New Estimands for Experiments with Strong Interference

TL;DR

This paper introduces new estimands for experiments with interference, enabling causal inference without strong assumptions and providing insights into how treatment effects vary with exposure levels.

Contribution

It proposes novel estimands for binary outcomes that can be estimated under minimal assumptions, expanding causal inference in social experiments with interference.

Findings

Estimands can be estimated without assumptions beyond randomization.

They reveal systematic variation of treatment effects with exposure.

They provide lower bounds on the number of affected units.

Abstract

In experiments that study social phenomena, such as peer influence or herd immunity, the treatment of one unit may influence the outcomes of others. Such "interference between units" violates traditional approaches for causal inference, so that additional assumptions are often imposed to model or limit the underlying social mechanism. For binary outcomes, we propose new estimands that can be estimated without such assumptions, allowing for interval estimates assuming only the randomization of treatment. However, the causal implications of these estimands are more limited than those attainable under stronger assumptions, showing only that the treatment effects under the observed assignment varied systematically as a function of each unit's direct and indirect exposure, while also lower bounding the number of units affected.