What School Food Can Do—and What It Cannot Do Alone

School food environment policies can change what children encounter, select and consume, but the central question is how far those changes travel toward childhood obesity. The evidence supports a conditional answer. Healthier school food can improve targeted dietary exposures and may contribute to healthier weight trajectories, yet it cannot guarantee lower obesity without substitution, sustained delivery, attention to the surrounding food environment, equity and evaluation across the whole causal chain.

That conclusion is less dramatic than either a promise or a dismissal, but it is more useful. It tells a policymaker what a school intervention can reasonably be asked to do, what evidence would be needed for a stronger claim, and where a disappointing outcome should be investigated before the policy is judged a failure.

A causal chain with unequal evidence

The strongest evidence concerns the links closest to the school food environment. Direct provision can increase consumption of targeted healthy foods. Nutrition standards can change what is available and reduce some less healthy options. Nudges can alter selection and purchase when healthier choices are already present. The PLOS meta-analysis gives these proximal effects concrete scale: direct provision increased habitual fruit intake by 0.27 servings per day, competitive standards reduced sugar-sweetened beverages by 0.18 servings per day and unhealthy snacks by 0.17, and school-meal standards increased fruit while reducing fat, saturated fat and sodium. These estimates are specific to the interventions and outcomes studied; they are not a single measure of “healthy eating.”

The next link is harder. Across the pooled analyses, total calories generally did not fall significantly. The policies changed composition more consistently than quantity. That does not erase the dietary effects. Less sodium, less saturated fat, more fruit or fewer sugary drinks can be relevant to health even when total energy does not change. But it does mean that a policy’s success at changing a nutrient or food target should not be reported as a measured reduction in obesity risk unless that outcome was actually assessed and supported.

The downstream evidence is less settled. The full PLOS review found adiposity measures generally unchanged, although individual school-meal studies reported some favourable findings. A separate 2021 review reports in its abstract a significant BMI z-score effect of −0.12 across 100 papers, alongside higher fruit consumption and no vegetable effect. Its abstract also reports substantial risk-of-bias concerns, including missing controls in many non-randomized studies, attrition in randomized trials and weak protection against contamination. These are not two results that can be averaged into a definitive universal BMI effect. They are evidence that the answer depends on what studies and outcomes are included, how implementation and bias are handled, and how far the follow-up extends.

The right interpretation of the disagreement is not that one review must be “correct” and the other useless. The PLOS article provides detailed full-text evidence for its intervention categories, dietary outcomes and limitations. The separate review contributes an abstract-level signal that school food-environment interventions may influence BMI, but it cannot establish from the available record which components or settings drove that result. The prudent conclusion is that school food can improve diet and may influence BMI, while the size, consistency and conditions of an obesity effect remain uncertain.

This is a distinction between evidence of absence and absence of evidence. A non-significant BMI result does not prove that no long-term benefit exists. It establishes that the study or pooled comparison did not show a reliable effect under its conditions. Conversely, a significant pooled result does not prove that every school programme lowers obesity. It shows an average association or intervention effect within a defined evidence set, with uncertainty about transfer and mechanism.