Does teaching music produce musical skill?

Verdict

Moderate support, and the modesty of that grade is the finding. Teaching music produces musical skill at about the effect size a competent instructional programme produces anywhere — d ≈ 0.5 on sight-reading, performance and aural skill. That is the honest justification for music in a school, and it is a good one.

What is startling is how little effort has gone into establishing it. The field has produced at least six meta-analyses asking whether music training raises maths, reading, IQ or executive function. It has produced one asking whether music teaching raises music attainment, published in 2025, and that one had to screen 93,291 records to find 20 studies reporting an effect size against a comparison group. There is no randomised trial of instrumental instruction with an independent measure of musical skill. The subject has spent fifty years litigating its spillovers and left its core claim to inference.

What the evidence shows

The one thing that is well established is that instruction beats no instruction on the taught skill, and the singing literature is the cleanest version of it: both treated and control children improved on pitch-matching from pretest to posttest, and the treated children improved more. That is what an instructional effect looks like.

Feedback is the lever, not the method. Across the singing studies, feedback — from self, teacher or computer — produced the largest effect of any teaching condition. Meanwhile the branded specialist methodologies that dominate music-education marketing (Kodály, Orff and relatives) produced lower effects than generalist teaching in the only meta-analysis that has compared them, and technology added a negligible amount. This is the same pattern the archive finds in feedback and in motor feedback: the general principle travels, the branded package does not.

Practice explains more of musical attainment than of anything else. Platz's corrected r꜀ = 0.61 is the top of the range across every domain in the deliberate-practice literature — music is the domain where practice matters most because it is highly structured and highly predictable (see deliberate practice). It still leaves the majority of variance outside practice hours, and the correlation is cross-sectional throughout.

Hereditarian-lens assessment

Risk: medium, and it bites on one specific sub-claim rather than on the topic as a whole.

The instruction meta-analyses compare taught to less-taught within samples, so heredity is not the main threat to them. The threat is to the inference most people draw next, which is that practice hours cause musical ability. Mosing's monozygotic co-twin control is the direct test, and it fails: holding genotype and rearing constant, the twin who practised up to 20,000 hours more was not better at discriminating rhythm, pitch or melody. Practice itself is 41–69% heritable, and the entire practice–ability correlation decomposed as genetic (r_A .33–.57) with the non-shared environmental component at .00–.06, all non-significant.

Gustavson's twin-and-adoption design puts a number on the choice: instrument engagement at age 12 is 78% heritable with no detectable shared environment. Whether a child picks up an instrument is about as heritable as height-adjacent traits, and adoptive siblings raised in the same musical household do not converge.

Two scoping corrections the evidence forces:

1. Mosing's null is about music discrimination, not about playing. The outcome was an auditory discrimination test; the authors explicitly concede that instrument-specific motor and notation skills may still respond to practice. Do not read it as "practice does not improve playing."
2. Heritable engagement is compatible with effective teaching. Per the premise, good instruction can raise the mean and increase heritability by removing environmental bottlenecks. A school that teaches every child music converts "who was offered lessons" into "who has the aptitude", which is evidence the teaching worked, not that it failed.

Boundaries & what critics say

• No randomised trial anchors this verdict. The singing meta-analysis restricted itself to pretest-posttest and posttest-only quasi-experiments by design; the music-teaching meta-analysis does not report its randomised share. Confidence is capped at medium for that reason alone.
• The music-teaching meta-analysis is a single 2025 synthesis and the specialist-versus-generalist finding inside it is an unreplicated moderator. Do not restructure a programme around it.
• d = 0.56 is measured mostly on instruments built by music educators, and measure type is not reported at the level the archive's effect-size rules want. Treat it as a plausible central estimate, not a precise one.
• Delayed retention is unmeasured, exactly as in the motor skill literature. Every effect here is end-of-instruction.
• The "start before seven" rule does not survive a genetic control. Wesseldijk 2021 tested it in 310 professional musicians plus 7,786 twins: once lifetime practice hours are held constant, early onset no longer predicts musical achievement (p > .14), and the residual association with aptitude was fully explained by familial factors. Age of onset is itself ~39% heritable. Their conclusion is that the data "provide little support for a sensitive period for music." Cross-sectional work finds early-onset advantages; the genetically informative work says those are accumulated hours and family.

Practical guidance

• Teach music on the musical case. Half a standard deviation on sight-reading and performance is a real instructional return, and it is the only claim about music you will not have to retract.
• Buy feedback, not a method. Recorded playback, teacher correction and computer pitch feedback are where the singing literature's effects concentrate; branded specialist systems underperform generalist teaching.
• Start singing early — 5 to 8 is where the singing-accuracy evidence is strongest, and singing requires no instrument budget.
• Do not screen children for musical aptitude. Engagement is 78% heritable and ability is substantially heritable, but heritability describes population variance and licenses no prediction about an individual child — the same conclusion the archive reaches for athletic talent.
• Do not let the transfer argument do the work. See music and cognitive transfer; staking the programme on spillovers is how music programmes get cut when the spillovers fail to appear.

Open questions

• No randomised trial of instrumental instruction with an independent measure of musical skill exists. It would be cheap and it would settle the base claim.
• Whether taught musical skill is retained after instruction stops is essentially unmeasured.
• Whether there is a genuine sensitive period for instrumental learning, as opposed to a cumulative-hours advantage for early starters, is untested with an adequate design.
• Whether whole-class classroom music produces the same skill gains as small-group or individual tuition is not separable from the current evidence.