Food for Thought?

Notes:

↵* p < 0.1,

** p < 0.05,

↵*** p < 0.01.

This table presents probability of remaining in the longitudinal sample estimated through linear probability models, with standard errors clustered at the community level. N= 3,433 children of target-age prior to attrition. Lower sample sizes reflect covariates that are missing or not applicable. Column 1 shows probabilities of child being followed up by treatment assignment; Column 2 presents odd ratios by baseline learning and cognition, while Column 3 interacts randomized assignment with key variables by which heterogeneity analysis was conducted throughout the paper. Household poverty is a dichotomous indicator having the value of one if the household had baseline per capita consumption levels falling below the national consumption poverty line in 2013. Northern regions include Upper West, Upper East, and Northern region. Southern regions include Western, Central, Greater Accra, Volta, Eastern, Asanti, and Brong Ahafo.

Notes: The school feeding—control difference column reports the school feeding coefficient of a basic OLS regression of each outcome over school feeding arm and controlling for child age in months. Standard errors are clustered at the community level. Lower sample sizes in the cognitive scores (as compared to the full longitudinal sample) reflect missing values in those scores.

Notes:

↵* p < 0.1,

↵** p < 0.05,

↵*** p < 0.01.

Confidence intervals clustered at community level in squared brackets. R-W p-values were adjusted for multiple testing using the Romano–Wolf (2005, 2016) step-down method with 2,000 iterations and standard errors clustered at community level. The table above presents intent-to-treat effects on each outcome estimated through ANCOVA for the full sample and stratified by child gender, household poverty, and place of residence. Models were estimated through OLS. For each outcome, the model controls for its baseline value, a dichotomous variable related to the randomized assignment to school feeding, and region dummies. Math and literacy scores are age-standardized. The composite index of learning was computed as the average of the math and literacy scores, and then they were standardized to the control group within each round. Household poverty is a dichotomous indicator having the value of one if the household had baseline per capita consumption levels falling below the national consumption poverty line in 2013. Northern regions include Upper West, Upper East, and Northern region. Southern regions include Western, Central, Greater Accra, Volta, Eastern, Asanti, and Brong Ahafo.

Notes:

↵* p < 0.1,

↵** p < 0.05,

↵*** p < 0.01.

Confidence intervals clustered at community level in squared brackets. R-W p-values were adjusted for multiple testing using the Romano–Wolf (2005, 2016) step-down method with 2,000 iterations and standard errors clustered at community level. The table above presents intent-to-treat effects on each outcome estimated through difference-in-differences for the full sample and stratified by child gender, household poverty, and place of residence. Models were estimated through OLS. Models include a dichotomous variable for treatment assignment, a dummy for endline survey, and the treatment effect relates to the interaction between these two variables. It also includes region dummies. Math and literacy scores are age-standardized. Composite indices were computed as averages of the standardized scores and then they were standardized to the control group within each round. Household poverty is a dichotomous indicator having the value of one if the household had baseline per capita consumption levels falling below the national consumption poverty line in 2013. Northern regions include Upper West, Upper East, and Northern region. Southern regions include Western, Central, Greater Accra, Volta, Eastern, Asanti, and Brong Ahafo.

Notes:

↵* p < 0.1,

↵** p < 0.05,

↵*** p < 0.01.

Confidence intervals clustered at community levels in squared brackets. R-W p-values were adjusted for multiple testing using the Romano–Wolf (2005, 2016) step-down method with 2,000 iterations and standard errors clustered at community level. The table above presents intent-to-treat effects on each outcome for the full sample and stratified by child gender, household poverty, and place of residence. Models were estimated through OLS. For each outcome, the model controls for its baseline value, a dichotomous variable related to the randomized assignment to school feeding, and region dummies. Enrollment is a dichotomous variable indicating whether the child is enrolled to any level of education; attendance is an indicator counting the number of days the child attended by the child in the past school week. The indicator ranges from zero to five days. Current grade provides the educational grade (in years) the child is currently enrolled in. Household poverty is a dichotomous indicator having the value of one if the household had baseline per capita consumption levels falling below the national consumption poverty fine in 2013. Northern regions include Upper West, Upper East, and Northern region. Southern regions include Western, Central, Greater Accra, Volta, Eastern, Asanti, and Brong Ahafo.