The impact of school feeding on nutrition, education, agriculture and other social outcomes: design of a randomised controlled trial of 'home-grown' school feeding in Ghana

The randomly assigned interventions are:
1. School feeding programme, where the standard Ghana School Feeding Programme is implemented
2. 'Home-grown' school feeding (HGSF+) group, including schools and communities where the Ghana School Feeding Programme is implemented in addition to district-level training of community based organizations and local government
3. Control group, including schools and household from communities where the intervention will be implemented at the end of the trial (preferably without informing schools and households of impending intervention).

2013-09-01

2016-06-01

Agriculture: Household income, production, sales.
Education: Child enrolment, attendance, completion, maths and literacy scores (5–15 year old).
Cognition: Raven’s test and forward/backward digit span scores (5–15 year old).
Physical health/Nutrition :Anthropometry (height-for-age, BMI-for-age, 2–15 year old,
haemoglobin levels (5–15 year old)

The selection of the study areas involved two key steps. I) Selecting 58 districts at random within Ghana from a sample frame including all districts in the country. The sample frame was stratified by region and district inclusion was prioritised using data from the GSFP retargeting exercise including data on the prevalence of poverty and food insecurity. II) Identifying 2 comparable schools within each of the 58 selected districts. A protocol was designed to ensure that the schools were comparable based on data from the Education Management Information system (EMIS) and that potential for contamination and crossover between the schools and pupils in each district was minimised. This step utilized a list from the GSFP secretariat of schools not currently covered by GSFP in each district. Data from the annual school census from 2011-2012 was then used to match schools not receiving the GSFP and identify “best matched” pair. The allocation into school feeding and control schools was then randomised within each pair. The 58 schools assigned to the intervention were then randomly assigned to regular GSFP and HGSF+. In this way the randomisation of the HGSF+ intervention occurs at the district level. The number of 58 schools is based on power calculations determined with the objective of achieving statistical validity and representativeness for the main outcomes of interest.

Lottery style

Selection of the target areas involved three key steps: 1) the first step involved selecting 58 districts at random
within Ghana from a sample frame including all districts in the country. The sample frame was stratified by region, and district inclusion was prioritised using data from the GSFP retargeting exercise including data on
the prevalence of poverty and food insecurity; 2) the second step involved identifying 2 comparable schools
within each of the 58 selected districts. A list was obtained through the GSFP secretariat including schools
not currently covered by the GSFP in each district. Data from the annual school census from 2011 to 2012 were
then used to match schools not receiving the GSFP and identify ‘ best matched’ pair. The allocation of school
feeding and control was then randomised (lottery style) within each pair; 3) the third step in the site selection
protocol involved the random allocation of districts to the HGSF+/GSFP groups by modelling pilot selection
using a set of community- and district-level variables and selecting the permutation of allocation that minimises
the R-squared for the predicted selection

Yes

58 districts, 116 schools

A total of 2626 households in 116 communities were surveyed at baseline

58

For the impact evaluation, power calculations and resource availability suggested the adoption of a sample of 25 households from the communities in the areas of the 58 schools receiving the intervention and of 20 households in the communities of the 58 control schools. Households were randomly selected from household listings in the catchment areas of the selected schools for the survey interviews. The household listings were stratified into farmer/non-farmer households, based on agriculture classification data from the national census. Farmer households were sampled in both areas in the following way: 10 out of the 25 households in the 60 intervention communities were farmer households and 5 out of the 20 households in the 60 control communities were farmer households. Non-farmer households with children in the 5– 15 years age group were randomly selected from the household listings. This distribution of the sample between farmer and non-farmer households and between project control groups allows the construction of comparable samples. Power calculations for the following outcomes were computed: a. School attendance: we used the rural sample of the Ghana DHS data of 2008 to estimate attendance rates of children in the age group 6 to 14 and we found rates of 79% for boys and 81% for girls. A sample of 60 clusters and collecting data on 40 children is sufficient to detect such an impact with 80% statistical power. b. Cognitive tests: we obtained data on outcomes of cognitive tests from a sample of rural children tested in 2003 using Raven’s matrices. The average score on the test was 15.3 out of 36 questions with a standard deviation of 5.9 and an intracluster correlation coefficient of 0.14. We assumed a number of 40 children per cluster considering 20 households interviewed in each cluster and an average of 2.3 children in the relevant age group per each household with children. c. Child nutrition: we used data of DHS Ghana 2008 to estimate mean and standard deviation of height-for-age Z-scores of rural children and we found these to be -1.03 and 1.57 respectively. The intracluster correlation coefficient is 0.08. d. Farm income: we used data from GLSS4 of 1998/99 to estimate average farm income of rural households (1,200 thousands Cedis) and relative standard deviation (1,400 Cedis). We found an extremely high intracluster correlation coefficient. Income is the most difficult outcome to estimate with sufficient precision.

Design and analysis plan

There is limited experimental evidence on the effects of large-scale, government-led interventions on human capital in resource-constrained settings. We report results from a randomized trial of the government of Ghana's school feeding. After two years, the program led to moderate average increases in math and literacy standardized scores among pupils in treatment communities and to larger achievement gains for girls and disadvantaged children and regions. Improvements in child schooling, cognition, and nutrition constituted suggestive impact mechanisms, especially for educationally disadvantaged groups. The program combined equitable human capital accumulation with social protection, contributing to the “learning for all” sustainable development agenda.

Food for Thought? Experimental Evidence on the Learning Impacts of a Large-Scale School Feeding Program Elisabetta Aurino, Aulo Gelli, Clement Adamba, Isaac Osei-Akoto and Harold Alderman Journal of Human Resources, January 2023, 58 (1) 74-111; DOI: https://doi.org/10.3368/jhr.58.3.1019-10515R1

https://jhr.uwpress.org/content/58/1/74

Background: Attention to nutrition during all phases of child and adolescent development is necessary to ensure healthy physical growth and to protect investments made earlier in life. Leveraging school meals programs as platforms to scale-up nutrition interventions is relevant as programs function in nearly every country in the world.

Objective: The aim of this study was to evaluate the impact of a large-scale school meals program in Ghana on school-age children's anthropometry indicators.

Methods: A longitudinal cluster randomized control trial was implemented across the 10 regions of Ghana, covering 2869 school-age children (aged 5-15 y). Communities were randomly assigned to 1) control group without intervention or 2) treatment group providing the reformed national school feeding program, providing 1 hot meal/d in public primary schools. Primary outcomes included height-for-age (HAZ) and BMI-for-age (BAZ) z scores. The analysis followed an intention-to-treat approach as per the published protocol for the study population and subgroup analysis by age (i.e., midchildhood for children 5-8 y and early adolescence for children 9-15 y), gender, poverty, and region of residence. We used single-difference ANCOVA with mixed-effect regression models to assess program impacts.

Results: School meals had no effect on HAZ and BAZ in children aged 5-15 y. However, in per-protocol subgroup analysis, the school feeding intervention improved HAZ in 5- to 8-y-old children (effect size: 0.12 SDs), in girls (effect size: 0.12 SDs)-particularly girls aged 5-8 y living in the northern regions, and in children aged 5-8 y in households living below the poverty line (effect size: 0.22 SDs). There was also evidence that the intervention influenced food allocation and sharing at the household level.

Conclusion: School meals can provide a platform to scale-up nutrition interventions in the early primary school years, with important benefits accruing for more disadvantaged children. This trial was registered at www.isrctn.com as ISRCTN66918874.

A School Meals Program Implemented at Scale in Ghana Increases Height-for-Age during Midchildhood in Girls and in Children from Poor Households: A Cluster Randomized Trial Aulo Gelli , Elisabetta Aurino , Gloria Folson , Daniel Arhinful, Clement Adamba , Isaac Osei-Akoto , Edoardo Masset , Kristie Watkins , Meena Fernandes , Lesley Drake , Harold Alderman

https://pubmed.ncbi.nlm.nih.gov/31100125/