Statistical Designs for Network A/B Testing

A/B testing is a common controlled experiment approach used to compare two versions of internet-based products. IT companies often conduct A/B testing on their users who are connected in a social network. The users’ responses could be related to the network connection, leading to two typical assumptions: network interference and network correlated outcomes. I will discuss the general problems of network A/B testing design and their relationship with graph cut objectives under the two assumptions. Further, I will show the asymptotic distributions of graph cut objectives to enable rerandomization algorithms for the design of network A/B testing.