Designing experiments to measure spillover effects

This paper formalizes the design of experiments intended specifically to study spillover effects. By first randomizing the intensity of treatment within clusters and then randomly assigning individual treatment conditional on this cluster-level intensity, a novel set of treatment effects can be identified. The paper develops a formal framework for consistent estimation of these effects, provides explicit expressions for power calculations, and shows that the power to detect average treatment effects declines precisely with the quantity that identifies the novel treatment effects. A demonstration of the technique is provided using a cash transfer program in Malawi.