🔍 Research Problem
Many studies of causal mechanisms in survey experiments estimate indirect effects by conditioning on nonrandomized variables. That approach relies on a “selection-on-observables” assumption, which undermines the central advantage of random assignment and raises concerns about the validity of mediation claims.
🧪 How the Design Reveals Mechanisms
A factorial experimental design is proposed that randomly provides or withholds information about potential mediators. Key features:
- Randomly manipulate both the primary treatment and whether subjects receive information about a putative mediator.
- Identify the overall average treatment effect (ATE) through the randomized treatment.
- Identify the controlled direct effect of the treatment when fixing the mediator via the information-manipulation arm.
- Do not identify indirect effects on their own without additional assumptions.
📊 Key Advantages and Limits
- Avoids the selection-on-observables assumption required by standard mediation analyses that condition on nonrandomized mediators.
- Preserves the inferential benefits of randomization while furnishing evidence about how treatment operates.
- Does not by itself recover indirect effects, but enables a broader investigation of mechanisms that captures both indirect pathways and treatment–mediator interactions.
📚 Illustrative Applications
Examples demonstrate the approach in two substantive domains:
- Evaluations of U.S. Supreme Court nominees
- Perceptions of the democratic peace
⚖️ Why It Matters
This design offers researchers a principled way to learn about causal mechanisms using experimental design rather than relying on untestable selection-on-observables assumptions. It informs when and how survey experiments can credibly speak to direct effects and to more complex mechanism questions that involve interactions and indirect pathways.
