Two Randomized Trials Provide No Consistent Evidence for Nonmusical Cognitive Benefits of Brief Preschool Music Enrichment

Experiment 1 n = 29 analysed (music 15, visual arts 14); Experiment 2 n = 45 analysed (music 23, no-treatment control 22)

US preschoolers, mean age approximately 4.7 years, attending parent-child arts classes

Two randomised trials with the design feature the field most lacks - Experiment 1 uses a VISUAL ARTS active control, Experiment 2 a waitlist control - and a deliberate internal replication attempt. The dose is the weakness and the authors say so: six weekly 45-minute parent-child sessions, substantially shorter than other randomised music trials. Grade C on sample size alone; n = 29 and n = 45 make this badly underpowered for the effects it is testing, and the archive should treat it as a failed replication rather than as a decisive null.

Experiment 1 produced two nominally significant one-tailed differences in OPPOSITE directions - music better on map use (65.9% vs 49.2%, d = 0.65) and WORSE on visual form analysis (40.6% vs 51.1%, d = 0.72) - neither surviving the multivariate test (Wilks lambda = .742, F(4,24) = 2.09, p = .113) nor correction for multiple comparisons. Numerical discrimination and receptive vocabulary were null. Experiment 2, designed specifically to check the spatial finding against a no-treatment control, reproduced nothing: map use 62.3% vs 63.1% (p = .911), visual form 44.6% vs 44.2% (p = .933), numerical 80.9% vs 81.1% (p = .971), vocabulary 116 vs 120 (p = .362), multivariate p = .934, and no Spatial x Group interaction (p = .887). Pooled across both experiments, p = .190. The authors' conclusion: "preschool music classes did not cause detectable skill increases in the cognitive domains of spatial, linguistic, or numerical reasoning."

Low - random assignment in both experiments, with a visual-arts active control in the first. Families self-selected into the study, but not into arm.

This is itself a failed internal replication: Experiment 2 was run to confirm Experiment 1's spatial signal and found nothing. It is the counterweight to Schellenberg (2004) in every subsequent meta-analysis, and in Sala & Gobet (2017) it is the single randomised active-control study of spatial and mathematical outcomes, contributing d = -0.25. Its small samples mean it cannot by itself establish a null; what it establishes is that a positive result in one small music trial does not survive being run again.