Multiple Randomization Designs: Estimation and Inference with Interference

TL;DR

This paper introduces new randomized experimental designs and estimands to effectively measure complex spillover effects in settings with multiple interacting populations, extending traditional methods.

Contribution

It develops novel experimental designs and estimators that account for strategic interactions and spillovers, with theoretical properties and inference methods.

Findings

Derived finite-sample properties of estimators

Established central limit theorems for estimators

Proposed designs improve measurement of spillover effects

Abstract

Classical designs of randomized experiments, going back to Fisher and Neyman in the 1930s still dominate practice even in online experimentation. However, such designs are of limited value for answering standard questions in settings, common in marketplaces, where multiple populations of agents interact strategically, leading to complex patterns of spillover effects. In this paper, we discuss new experimental designs and corresponding estimands to account for and capture these complex spillovers. We derive the finite-sample properties of tractable estimators for main effects, direct effects, and spillovers, and present associated central limit theorems.