing. That was never the reason to teach it. The reason walks out of the concert hall humming.