Randomizing under non-response and non-compliance : Evidence from school meals experiment in rural Senegal

School meals interventions have been introduced as an important policy tool to improve education outcomes and pupil’s nutritional status. This study uses a unique and largescale randomized field experiment to assess the effectiveness of such programmes on pupils’s performance (test scores in French, mathematics and the aggregate) and on the internal efficiency of schools (enrollment, promotion, repetition and dropout) in rural Senegal. We show that attrition and non-compliance occurred not at random in the experiment, calling for a framework that embeds both issues. Relying on the average treatment effect and the complier effects, we find that the programme has a positive and significant impact on pupils’ scores and on the enrollment rate. However, the repetition rate increased. The intervention has a marked gender effect. Cost-effectiveness analysis shows that deworming intervention is more cost-effective than school meals.