Policy relevance of causal quantities in networks

2025-07-18

AI Summary

In network interventions, interference between units—where treatment assignment to one unit affects others’ outcomes—renders conventional causal estimands (e.g., the average treatment effect) policy-irrelevant. Method: Grounded in the potential outcomes framework, this paper systematically analyzes the policy interpretability of various causal estimands in network settings and proposes two policy-oriented average treatment allocation strategies: population-level aggregation and cross-policy comparison. Contribution/Results: We establish that the “expected average outcome” under a given policy—and its differences across policies—offers greater decision relevance than aggregated individual-level treatment effects. Moreover, we identify several causal quantities that resist interpretation as individual effects yet yield robust optimal allocation guidance through policy comparisons. The work introduces a decision-centric metric system for causal policy evaluation in networked environments, significantly enhancing the practical utility of causal inference in real-world interventions.

Technology Category

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

Application Category

• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Social Networks and Social Media: Media and governance, opinion dynamics, filter bubbles, polarization

Abstract

In settings where units' outcomes are affected by others' treatments, there has been a proliferation of ways to quantify effects of treatments on outcomes. Here we describe how many proposed estimands can be represented as involving one of two ways of averaging over units and treatment assignments. The more common representation often results in quantities that are irrelevant, or at least insufficient, for optimal choice of policies governing treatment assignment. The other representation often yields quantities that lack an interpretation as summaries of unit-level effects, but that we argue may still be relevant to policy choice. Among various estimands, the expected average outcome -- or its contrast between two different policies -- can be represented both ways and, we argue, merits further attention.

Problem

• Quantifying treatment effects in networked settings
• Identifying policy-relevant causal estimands
• Comparing representations for optimal policy choice

Innovation

• Averages over units and treatment assignments
• Focus on expected average outcome contrasts
• Policy relevance over unit-level effects