Summary
Key findings
- About 71% of sampled schools reported a feeding programme, but simply having one showed no clear benefit. Comparing schools with and without a programme, the effect on attendance was at most 1.6 percentage points, and the upper bounds for literacy and numeracy were below zero. The author attributes this to programmes being concentrated in more deprived areas, not to any harm from feeding.
- Coverage mattered. In schools where more than three-quarters of learners were fed, the main estimates allow gains of up to 8.3 percentage points in attendance, 7.7 points in the share of children who can read a Grade 2 English story and 9.5 points in the share who pass the Grade 2 numeracy task, compared with schools with no programme. These upper limits are below the simple regression estimates.
- Compared with schools feeding 25% of learners or fewer, high-coverage schools show upper limits of about 11 percentage points for attendance, 13 points for English reading and 19 points for numeracy. Under the strictest assumptions the effects lie between zero and these values.
- Local-language reading showed no positive effect in any comparison. The benefits of high coverage appeared larger in schools within the recommended pupil–teacher ratio of 53:1, and the upper limit for English reading was higher for boys than for girls.
Methodology & context
- Uses the 2024 Uwezo Uganda national learning assessment (July–August 2024): 4,163 Primary 3–7 learners in 410 schools for English and numeracy, and 1,602 learners in 12 districts for local-language reading. Attendance is the share of enrolled pupils present on the survey day; proficiency means completing a Grade 2-level reading or division task.
- Uses Manski-style non-parametric bounds that layer monotone treatment selection, monotone treatment response and a monotone instrument (reliability of the school's water source), with bootstrapped confidence intervals. Schools with a programme are compared with those without, and schools are compared across coverage levels (the share of learners fed).
- The results are ranges, not point estimates, and come from a single cross-sectional survey. Head teachers reported feeding at school level, so whether each child actually received meals is not observed, and informal arrangements such as packed lunches are not counted.
Summary drafted with AI from the AFEP working paper and policy brief, and checked against them. See the full study for details.
Abstract
This paper estimates the causal effect of school feeding programs on primary-school attendance, English and local-language literacy, and numeracy using Manski-style non-parametric bounds from the 2024 Uwezo Uganda learning assessment. The findings are twofold: (i) the binary “feeding vs no feeding” comparison under the MTS+MIV assumption yields upper bounds that lie below zero for all education outcomes except attendance, but this reflects the concentration of feeding programs in more deprived schools rather than a harmful effect of feeding; (ii) exploiting feeding intensity shows non-negative upper bounds lower than the OLS point estimates for attendance, English literacy, and numeracy (but not local language literacy) when comparing schools with the highest feeding intensity to those with no feeding and the lowest intensity under the MTS+MIV assumption. These findings show that in contexts where school feeding is overwhelmingly parent-led and unevenly covered, the binary ‘has-a-program-or-not’ distinction is uninformative; what matters is the share of children who are actually fed.