Compare Mediation Effects Without Assuming No Confounding — An Audience Costs Case

🧭 What Was Compared

This study introduces a way to compare the causal "anatomies" of multiple treatments by focusing on how different treatments work through a common mediator. The core contribution is a set of comparative causal mediation (CCM) estimands that directly compare mediation effects across treatments rather than estimating each mediation effect in isolation.

🛠️ What The New Method Does

The paper develops estimators for the CCM estimands and derives their statistical properties. Crucially, these estimators are shown to be consistent (or conservative) under a set of assumptions that do not require the usual—and typically untestable—assumption of no unobserved confounding between the mediator and the outcome. Key features include:

• Comparative causal mediation (CCM) estimands that contrast mediation effects of different treatments via a shared mediator
• Estimators whose consistency or conservativeness holds even when mediator–outcome confounding may be present
• Formal derivations of estimator properties and the assumptions needed for valid inference

🔎 How the Approach Is Applied

An original application assesses whether and how the international legal status of a foreign policy commitment affects domestic "audience costs"—the political penalties democratic governments face for violating commitments. The CCM framework is used to trace whether legalization increases audience costs and to identify the causal channels at work.

📌 Key Findings

• International legalization can enhance audience costs.
• The increase operates through multiple causal channels rather than a single pathway.
• One identified channel is amplified perceived immorality of violating the commitment, which mediates part of the effect.

⚖️ Why It Matters

This method allows researchers running multi-treatment experiments to compare how treatments operate without relying on the strong, nonrefutable assumption that there is no unobserved mediator–outcome confounding. The approach improves causal mediation analysis in experimental and observational settings and yields substantive insights into how international law shapes democratic accountability.