Musical instrument engagement in adolescence predicts verbal ability 4 years later: A twin and adoption study.

1,684 adolescents (232 MZ pairs, 203 DZ pairs, 177 biological sibling pairs, 176 ADOPTIVE sibling pairs); age-12 wave n = 1,373, age-16 wave n = 1,641, both waves n = 1,287

Colorado Longitudinal Twin Sample and Colorado Adoption Project (CATSLife), primarily white American families; music engagement self-reported at ~12.5 years, verbal ability at ~16.6

The design that matters here is the adoptive-sibling arm: adoptive siblings share a household but no genes, so a shared-environment account of "musical families produce verbal children" is directly testable. This is the closest thing the music literature has to the discordant-twin evidence the archive uses elsewhere, and it is genetically informative about the ENGAGEMENT variable - who takes up an instrument - which is exactly the variable every observational music-and-achievement study treats as an exposure. Numbers below come from the PMC full text (results tables and limitations section), not from the abstract.

Instrument engagement at age 12 is one of the most heritable "choices" in this archive: a2 = 0.78 [0.55, 0.94], with shared environment indistinguishable from zero (c2 = 0.11 [0.00, 0.29]) and non-shared environment 0.11 [0.03, 0.21]. Singing engagement a2 = 0.43 [0.18, 0.63] and dance a2 = 0.66 [0.38, 0.80]. Instrument engagement - but not singing or dance - was genetically correlated with age-12 full-scale IQ (rg = 0.44 [0.21, 0.77] in twins; 0.80 [0.35, 1.0] in siblings), i.e. the same genes that predispose a child to pick up an instrument predispose them to score well. The steelman for music is the residual: instrument engagement at 12 still predicted age-16 verbal ability after controlling for age-12 full-scale IQ, beta = 0.09 [0.04, 0.15] in twins and 0.10 [0.04, 0.16] in siblings. The authors call this evidence of "some (weak) direct benefits." They also flag that they could not run a direction-of-causation model because engagement and verbal ability had similar genetic architectures, and that openness and SES remain unmeasured confounds.

This source IS the genetic-confound evidence. It converts "children who take music lessons differ from those who do not" from an assumption into a measured a2 = 0.78 with a genetic correlation to verbal IQ. Any observational study that reports a music-participation advantage without genetic controls is estimating this correlation, not a treatment effect. Residual limits: self-report single-item indicators, an American mostly-white sample, and biological-sibling correlations markedly below DZ correlations, which strains the twins-and-siblings equivalence assumption.

No direct replication located. A separate ABCD-cohort analysis by the same first author reports heritable childhood music engagement with language and executive-function associations, which is consistent but not an independent replication of these parameters.