![\"\"](\"\/cise\/files\/2022\/07\/Yuan-Yuan-453x636.jpeg\")
Yuan Yuan<\/span><\/strong>\nAssistant Professor<\/span><\/strong>
\nPurdue University<\/span><\/strong><\/h3>\n
Using Causal Network Motifs to Characterize Network Interference in Randomized Controlled Experiments<\/strong><\/p>\n Randomized control trials, or \u201cA\/B tests\u201d, have been crucial to understanding the impact of an intervention. Traditional causal inference methods rely on a critical assumption called the \u201cstable unit treatment value assumption\u201d (SUTVA), which assumes that a user\u2019s treatment response only depends on only their treatment assignment. However, SUTVA is an unrealistic assumption in settings such as networks, where a unit\u2019s response may be affected by other users. This violation of SUTVA is referred to as network interference. The current literature on network interference has two major limitations \u2014 failing to account for social theories of interference (e.g. structural diversity or social contagion) and relying on human experts to model interference patterns. To tackle these issues, we propose a two-part machine learning approach to automatically characterize network interference conditions based on both the local network structures and the treatment assignment among users in the network neighborhood. Specifically, we first construct network motifs with treatment assignment information, referred to as causal network motifs, to characterize the network interference conditions for each unit. We then develop machine learning methods based on decision trees and nearest neighbors to map these causal network motif representations to an \u201cexposure condition\u201d proposed by Aronow and Samii (2017). We demonstrate the validity of our approach on simulated A\/B tests on real-world networks and a re-analysis from a prior experiment study, which show how causal network motifs are able to more accurately account for complex interference patterns and reduce bias for treatment effect estimation with the presence of interference.<\/span><\/p>\n