Efficient Semiparametric Estimation of Network Treatment Effects Under Partial Interference

TL;DR

This paper develops a framework to derive the most efficient estimators for network treatment effects under partial interference, improving causal inference in networked settings.

Contribution

It introduces a semiparametric efficiency framework for network causal effects, showing existing estimators are locally efficient and proposing new efficient estimators.

Findings

Existing estimator by Liu et al. (2019) is locally efficient.

New efficient estimators are proposed and analyzed.

Application to Colombian cash transfer programs demonstrates practical utility.

Abstract

Recently, many estimators for network treatment effects have been proposed. But, their optimality properties in terms of semiparametric efficiency have yet to be resolved. We present a simple, yet flexible asymptotic framework to derive the efficient influence function and the semiparametric efficiency lower bound for a family of network causal effects under partial interference. An important corollary of our results is that one of the existing estimators by Liu et al. (2019) is locally efficient. We also present other estimators that are efficient and discuss results on adaptive estimation. We conclude by using the efficient estimators to study the direct and spillover effects of conditional cash transfer programs in Colombia.