Is It the Music or Is It Selection Bias? A Nationwide Analysis of Music and Non-Music Students’ SAT Scores

ELS:2002 nationally representative cohort; n = 8,800 for SAT models, n = 12,840 for the ELS mathematics-test replication; transcript analysis covers 1.127 million students, 36.4% of the US class of 2004

US high-school class of 2004, National Center for Education Statistics restricted-use data with full transcripts

The cleanest available dissection of the single most-quoted claim in American music advocacy - that music students score higher on the SAT. Method is school fixed-effects OLS with covariate domains added sequentially, not propensity-score matching (a widespread secondary-source error). Alpha set at .01. Grade C: it is observational and the identification rests on covariate adjustment rather than on a design, but the sequential structure is what makes it useful, because you can watch the gap die one covariate block at a time. Note also that the coefficient tables are in a SAGE online supplement; the values below are the point estimates given in the article narrative, without standard errors.

The raw SAT advantage for music students is +37.30 points, "more than one sixth of a standard deviation." Adding school fixed effects shrinks it modestly. Adding demographics takes it to +25.61, a 31% decline. Adding PRIOR ACADEMIC ACHIEVEMENT and IEP status takes it to +0.23 - statistically indistinguishable from zero. Adding time use and school attitudes flips the sign negative and non-significant. The dosage story dies the same way: +15.3 points per additional year-long music course raw, +12.82 with school fixed effects, and no practical or statistical significance once demographics and prior achievement enter; at a looser alpha, four years of music buys +8.68 points, 0.04 SD, less than the College Board's own measurement band. The pattern replicates on the ELS mathematics test (n = 12,840): +2.14 points raw, +1.64 after demographics, gone after prior achievement. Two disaggregations worth keeping: the raw association is driven by INSTRUMENTAL (band and orchestra) students, and in the fully controlled model CHORAL students carry a significant 10-point SAT DEFICIT. The strongest predictors throughout are prior achievement, IEP status and socioeconomic status.

This is the confound made visible. Prior academic achievement is the covariate that annihilates the effect, and prior achievement is the most heritable variable in the model. Elpus cannot separate genetic from environmental selection, but he establishes that the entire raw gap lives in who enrols, not in what enrolling does.

No direct replication on US national data. Guhn et al. (2020) run a conceptually similar analysis on 112,916 British Columbia students with prior achievement controlled and find the association attenuated by 41-53% but NOT eliminated - the central disagreement in this literature. Foster and Marcus Jenkins (2017), using propensity weighting on PSID, side with Elpus. Elpus's own recommendation is blunt: "it may be time to question seriously, or even retire, the advocacy argument that implies music students are more academically successful on standardized tests than their non-music peers because of their enrollment in music."