Design and Analysis of Experiments on Networks

Abstract

In the design and analysis of experiments, it is often assumed that experimental units are independent, in the sense that the treatment assigned to one unit will not affect the potential outcome of another unit. However, this assumption may not hold if the experiment is conducted on a network of experimental units. The treatment assignment of one unit can spread to its neighbors via their network connections. The growing popularity of online experiments conducted on social networks calls for more research on this topic. We investigate the problem of experiments on networks and propose new approaches to both the design and analysis of such experiments.