|
Abstract:
|
This article introduces a new randomization method for conducting networked experiments. Balanced randomization configuration is one of the most important concerns for successful comparative studies. However, chance imbalance still exists in many traditional experiments using the complete randomization. Under the assumption of random dot product graphs, the proposed method generates more balanced assignments of the units to the treatment and control groups compared with the traditional randomization methods. The proposed method can be used for directed or undirected graphs with or without vertex covariate information. In addition, under the linear regression setting, the proposed method achieves the optimal randomization balance asymptotically in the sense that the estimated treatment effect attains its minimum variance. Numerical studies provide further evidence of the advantages of the proposed method.
|
