Clustered Switchback Designs for Experimentation Under Spatio-temporal Interference
The paper proposes a clustered switchback design to estimate global average treatment effects when experiments have both spatial interference (units influencing neighbors) and temporal interference (time steps affecting each other). The method groups units into clusters and time steps into blocks, assigning one treatment to each whole cluster-block combination. This design allows for estimating the effect under these complex conditions with a low mean squared error.