Does music training improve inhibition control in children? A systematic review and meta-analysis

22 longitudinal controlled studies (N = 1,734), of which 8 are randomised controlled trials with active control groups

Children aged 3-11; records screened 1980-2023

The strongest live challenge to the archive's deflationary reading of music transfer, and it is included precisely because it is inconvenient. Where every other synthesis pools heterogeneous cognitive outcomes, this one targets a single theory-motivated construct - inhibition control - and finds the effect gets LARGER, not smaller, in the randomised active-control subset, with zero heterogeneity. Grade C: 22 studies, N = 1,734, and only 8 randomised; the authors restrict to peer-reviewed journals and acknowledge that this leaves a publication-bias risk they can only test with funnel-plot symmetry and Egger's test. One methodological caution the archive should state plainly: the authors benchmark their result against Hattie's d = 0.40 "barometer of influence," a naive meta-meta average this archive explicitly rejects, so the "practically important" gloss should not be inherited.

Across 8 RCTs with non-musical ACTIVE control training, music training improved inhibition control by g = 0.60 (SE 0.11, 95% CI 0.39-0.82, p < .0001) with I2 = 0.0% - a moderate- to-large effect with no between-study heterogeneity. Across all 22 longitudinal studies the effect is g = 0.31 (SE 0.07, CI 0.19-0.44). Every study except two had a positive effect size. The moderator pattern runs OPPOSITE to Sala & Gobet: RCT design added +0.19 (CI 0.07-0.32, p = .002) over non-RCT, and active controls added +0.13 (CI 0.02-0.25, p = .022) over passive - i.e. the more rigorous the design, the larger the effect. Training in a specialised music facility added +0.13 over school-based training (p = .047). Age (p = .68) and training intensity (p = .69) did not moderate. No funnel-plot asymmetry and Egger's test non-significant (p > .33). The authors go further than the data compel: "Music training plays a privileged role compared to other activities (sports, visual arts, drama) in improving children's executive functioning."

Low within the RCT subset by construction - randomisation plus active controls removes both selection and generic activity effects. The 14 non-randomised studies carry the usual selection confound and produce the smaller estimate.

Unreplicated, and squarely in tension with Sala & Gobet (2020), whose active-control estimate across a broader outcome set is g = 0.056 falling to -0.021. The two are not strictly contradictory - Jamey et al. target one construct across 8 studies while Sala & Gobet pool 254 effect sizes across many - but they cannot both be a fair summary of what music training does to executive function. Note also that Román-Caballero et al. (2022) independently found executive functions to be the strongest domain cell (gDelta = 0.41), so two independent teams now locate the surviving effect in the same place. That convergence is the single best argument against treating music transfer as closed.