Peer effects and global treatments: Design and analysis of experiments in networks - Dean Eckles

Dean Eckles, a Social Scientist and Statistician working on the Facebook Data Science Team

Abstract

Peer effects (i.e., social interactions, interference, social influence, spillovers) are common in many settings of interest to social scientists, system designers, and policy-makers. Experimenters may wish to estimate these peer effects themselves and/or estimate what would happen under a global (i.e., network-wide) treatment that may function largely through peer effects.

For estimating peer effects, we use experimental designs that manipulate one or more mechanisms by which peer effects occur. Examples include social influence in online advertising and information diffusion via Facebook News Feed.

For estimating effects of global treatments, estimates from simple random assignment and analysis ignoring peer effects can suffer from substantial bias. We use experimental designs that reduce bias by producing treatment assignments that are correlated in the network. For example, experimenters can use graph partitioning methods to construct clusters of individuals who are then assigned to treatment or control together. This clustered assignment alone can substantially reduce bias, as can incorporating information about peers' treatment assignments or behaviors into the analysis. Simulation results show how this bias reduction varies with network structure and the size of direct and peer effects. We illustrate this method with a real experiment on Facebook conducted on Thanksgiving Day 2012.

This covers joint work with Eytan Bakshy, Brian Karrer, and Johan Ugander.