Effects of a school-based physical activity intervention on academic performance in 14-year old adolescents: a cluster randomized controlled trial - the School in Motion study

30 Norwegian lower-secondary schools cluster-randomized into 3 arms of 10, but one control school withdrew before baseline (29 schools, 9 control) and one DWBH school withdrew at 3 months; 2,084 consenting students (76% consent rate), mean age 14; ~2,000 with valid baseline data and only 1,682 with valid follow-up on both tests

14-year-old Norwegian lower-secondary students

Nine-month three-arm cluster RCT (NCT03817047, RETROSPECTIVELY registered in 2019 after the 2017-18 trial period). Arms: Physically Active Learning (PAL), "Don't Worry - Be Happy" (DWBH), and control. Both intervention arms targeted 120 min/week of additional in-school physical activity. Outcomes were Norwegian standardized national tests in reading and numeracy. Only 10 clusters per arm, which is thin for a cluster design; retrospective registration weakens the preregistration claim; and the authors report that estimates attenuated and some lost significance under intention-to-treat analysis with imputed data. Sex modified the effect in every academic model (p<0.001 for interaction), so the headline pooled estimates average over two very different subgroups. Adherence was 83% (PAL) and 78% (DWBH), i.e. ~100 and ~94 min/week actually delivered against the 120 min/week target, and only 10 of 19 intervention schools cleared 80% delivery.

This is the strongest positive cluster RCT in the school-PA literature and it is the steelman against a flat null verdict — but the effect sizes are at or below the noise floor of the field: numeracy d=0.12 (PAL) and d=0.23 (DWBH), reading d=0.06 (PAL) and d=0.18 (DWBH). Two features cut against the aerobic-dose mechanism. First, the non-aerobic-framed DWBH arm outperformed the physically-active-learning arm on every outcome. Second, and decisively, the DWBH arm produced its academic effect with NO measured change in physical activity at all (Kolle et al. 2020 companion paper); only the PAL arm moved PA. An academic gain in an arm whose PA did not move cannot be a physical-activity effect. Third, the estimates attenuate under ITT with imputation — PAL reading falls to d=0.03 (0.5 points, 95% CI -0.4 to 1.3), and PAL numeracy in girls and DWBH reading in boys also lose significance. Reading is null in three of the four arm-by-sex cells. Against four larger or longer null trials (A+PAAC, ASK, LCoMotion, Fit to Study) this is a single positive result at d<=0.23 with 10 clusters per arm and retrospective registration.

Randomization removes genetic confounding of the arm contrast. With only 10 clusters per arm, chance imbalance in school-level composition (which is itself heritability-loaded through parental sorting) is a live threat that randomization does not eliminate at this cluster count.

The ScIM result itself has not been directly replicated, but the hypothesis it tests has been tested repeatedly and mostly failed - LCoMotion (Tarp 2016), ASK (Resaland 2016), A+PAAC (Donnelly 2017) and Fit to Study (Wassenaar 2021) are null on achievement or cognition. Recorded as `mixed` rather than `unreplicated` because the literature is not silent on this question, it disagrees.