The Impact of Skipping Breakfast on Academic Performance in Youths: A Meta-Analysis of Observational Studies

24 observational studies contributing 37 reports, 172,988 participants; literature searched to October 2024 in PubMed, ScienceDirect, SpringerLink and Scopus. Individual study sizes 402 to 75,643.

School-aged youths internationally, ages 6-29 (several samples are university students, not schoolchildren). Every one of the 37 pooled reports is CROSS-SECTIONAL - the single longitudinal cohort study identified had to be excluded from the quantitative analysis.

Meta-analysis of OBSERVATIONAL studies only, pooled as odds ratios with random effects in RevMan 5.4. Study quality assessed with a Newcastle-Ottawa Scale adapted for cross-sectional designs (2 unsatisfactory, 5 satisfactory, 16 good, 2 very good; mean 7.1). Heterogeneity is severe: I2 = 83%, p < 0.00001, and no subgroup cut brings it below 41%. Begg test found no evidence of publication bias (p = 0.29) - but Begg test is badly underpowered at this heterogeneity, and the paper own subgroup table shows a textbook small-study gradient it did not interpret as one: studies under 1,000 participants pool to OR 2.66, studies of 10,000 or more pool to OR 1.50. No trim-and-fill or PET-PEESE correction was attempted. Per METHODOLOGY.md a meta-analysis inherits the confounding of what it pools, so this is grade D regardless of its internal execution. It is recorded here precisely because it quantifies the naive correlation that the causal literature must be compared against.

Breakfast skipping is associated with roughly a DOUBLING of the odds of poor academic performance (pooled OR 2.08, 95% CI 1.82 to 2.37; I2 = 83%). The association is directionally consistent across every subgroup the authors cut - gender, region, sample size, assessment tool, outcome definition and covariate adjustment - and significant in all but one thin cell. CORRECTION TO OUR EARLIER READING: it is NOT invariant in MAGNITUDE, and the paper own Table 2 says so. Adjusting for sleep drops the pooled OR from 2.21 to 1.61; adjusting for socioeconomic status drops it from 2.57 to 1.64. Roughly half the excess odds is accounted for by two measured covariates, in a literature that measures almost nothing else. That is the opposite of the invariance our record previously claimed, and it points the same direction for a different reason: if two crude proxies for household routine and family resources absorb half the association, the unmeasured heritable and environmental correlates plausibly absorb much of the rest. The effect also shrinks as studies get larger (OR 2.66 under 1,000 participants, 1.50 at 10,000 or more) and as outcome measures get more objective in the opposite direction (self-reported performance OR 1.59, maths tests OR 2.90) - both signatures of a literature dominated by small, weakly controlled studies. Randomized breakfast provision in comparable populations moves achievement by approximately 0 SD.

Very high. Breakfast skipping tracks household chaos, poverty, parental conscientiousness, adolescent sleep timing and parental supervision - all traits with substantial heritability that also predict school performance directly. No included study uses a within-family, discordant-twin or genetically informed design; all 37 pooled reports are cross-sectional, so reverse causation (disengaged, poorly performing adolescents keeping later and more chaotic hours) is unexcluded. Covariate adjustment removes about half the association and cannot remove passive gene-environment correlation at all.

The association itself is replicated to the point of monotony - 37 reports from 24 studies across five continents all point the same way, and the archive's other observational entry (nutr-sincovich-2025-breakfast-skipping-cross-sectional, 28,651 South Australian students) reproduces it again. What is NOT replicated is any causal reading of it - randomized breakfast provision in comparable populations (Luan 2022, Ni Mhurchu 2013) finds approximately zero.