When Better Food Changes the Diet—but Not the Calories

This chapter reads the strongest quantitative evidence on school food environment policies: a systematic review and meta-analysis of 91 interventions. The focus is deliberately narrower than “did school food work?” We will ask which dietary targets moved, how large those changes were, and why a result for fruit, sodium or sugary drinks cannot be translated directly into a claim about total energy or body weight.

That narrowness is a strength. A policy can be valuable without changing every dietary measure, and a pooled result becomes misleading when it is detached from the food, outcome and comparison that produced it. The review separates direct provision, competitive-food standards and school-meal standards, then distinguishes what children ate at school from what they ate habitually across the day.

What the intervention studies actually found

The PLOS One review began with 6,636 abstracts and included 91 interventions: 39 randomized studies and 52 non-randomized studies. The interventions were not all the same. Some provided healthier foods or beverages, some set rules for competitive foods and drinks sold outside the ordinary meal programme, and some changed standards for school meals. Several studies assessed more than one policy, so the category counts should not be added as if they were mutually exclusive.

That design matters when the numbers are compared. A randomized study and a quasi-experimental study can both contribute useful evidence, but they do not create the same comparison or control the same sources of bias. The pooled estimate is an average across the included interventions, not a forecast for a named school. It incorporates differences in age groups, settings, policy intensity, duration and the way diet was measured. The review’s strength is that it separates the intervention families and standardizes outcomes where possible; its limitation is that standardization cannot make unlike exposures identical.

The units also need a little care. A “serving per day” is a standardized dietary measure, not a count of children or a promise about each child’s plate. A percentage of energy from fat describes composition, not the number of calories eaten. A milligram change in sodium is not interchangeable with a serving of fruit. These measures answer different nutritional questions, so combining them into one score called “healthier eating” would erase the very distinctions the review was designed to preserve.

The clearest pattern appeared in fruit and vegetables. Direct provision increased habitual fruit intake by 0.27 servings per day, with a 95% confidence interval from 0.17 to 0.36. A confidence interval is the range of values compatible with the pooled estimate under the study’s assumptions; it is not a guarantee that every school will produce the same change. Combined fruit and vegetable intake increased by 0.28 servings per day (95% confidence interval 0.17 to 0.40). The pooled vegetable effect was much smaller: 0.04 servings per day (0.01 to 0.08).

Those results are meaningful but specific. They say that interventions supplying or making healthier foods available were associated with more reported habitual fruit or vegetable consumption in the included studies. They do not say that every healthy food increased, that every child changed their diet, or that the average child consumed a full additional serving of vegetables. The authors note that fruit effects may be larger partly because fruit is often more palatable and easier to serve than vegetables. That is a plausible explanation for the pattern, not a universal law of school meals.

Direct provision did not produce a significant pooled change in habitual calories in the six studies that measured it: the estimate was −56 kilocalories per day, with a confidence interval from −174 to 62. The interval includes both a moderate decrease and a small increase, so the evidence does not establish a reliable calorie reduction. In the limited studies that targeted water, habitual intake also did not show a significant change: 0.33 glasses per day, with a confidence interval from −0.27 to 0.93. Measuring water intake is difficult, and the small evidence base makes that estimate less informative than the fruit findings.

Competitive-food and beverage standards produced a different pattern. They generally targeted sugary drinks and unhealthy snacks rather than supplying fruit directly. Habitual sugar-sweetened beverage intake fell by 0.18 servings per day (95% confidence interval −0.31 to −0.05), and unhealthy snack intake fell by 0.17 servings per day (−0.22 to −0.13). But habitual calorie intake again did not fall significantly: −79 kilocalories per day, with a confidence interval from −179 to 21. The policy could change the composition of what children consumed without establishing a reduction in the total amount of energy consumed.

School-meal standards showed yet another profile. They increased habitual fruit intake by 0.76 servings per day in the two estimates available (95% confidence interval 0.37 to 1.16). They reduced habitual total fat by about 1.49 percentage points of energy (−2.42 to −0.57), saturated fat by 0.93 percentage points of energy (−1.15 to −0.70), and sodium by 170 milligrams per day (−242 to −98). These are changes in dietary composition and nutrient exposure. Habitual calories did not significantly decrease: the pooled estimate was −38 kilocalories per day, with a confidence interval from −137 to 62.

The contrast is the point. In the same broad domain, policies could increase fruit, reduce sugary drinks, lower fat or lower sodium while leaving total calories statistically indistinguishable from the comparison condition. “Healthier” is not one dimension. A diet can contain less sodium or saturated fat and more fruit without producing the same total energy intake. That combination may matter for health, but it must not be reported as if it were a measured weight-loss effect.

There are several ways to understand this pattern without forcing the data to choose between “healthy” and “effective.” First, substitution may change nutrients more readily than it changes energy: a meal can replace saturated fat with another source of energy while preserving its calorie content. Second, children may eat a healthier item and later obtain comparable energy elsewhere. Third, the intervention may be too short or too modest to produce a detectable change in a body-size measure, even if the diet is moving in a desirable direction. These are explanations to investigate, not conclusions that the pooled data prove separately.

The calorie result also illustrates why a null estimate is not the same as a precisely measured zero. The confidence interval around habitual calories includes no change, but it also includes decreases and, for some interventions, increases. The appropriate statement is that the review did not establish a significant pooled reduction in calories. A stronger statement—that calories could not have changed, or that the intervention had no effect on energy balance—would exceed the evidence. The same discipline applies to water, where the small number of studies and measurement difficulty make a firm interpretation especially weak.

flowchart LR P["Direct provision"] --> PF["Fruit +0.27 servings/day"] P --> PV["Vegetables +0.04 servings/day"] P --> PC["Calories: no significant pooled change"] C["Competitive standards"] --> CS["Sugary drinks −0.18 servings/day"] C --> CN["Unhealthy snacks −0.17 servings/day"] M["School-meal standards"] --> MF["Fruit +0.76 servings/day"] M --> MN["Fat, saturated fat and sodium reduced"] M --> MC["Calories: no significant pooled change"] H["Habitual intake"] -. differs from .-> I["In-school measures"]

Why the target and the measurement level matter

The review’s most important methodological choice was to separate in-school intake from habitual intake. A cafeteria measure answers a local question: what happened during the meal or at the point of sale? A habitual-intake measure attempts to capture what children consumed across school and non-school settings. The latter is more relevant to a claim about the whole diet, but it is also harder to measure and more exposed to recall error and outside influences.

The distinction changes how calorie results should be read. School-meal standards reduced the energy content of meals offered in some analyses, but they did not significantly reduce calories consumed at lunch or across lunch and breakfast, and they did not significantly reduce habitual calories. The food on the menu and the food a child eats are not the same variable. Portion choice, substitutions, leftovers and food consumed elsewhere sit between them.

The results for sodium illustrate the opposite possibility. Sodium reductions were more similar in school meals and habitual intake, suggesting less compensation outside school for that target in the available studies. This does not prove that sodium policies always travel beyond school or that calorie compensation always occurs. It shows why the mechanism must be examined target by target instead of assumed from a general theory about children’s behaviour.

The same caution applies to sales and purchases. A decline in sugary-drink sales can be a useful sign that a standard changed the school food environment. It does not by itself prove a decline in the child’s full-day sugary-drink intake. Conversely, a small or null cafeteria result could coexist with a broader dietary change if the measure is noisy or if children obtain the food elsewhere. The outcome must match the claim being made.

This is also why “statistically significant” is not the only useful phrase. The fruit and sodium estimates are accompanied by intervals that exclude no change in the pooled analysis, while the calorie intervals include no change and a range of decreases or increases. But even a statistically clear dietary effect has to be interpreted for magnitude and relevance. A change of 0.27 fruit servings per day is not the same as a complete dietary transformation. It may be meaningful at population scale, especially if sustained and distributed equitably, but it does not establish that the intervention corrected children’s broader diets.

The review found that fruit effects were generally stronger than vegetable effects, and that water intake did not change in the limited studies that measured it. These differences may reflect taste, convenience, preparation, measurement difficulty and the way each intervention was designed. They also reveal a common analytical mistake: grouping all foods that contribute to a healthy diet into one outcome and then treating a result for the easiest target as evidence for the entire category.

The evidence base itself sets limits on transfer. Of the 91 interventions, 55 were in the United States or Canada and 36 were in Europe or New Zealand. Most came from high-income Western settings. Schools, food prices, meal traditions, household diets and the availability of food outside the gate differ elsewhere. The pooled estimates are therefore best understood as evidence that particular policy mechanisms can change particular dietary targets under the studied conditions—not as a universal effect size for every country or school system.

The review also notes that some studies combined food-environment policies with other components and that implementation intensity was difficult to quantify. A written standard may be applied carefully in one district and weakly in another. A fruit programme may be free and well accepted in one school but reduced-price, poorly stocked or culturally mismatched in another. Pooling such interventions gives useful average evidence, but it cannot remove the differences in exposure.

Nor do these dietary results answer the obesity question by themselves. Total calories are only one bridge between diet and adiposity, and the studies varied in duration and in how well they measured what happened outside school. The absence of a significant calorie change means that the pooled evidence did not establish a reliable reduction in total energy intake for those comparisons. It does not prove that no child consumed fewer calories, that no long-term pathway exists, or that improvements in nutrient composition have no health value. It also does not justify claiming that a healthier menu lowered obesity when that outcome was not measured or did not change.

A disciplined reading therefore keeps three statements separate. First, specific school food policies changed specific dietary behaviours in the intervention evidence. Second, those changes were uneven across foods and often did not extend to total calories. Third, the evidence in this review generally did not show a consistent change in adiposity. The first statement is well supported for several targets; the second is a central result; the third is a downstream conclusion with a different level of uncertainty.

That separation leads to the next problem rather than closing it. If children eat more fruit or less sodium but total calories and adiposity do not reliably move, the explanation may involve compensation, duration, implementation, measurement or the difference between diet quality and diet quantity. The correct response is not to discard the dietary findings or to call them proof of obesity prevention. It is to examine how proximal changes travel—or fail to travel—through the rest of the child’s food environment and over time.

References