|
Abstract:
|
From a causal inference perspective, the typical assumption of no interference becomes untenable in experiments in a social context. In many instances, however, the patterns of interference may be informed by the observation of network connections among the units of analysis. We develop elements of optimal estimation theory for causal effects by leveraging an observed network. Considering the class of linear unbiased estimators of the average direct treatment effect under various exclusion restrictions for the potential outcomes, we offer analytical insights on the weights that lead to minimum integrated variance estimators. These estimators offer superior performance to previously proposed estimators and we seek to develop complementary variance estimators for these estimates based on similar principles.
|
