Estimating Total Effects in Bipartite Experiments with Spillovers and Partial Eligibility

2025-11-14

AI Summary

This paper addresses causal effect estimation in bilateral systems where some units lack treatment eligibility yet remain subject to interference. Recognizing that conventional methods—by ignoring interference—yield biased estimates and even sign reversals, we formally introduce the “partially admissible” bipartite experiment design. We propose two novel causal estimands: the Primary Total Treatment Effect (PTTE), capturing the aggregate impact on eligible units, and the Secondary Total Treatment Effect (STTE), quantifying the net effect on ineligible units. Methodologically, we develop an interference-aware ensemble estimator: leveraging exposure mappings and generalized propensity scores, we employ projection mapping under linear edge assumptions to precisely link treatment-side exposures to outcome-side responses, and incorporate a deterministic aggregation scheme to enhance estimation efficiency for sparse treatment-side data. Simulation and real-world field experiments demonstrate that our framework substantially reduces both bias and variance, effectively corrects pre-specified metric distortions induced by interference, and—in practical applications—reverses both statistical significance and sign of key decision-relevant metrics.

Technology Category

• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning
• Multiagent Systems: Mechanism Design

Application Category

• Security and Privacy: Large-scale security measurements
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

We study randomized experiments in bipartite systems where only a subset of treatment-side units are eligible for assignment while all units continue to interact, generating interference. We formalize eligibility-constrained bipartite experiments and define estimands aligned with full deployment: the Primary Total Treatment Effect (PTTE) on eligible units and the Secondary Total Treatment Effect (STTE) on ineligible units. Under randomization within the eligible set, we give identification conditions and develop interference-aware ensemble estimators that combine exposure mappings, generalized propensity scores, and flexible machine learning. We further introduce a projection that links treatment- and outcome-level estimands; this mapping is exact under a Linear Additive Edges condition and enables estimation on the (typically much smaller) treatment side with deterministic aggregation to outcomes. In simulations with known ground truth across realistic exposure regimes, the proposed estimators recover PTTE and STTE with low bias and variance and reduce the bias that could arise when interference is ignored. Two field experiments illustrate practical relevance: our method corrects the direction of expected interference bias for a pre-specified metric in both studies and reverses the sign and significance of the primary decision metric in one case.

Problem

• Estimating causal effects in bipartite experiments with spillovers and partial eligibility
• Developing interference-aware estimators combining exposure mappings and machine learning
• Addressing bias from interference when only subset of units receives treatment

Innovation

• Interference-aware ensemble estimators combining exposure mappings
• Generalized propensity scores with flexible machine learning
• Projection linking treatment- and outcome-level estimands