To identify the poorest of the poor, researchers collected data on nearly 26,000 households in Assiut and Sohag, and utilized a proxy means test that uses observable assets to predict poverty. Households that were assessed to be not poor enough to participate based on basic criteria were then filtered out, and, among the remainder, community leaders and representatives were asked to identify the poorest half of households in their communities. The identified households were surveyed using a more detailed questionnaire to further screen the sample. Based on this process, researchers determined the final sample of 3,469 ultra-poor households.
Researchers randomly assigned agglomerations (subdivision of a village) into four groups : a standard graduation program, a low cost graduation program, a female-targeted program while the comparison group did not receive any interventions.