ABSTRACT

This paper examines a new set of average cost data from a large international NGO, finding that costs for the same intervention can vary as much as twenty times when scale or context is changed. Despite this challenge to the generalisability of cost estimates, a high proportion of the variation can be explained by observable program and contextual characteristics. Binary questions about whether cost estimates are externally valid do not provide a useful framework for wider inference; instead, researchers can gain analytical traction if they study what factors cause the costs of specific interventions to change, and by how much.