Causal Estimands and Multiply Robust Estimation of Mediated-Moderation

Area of Science:

• Social Sciences
• Psychology
• Biostatistics

Background:

• Effect heterogeneity (moderation) is often studied alongside underlying mediation mechanisms (mediated moderation).
• Traditional methods for mediated moderation rely on parametric models, posing challenges for causal inference and potential misspecification.
• Causal mediation analysis is established but lacks development for mediated moderation.

Purpose of the Study:

• To extend causal mediation analysis to address mediated moderation.
• To propose a novel method for estimating mediated moderation using the potential outcomes framework.
• To provide a multiply robust estimation approach incorporating machine learning.

Main Methods:

• Developed two causal estimands to decompose total moderation into mediated and unexplained components.
• Utilized the potential outcomes framework for causal interpretation.
• Proposed a multiply robust estimation method for mediated moderation analysis.

Main Results:

• The proposed method successfully decomposes total moderation.
• The approach allows for causal interpretation of mediated moderation.
• Simulations demonstrated the method's validity.

Conclusions:

• The novel method advances causal mediation analysis for mediated moderation.
• This approach offers a robust alternative to traditional parametric methods.
• It enables a deeper understanding of mediation mechanisms underlying subgroup differences.