Investigating the Causal Effects of Arts Education

42 schools randomised in matched pairs (36 elementary, 6 middle); 15,886 students in grades 3-8 in the intent-to-treat frame, 10,548 in the analytic frame; preferred administrative models run on 11,300-12,330 students, writing on 4,842, survey outcomes on ~7,600

Houston Independent School District, grades 3-8, two cohorts (2015-16 and 2016-17). Stratified matched-pair randomisation on grade levels served, demographics, existing arts resources and school achievement.

The Houston Arts Access Initiative is the only school-randomised test of a general increase in arts provision on standardised academic achievement, which makes it the single most decision-relevant study for a school board. Peer-reviewed version in Journal of Policy Analysis and Management; a 2019 Houston Education Research Consortium report (ERIC ED598203) covers the same trial with a slightly different analytic frame and reports several results that the peer-reviewed version withdrew or qualified. Grade B rather than A: it is a single unreplicated trial, the 2019 report concedes a statistically significant baseline imbalance in reading achievement that "calls into question the integrity of the randomization process," survey response was 7-9 percentage points higher in control schools, and the headline non-academic constructs are single survey items. Dose and cost are unusually well documented, which is rare and valuable: a net increase of 5.03 arts partnerships and 9.86 arts educational experiences per school-year, at $14.67 per student per year including matched funds - 0.08% of the district's $9,663 per-pupil spending.

On the outcomes a school board actually argues about, the trial is null. Standardised reading -0.019 (SE 0.020, n = 11,343) and mathematics -0.001 (SE 0.035, n = 11,300) - both point-negative on the two best-powered administrative outcomes. Science -0.023 (SE 0.057). Attendance -0.001. The one positive academic result is WRITING, +0.128 SD (SE 0.043, p < .01), and it sits on the smallest administrative sample (n = 4,842, because the Texas writing test is given only in grades 4 and 7) and on the outcome closest to the treatment; decomposed, it is +0.08 SD on mechanics and +0.18 SD on the expository essay. Disciplinary infractions fell by 3.6 percentage points (SE 0.008), but the authors report that this effect "falls shy of statistical significance" (p = 0.14) under randomisation inference and say they have "somewhat less confidence" in it. Emotional empathy +0.072 (SE 0.025) rests on a SINGLE survey item ("I want to help people who are treated badly") after the authors dropped the construct scale for reliability below 0.6. School engagement (+0.080, SE 0.056) and college aspirations (+0.012) are null overall; the widely quoted larger figures are an elementary subgroup (engagement +0.260, compassion +0.150, aspirations +0.069), and the middle-school subgroup shows significant NEGATIVE effects on engagement (-0.136) and college aspirations (-0.053). Lee bounds for differential survey attrition reduce the engagement and compassion effects to non-significance.

Low. School-level randomisation with matched pairs removes selection into arts provision, which is the confound that contaminates the entire observational arts-participation literature. The residual threats here are baseline imbalance and differential survey response, not heredity.

No replication of this trial exists. Its academic nulls are consistent with the five EEF Learning About Culture randomised trials and with the REAP meta-analyses; its writing result has no independent confirmation. The authors themselves frame the academic finding defensively - "substantial influxes of arts educational experiences do not appear to be detrimental to student growth in outcomes tied to accountability assessments" - which is a does-no-harm claim, not a raises-achievement claim.