Decomposition of the total effect for two mediators: A natural mediated interaction effect framework

Area of Science:

• Causal inference
• Statistical modeling
• Biostatistics

Background:

• Mediation analysis explains relationships via mediators.
• Decomposing total effects into mediation and interaction is increasingly important.
• Existing methods lack comprehensive interaction analysis for causally ordered mediators.

Purpose of the Study:

• To develop a unified framework for decomposing total effects in mediation analysis.
• To introduce the natural mediated interaction (MI) effect for two-mediator scenarios.
• To account for causally sequential and non-sequential mediators with interactions.

Main Methods:

• Developed counterfactual framework for mediation decomposition.
• Proposed natural mediated interaction (MI) effect for two-way and three-way interactions.
• Unified approach to decompose total effect into mediation, interaction, or both/neither.

Main Results:

• The proposed method decomposes total effects into distinct components: mediation only, interaction only, both, or neither.
• Natural mediated interaction (MI) effect extends existing two-way MI.
• Demonstrated effectiveness via simulation and real-data analysis for sequential and non-sequential mediators.

Conclusions:

• The novel decomposition framework provides a comprehensive understanding of mediation and interaction effects.
• The natural mediated interaction (MI) effect offers a valuable tool for analyzing complex causal pathways.
• This approach enhances mediation analysis in various scientific disciplines.