Gender-specific effects of physical activity on children's academic performance: The Active Smarter Kids cluster randomized controlled trial

1,129 fifth-graders, 57 Norwegian elementary schools (secondary analysis of the ASK trial)

10-year-old Norwegian schoolchildren

Secondary moderator analysis of the ASK cluster RCT (NCT02132494). Tests a three-way group x gender x baseline-performance-tertile interaction on a composite of numeracy, reading and English. Graded C rather than B because it is a post-hoc three-way interaction on a trial whose primary outcome was null — the design is randomized but the estimand is exploratory, and three-way interactions require several times the sample a main effect does.

The three-way interaction is significant at p=0.044 — a value that would not survive any correction for the number of moderator analyses the ASK team ran. The pattern the authors report is itself a warning: low-performing boys ES=0.11 and low-performing girls ES=0.18 trend positive, middle- and high-performing boys ES=0.03 and 0.09, while middle- and high-performing girls trend NEGATIVE (ES=-0.11 and -0.06). This is the archetype of a subgroup story assembled after a null primary result: the direction of effect reverses across cells, the mechanism proposed (gendered response to a sedentary classroom) was not prespecified, and the largest cell effect is 0.18.

Randomized contrast, so between-arm genetic differences are ruled out. But baseline-achievement tertiles stratify on a trait that is roughly 60% heritable by this age, so "effects" that appear only within tertiles are the estimates most vulnerable to regression to the mean.