Heterogeneous Treatment Effects in Social Networks

Abstract

We study treatment effect modifiers for causal analysis in a social network, where neighbors' characteristics or network structure may affect the outcome of a unit, and the goal is to identify subpopulations with varying treatment effects using such network properties. We propose a novel framework called Testing-for-Effect-Modifier (TEEM) for this purpose that facilitates data-driven decision making by testing hypotheses about complex effect modifiers in terms of network features or network patterns (e.g., characteristics of neighbors of a unit or belonging to a triangle), and by identifying sub-populations for which a treatment is likely to be effective or harmful. We describe a hypothesis testing approach that accounts for unit's covariates, their neighbors' covariates and patterns in the social network, and devise an algorithm incorporating ideas from causal inference, hypothesis testing, and graph theory to verify a hypothesized effect modifier. We perform extensive experimental evaluations with a real development economics dataset about the treatment effect of belonging to a financial support network called self-help groups on risk tolerance, and also with a synthetic dataset with known ground truths simulating a vaccine efficacy trial, to evaluate our framework and algorithms. * Both authors contributed equally to this research. 2 Rubin's potential outcome framework [51], which is a prominent approach for causal inference in statistical studies, relies on the "Stable Unit Treatment Value Assumption (SUTVA)", which includes the assumption of no-interference.

Feature Selection for Discovering Distributional Treatment Effect Modifiers

“…The Causal Rule Ensemble method [Lee et al, 2020] seeks the important features by learning a rule-based model that emulates the input-output relationship of a fitted treatment effect estimation model. Gilad et al [2021] considered a hypothesis test for discovering the treatment effect modifiers from social network data. However, none of these methods can find the features related to distributional treatment effect heterogeneity because they are also based on the average treatment effect and cannot find the features related to other functionals of the joint distribution of potential outcomes.…”