Primary Submission Category: Causal Inference in Networks
Causal Inference in Dynamic Networks
Authors: Peem Lerdputtipongporn, David Choi, Nynke Niezink,
Presenting Author: Peem Lerdputtipongporn*
For longitudinal network settings, this study introduces a novel methodology for evaluating the effects of dyadic interventions (i.e., intervening on a network connection). The approach can be applied to either dyadic-level, such as whether the presence of a connection causes the tie to persist or be reciprocated in subsequent periods, or to individual-level outcomes, such as whether the presence of a connection between two individuals causes their behaviors in subsequent periods to become more similar. The methodology employs a doubly-robust estimation technique that combines low-rank matrix approximation with longitudinal network models, for which we establish unbiasedness and consistency under certain conditions.