Testing for arbitrary interference on experimentation platforms∗

Experimentation platforms are essential components of modern large IT companies, as they are used to carry out a large number of randomized experiments daily. On such platforms, the classic assumption of no interference among users—that is, the fact that the outcome of a user does not depend on the treatment assigned to other users—is rarely tenable. Here, we introduce an experimental design strategy for testing whether this assumption holds. Our approach is in the spirit of the Durbin-Wu-Hausman test for endogeneity in econometrics, where multiple estimators return the same estimate if and only if the null hypothesis holds. The design that we introduce makes no assumptions on the interference model between units, nor on the network among the units, and comes with a sharp bound on the variance, and an implied analytical bound on the type I error rate. The idea is to obtain two different estimates of the total treatment effect leveraging two randomization strategies—namely, complete randomization and clusterbased randomization—and to analyze the distribution of their difference. We propose a multilevel assignment strategy for obtaining these two estimates simultaneously, and we develop theoretical guarantees for rejecting the null hypothesis that no interference holds, regardless of the type of interference. We discuss how to apply the proposed design strategy to large experimentation platforms, and we illustrate it in the context of a live experiment on the LinkedIn experimentation platform.