More powerful cluster randomized control trials

Abstract

Balanced experimental designs, in which the number of treatment and control units are the same, do not maximize power subject to a cost constraint when treatment units are more expensive than control ones. Despite this, such balanced designs are the norm in economics. This paper describes methods to optimally choose the number of treatment and control clusters, and the number of units within treatment and control clusters, allowing for full flexibility. We use three archetypal examples from the development literature to illustrate the magnitude of the power gains, which lie between 8.5 and 19 percentage points.