The Evidence on Teaching

The Effects of Musical Training on Child Development: a Randomized Trial of El Sistema in Venezuela

Alemán X, Duryea S, Guerra NG, McEwan PJ, Muñoz R, Stampini M, Williamson AA · 2017

grade Arctindependentunreplicatednumbers spot-checked
Sample
2,914 children (2,529 guardians as randomisation clusters) across 16 oversubscribed music centres
Population
Venezuelan children aged 6-14 applying to El Sistema orchestra centres, many exposed to community violence; follow-up at one year
Design
A lottery over admission dates - early admission September 2012 versus delayed admission September 2013 - at 16 oversubscribed centres, which is the cleanest identification strategy available for a voluntary programme and the one this archive grades highest. Registered as NCT02369315 and funded by the Inter-American Development Bank rather than by the programme. Grade A: large, randomised, independently evaluated, with an intention-to-treat analysis and a pre-specified family-wise error correction. The material limitation is that all outcomes are self- or guardian-reported scales, and take-up was partial (69% then 58% of the treatment group against 15% then 14% of controls; average difference 0.98 semesters).
Key findings
The most decision-relevant result in the music cluster, because it is a large lottery and because of WHERE the effects land. After one year the intention-to-treat effects on child-reported self-control (+0.095 SD, 90% CI 0.013-0.177) and behavioural difficulties (-0.081 SD, 90% CI -0.160 to -0.003) are significant at the 10% level after family-wise correction, and there are NO significant full-sample effects on aggression, prosocial behaviour, or COGNITIVE skills. The effects concentrate in vulnerable subgroups: boys exposed to violence show self-control +0.205 (0.039-0.370), aggression -0.242 (-0.450 to -0.034) and behavioural difficulties -0.253 (-0.415 to -0.092); children of less-educated mothers show self-control +0.177; younger children +0.134. The authors' own explanation of the SES gradient is instructive - it "may reflect the ability of more-educated mothers to finance alternative activities", i.e. the counterfactual matters more than the treatment.
Genetic confound
Low. Randomised admission date within an oversubscribed applicant pool means both arms self-selected equally into wanting music; genotype cannot differ between arms.
Replication notes
Unreplicated. No other lottery-based evaluation of El Sistema or a comparable large-scale music programme exists; Sistema Scotland, In Harmony and Musical Futures have non-randomised evaluations only. Its cognitive null is consistent with Sala & Gobet (2020) and with the EEF First Thing Music trial; its non-cognitive positives are consistent in direction with the Houston arts RCT's behaviour findings, though both are fragile.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Self-control (child-reported), full sample, intention-to-treatstandard deviations+0.095 (90% CI 0.013-0.177), p = .056 after k-familywise error controlstandardized1 year after admissionbusiness-as-usual1-2yrnon-cognitive
Behavioural difficulties (child-reported), full samplestandard deviations-0.081 (90% CI -0.160 to -0.003), p = .088standardized1 yearbusiness-as-usual1-2yrbehaviour
Cognitive skills, prosocial behaviour and aggression, full samplestandard deviationsno significant effects on cognitive skills, prosocial behaviour, connections or aggression in the full sample. The authors attribute the cognitive null partly to the one-year horizon and to the fact that "many children were only exposed to introductory training in a single instrument."standardized1 yearbusiness-as-usual1-2yrfar-transfer
Boys exposed to violence at baseline (n approx. 443)standard deviationsself-control +0.205 (0.039-0.370, p = .042); aggressive behaviour -0.242 (-0.450 to -0.034, p = .056); behavioural difficulties -0.253 (-0.415 to -0.092, p = .01)standardized1 yearbusiness-as-usual1-2yrbehaviour
Children of less-educated mothersstandard deviationsself-control +0.177 (0.057-0.298, p = .016); behavioural difficulties -0.134 (-0.253 to -0.016, p = .062). The authors read the gradient as reflecting the alternative activities better-off families can buy.standardized1 yearbusiness-as-usual1-2yrnon-cognitive
Take-upparticipation ratestreatment 69% first semester, 58% second, 56% both; control 15%, 14%, 11%. Average treatment-control difference 0.98 semesters (p < .001). Follow-up completion 74.5% treatment, 77.3% control.standardized1 yearbusiness-as-usual1-2yrbehaviour

Cited by