Peer Encouragement Designs in Causal Inference with Partial Interference and Identification of Local Average Network Effects

TL;DR

This paper introduces peer encouragement designs for causal inference in networks, enabling the study of treatment effects without perfect compliance, and defines new estimands called local average network effects.

Contribution

It proposes a novel peer encouragement design for network experiments that relaxes the need for perfect compliance and identifies new causal estimands.

Findings

Point-identification of familiar estimands in encouragement designs

Analysis of non-compliance effects in network experiments

Definition of local average network effects

Abstract

In non-network settings, encouragement designs have been widely used to analyze causal effects of a treatment, policy, or intervention on an outcome of interest when randomizing the treatment was considered impractical or when compliance to treatment cannot be perfectly enforced. Unfortunately, such questions related to treatment compliance have received less attention in network settings and the most well-studied experimental design in networks, the two-stage randomization design, requires perfect compliance with treatment. The paper proposes a new experimental design called peer encouragement design to study network treatment effects when enforcing treatment randomization is not feasible. The key idea in peer encouragement design is the idea of personalized encouragement, which allows point-identification of familiar estimands in the encouragement design literature. The paper also defines new causal estimands, local average network effects, that can be identified under the new design and analyzes the effect of non-compliance behavior in randomized experiments on networks.