Statistical Mediation Analysis for Models with a Binary Mediator and a Binary Outcome: the Differences Between Causal...

Area of Science:

• Statistics
• Prevention Research
• Causal Inference

Background:

• Mediation analysis is crucial for identifying effective intervention components in prevention research.
• Traditional mediation analysis relies on linear regression coefficients, with unclear estimation for binary variables.
• Causal mediation analysis offers a robust framework using potential outcomes, applicable to complex models.

Purpose of the Study:

• To clarify similarities and differences between causal and traditional mediation effect estimates.
• To evaluate these methods in mediation models with binary mediators and outcomes.
• To demonstrate the application using an empirical example.

Main Methods:

• Applied traditional mediation analysis using linear regression coefficients.
• Applied causal mediation analysis using potential outcome definitions.
• Compared effect estimates from both methods on a binary mediator/binary outcome model.

Main Results:

• Causal and traditional methods yielded similar controlled direct effect estimates.
• Estimates for natural direct effects, natural indirect effects, and total effects differed significantly.
• Traditional methods showed limitations in generalizing to binary variable mediation models.

Conclusions:

• Causal mediation analysis is the preferred method for mediation models involving binary variables.
• The natural effect definitions in causal mediation analysis are broadly applicable.
• This framework enhances the precision of intervention component analysis in prevention research.