The Evidence on Teaching

Longitudinal Relationship between Cardiorespiratory Fitness and Academic Achievement

Sardinha LB, Marques A, Minderico C, Palmeira A, Martins S, Santos DA, Ekelund U · 2016

grade Dlongitudinalindependentunreplicated
Sample
1,286 students in grades 5-7, aged 11-14 at baseline, from 14 Portuguese schools, followed 3 years
Population
Portuguese schoolchildren aged 11-14 at baseline
Design
Three-year prospective cohort. Cardiorespiratory fitness measured by the FITNESSGRAM PACER shuttle test at baseline and follow-up; students classified fit-fit, unfit-fit, fit-unfit, unfit-unfit. Outcome is teacher-assigned school marks in Portuguese, mathematics, English and science. Ordinal regression. Graded D: the outcome is teacher marks rather than an independent standardized test, the exposure is self-selected, and there is no genetically informative or within-family component. Recorded here as the best-executed exemplar of the observational fitness-achievement literature, not as causal evidence.
Key findings
The effect sizes here are enormous compared with anything a randomized trial produces, which is the point. Persistently fit students had 3.49 times the odds (95% CI 1.97-6.20) of high achievement in Portuguese and 2.41 (1.39-4.14) in English, versus persistently unfit students. Students who became fit had OR 2.52 and 2.13 on the same two subjects. Two things break the causal reading. First, the effects appear in mother-tongue and foreign language and NOT in mathematics or science — the reverse of the mathematics-specific pattern every intervention meta-analysis reports, so the two literatures do not even agree about which subject responds. Second, odds ratios of 2.4-3.5 are one to two orders of magnitude larger than the d=0.00-0.23 range randomized trials return for the same exposure. A causal effect does not shrink that far when you randomize; a confound does.
Genetic confound
HIGH. Cardiorespiratory fitness is roughly 55-72% heritable with no detectable shared-environment component (Schutte et al. 2016), and school achievement is roughly 60% heritable by adolescence. Persistently fit and persistently unfit adolescents are not randomly assigned groups: they differ in genotype, body composition, family circumstance and conscientiousness. No within-family, twin or sibling-fixed-effects version of this analysis exists anywhere in this literature.

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
High achievement in Portuguese (mother tongue), fit-fit vs unfit-unfitodds ratioOR 3.49 (95% CI 1.97 to 6.20), p<0.001researcher-designed3-year follow-up; teacher-assigned marksnonenot-applicabledomain-skill
High achievement in English (foreign language), fit-fit vs unfit-unfitodds ratioOR 2.41 (95% CI 1.39 to 4.14), p<0.01researcher-designed3-year follow-upnonenot-applicabledomain-skill
High achievement in Portuguese, unfit-fit vs unfit-unfitodds ratioOR 2.52 (95% CI 1.42 to 4.45), p<0.01researcher-designed3-year follow-upnonenot-applicabledomain-skill
Mathematics and science achievementodds rationo reported association — the subjects intervention meta-analyses claim respond mostresearcher-designed3-year follow-upnonenot-applicabledomain-skill

Cited by