A population-level analysis of associations between school music participation and academic achievement.

112,916 students (15,483 music participants, 97,433 non-participants) across four cohorts in all 60 British Columbia school districts

British Columbia public-school students followed through linked K-12 administrative records; outcomes are provincial EXAM grades in English 10, English 12, Maths 10 and Science 10

The strongest observational case for music, and the one arts advocacy quotes most, so it is worth stating what makes it strong and what makes it insufficient. Strong: population-level administrative data with no volunteer sampling, provincially standardised exams rather than course grades, multilevel models nesting students in schools, and prior achievement (Grade 7 numeracy and reading) controlled. Insufficient: the dataset has NO parental education and NO household socioeconomic status - the authors substitute neighbourhood median income and neighbourhood share with postsecondary education - and no measure of conscientiousness, openness or parental involvement, all of which predict both music enrolment and achievement. Grade C by the archive's grid: a longitudinal study with strong covariate and prior-achievement controls, which is exactly what C describes.

Controls cut the raw advantage roughly in half but do not eliminate it: English 12 d = 0.36 unadjusted to 0.22 adjusted; English 10 0.40 to 0.23; Maths 10 0.36 to 0.25; Science 10 0.46 to 0.34. That 41-53% attenuation is the number to carry - it is the size of the selection effect this design can see, and there is no reason to think it is the whole of it. The instrumental versus vocal contrast is the paper's most interesting internal result: adjusted instrumental d = 0.28-0.44 against vocal d = 0.05-0.13, with a dose gradient (five or more instrumental courses gives d = 0.37-0.55). The authors themselves list what is missing with unusual candour, including "children's conscientiousness and openness to new experiences, as well as family context, such as parental involvement at home, attitudes about music, and parenting style", and concede the design "cannot establish cause-effect relationships." They also flag possible OVERcontrol: some children began music before Grade 7, so the prior-achievement covariate may absorb genuine early effects.

High and unaddressed. Prior achievement at Grade 7 is heavily heritable, parental education is absent entirely, and the substitute is a neighbourhood aggregate. Gustavson et al. (2021) put instrument engagement heritability at 0.78 with a genetic correlation to verbal IQ, which is the mechanism that would generate exactly this residual. Schellenberg and Lima (2024) read the paper this way explicitly: "high-functioning individuals appear to do well in school and to take music lessons, and their intellectual advantages may grow larger over time."

No published critique or reply exists despite 113 citing works. The nearest thing is Schellenberg and Lima's 2024 Annual Review, which accepts the finding and reinterprets it as selection. It is in direct tension with Elpus (2013), where the analogous US association goes to zero once prior achievement enters, and with Foster and Marcus Jenkins (2017). Note the two studies are not measuring the same population: Guhn's participation rate is 14% because guitar and general music were excluded, against Elpus's 36% nationally.