Demographic controls erase most of the test-score gap between schools with and without universal meals

Scientific American's own analysis looks most damning and explains the least. It paired test-score estimates from the Stanford Education Data Archive with federal records of which schools participate in the Community Eligibility Provision and Provisions 2 and 3 of the National School Lunch Program ^[16]. Schools with universal meals came out about 1.6 grade levels lower in reading and 1.66 lower in math, and the gap narrowed to roughly two-thirds of a grade level once demographic factors such as poverty and race were accounted for ^[18]. That single adjustment removes about 58 percent of the raw reading gap ^[19]. The remaining third reflects unmeasured differences between adopting and non-adopting schools, not what lunch does.

The studies built to isolate the policy give small positive numbers. Amy Ellen Schwartz of the University of Delaware and a colleague, looking at New York City middle schools, estimated math gains of as much as 0.083 standard deviations and English gains of 0.059 for students who were not poor, against 0.032 and 0.027 for poor students ^[8][9]. The non-poor math estimate is about 2.6 times the poor-student one ^[10]. Orgul Ozturk, the University of South Carolina economist who coauthored the South Carolina study, was direct about the spread: "I think you see variation on who benefits, and definitely the effects are heterogeneous," she said ^[23]. "And benefits are positive, even when they are small, considering it's not a very costly program in some ways," Ozturk said ^[12].

Cost holds up best. The same federal appropriation funded more than 4.87 billion lunches ^[4]. That works out to no more than about $3.53 of federal money per lunch ^[5]. It covers the federal program only, since Seattle and New York State pay for universal programs of their own ^[21]. "In terms of an intervention, it's cheap compared to almost any of the other things that you might be thinking of to improve your school," Schwartz said ^[7].

Test scores also happen to have a five-year national panel behind them. The researchers who spoke to Scientific American named absenteeism and food insecurity alongside test scores as benefits ^[20]. Standardized reading and math scores measure neither. On this evidence I would not defend a universal meals program on academic performance. The narrower claim the numbers support is that the program is cheap and its measured academic effects are small and unevenly spread ^[9][12].

The budget argument is moving ahead of the literature. "Subsidizing school food is an outdated use of federal dollars," Chris Edwards, an economist at the libertarian Cato Institute, said in a report last year ^[14]. "Congress should repeal school food programs to reduce budget deficits and hand power back to the states," he said in the same report ^[15].