A brief primer on conducting regression-based causal mediation analysis.

Area of Science:

• Psychology
• Epidemiology
• Biostatistics

Background:

• Mediation analysis is crucial for understanding complex relationships in traumatic stress research.
• Traditional mediation methods often fail to account for crucial interactions between exposure and mediator variables.
• Causal mediation analysis provides a more robust framework for estimating direct and indirect effects.

Purpose of the Study:

• To provide an overview of regression-based causal mediation analysis in traumatic stress research.
• To offer guidance on conducting mediation analysis using the R package regmedint.
• To highlight the importance of accounting for exposure-mediator interactions.

Main Methods:

• Discussed causal interpretations of total, direct, and indirect effects, particularly with exposure-mediator interactions.
• Outlined assumptions for valid mediation analysis and suitable study designs.
• Illustrated mediation analysis using the regmedint R package with longitudinal data from the COVID-19 lockdown.

Main Results:

• Traditional methods can yield different direct and indirect effect estimates when exposure and mediator interact.
• Causal mediation methods provide more accurate estimates by explicitly allowing for interactions.
• When no interaction is present, results from traditional and causal methods may converge depending on model specification.

Conclusions:

• Regression-based causal mediation methods estimate specific interventional quantities, not just associations.
• Causal methods explicitly accommodate exposure-mediator interactions, offering greater flexibility.
• Recommended adopting causal mediation methods over traditional approaches due to their accuracy and explicit handling of interactions.