Presentation: Estimation of Causal Effects Among Network-Dependent Units
Speaker: Oleg Sofrygin, PhD, Postdoctoral Research Fellow, Group in Biostatistics, University of California, Berkeley
Abstract: Consider an observational study in which data are collected on subjects linked by social network ties. Our goal is to identify and estimate the average causal effect of treatment in settings where one individual’s treatment assignment may affect the outcomes of his or her social contacts. Our prior work introduced a semi-parametric estimation approach that allows drawing statistical inference in network-dependent settings. One important limitation of that work was the requisite that each unit is independent of all, but a fixed number of other units. Our inference relied on an asymptotic regime which assumed that the size of each unit’s dependence-set is fixed, even as the total number of units in the network grows to infinity. However, this assumption is known to be violated in many realistic network models. In this work we discuss approaches to valid inference when the assumption is violated. In particular, we prove the CLT under the asymptotic regime whereby the size of each unit’s dependence-set grows to infinity. We then apply our estimation framework under the latter larger model.