Does Music Training Enhance Literacy Skills? A Meta-Analysis
Gordon RL, Fehd HM, McCandliss BD · 2015
grade Cmeta-analysisindependentmixed
Sample
12 papers yielding 13 independent samples, n = 901; 18 phonological-awareness outcomes and 5 reading-fluency outcomes
Population
Children with weighted mean age 6.25 years (sample means 4.53-9.33). Native languages English, Portuguese, German, French, Spanish and Tamazight. Music training totalled 3-90 hours.
Design
Narrow and careful: inclusion required a music versus control comparison, pre- and post-measures, AND reading instruction held constant across arms - which rules out the commonest way music-and-reading studies fake a result. Grade C because, as the authors state, most included studies were quasi-experimental without random assignment and only 6 of 12 reported controlling for socio-economic status. The pooled sample is small enough that the authors declined to test most moderators.
Key findings
Music training moved phonological awareness by d = 0.20 [0.04, 0.36], p = .01, across 18 outcomes - a real but small effect on the skill that sits closest to music (both are auditory segmentation). It did NOT move reading fluency: d = 0.16 [-0.03, 0.35], p = .10, k = 5, I2 = 0. The rhyming subcomponent alone was null (0.18 [-0.06, 0.42], p = .14). Training hours moderated rhyming (p = .04, roughly +0.01 per hour), with the model implying at least 40 hours before a significant rhyming effect appears - the only dose-response signal in the music cluster. Control type was NOT a significant moderator here (p = .34-.75), which is the one place this meta-analysis diverges from Sala & Gobet, though with far fewer studies. No publication bias detected. The authors are notably unwilling to overclaim: "the findings of the current meta-analyses are somewhat inconclusive with regards to the hypothesized impact of music education on reading-related skills."
Genetic confound
Unmanaged in most included studies - only half controlled for SES and most did not randomise. The authors' own counterpoint is that the three studies with random assignment plus matched SES and IQ all showed large effects on at least one reading-related outcome, which is the strongest pro-transfer argument inside this paper.
Replication notes
The phonological-awareness effect is roughly consistent with Sala & Gobet (2017), who report d = 0.17 [0.04, 0.29] on the same construct across 32 effect sizes; the two syntheses agree on this cell even though they disagree everywhere else. Both are contradicted by Román-Caballero et al. (2022), who report phonological processing at -0.10 for instrumental training specifically (single study). The reading-fluency null is consistent across every synthesis.
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Phonological awareness (all measures) | Cohen d | 0.20 (95% CI 0.04-0.36), p = .01, k = 18 outcomes; Q(17) = 28.8, p = .04 | standardized | post-training | unclear | end-of-treatment | near-transfer |
| Reading fluency | Cohen d | 0.16 (95% CI -0.03 to 0.35), p = .10, k = 5, I2 = 0% - null despite some individual studies reporting large training effects | standardized | post-training | unclear | end-of-treatment | far-transfer |
| Rhyming specifically, and its dose-response | Cohen d; meta-regression on training hours | rhyming 0.18 (-0.06 to 0.42), p = .14 (null); training hours moderated it (p = .04, +0.01 per hour, CI 0 to 0.03), implying 40+ hours before a significant effect | standardized | post-training | unclear | end-of-treatment | near-transfer |
| Control-group type as a moderator | moderator p-value | not significant (rhyming p = .75; other phonological p = .34) - the one meta-analysis in the cluster where control type does not move the estimate, though on far fewer studies | standardized | not-applicable | unclear | not-applicable | near-transfer |
Cited by
- Does learning music make children smarter or better at school?mixedconf: mediumgc: medium