Comparing inference methods for causal mediation analysis with nominal mediators: A simulation and empirical study

Area of Science:

• Social Sciences
• Behavioral Sciences
• Health Sciences

Background:

• Causal mediation analysis is crucial for understanding complex relationships.
• Existing methods often struggle with nominal (categorical) mediators.
• A robust framework is needed for categorical causal pathways.

Purpose of the Study:

• To advance causal mediation analysis for nominal mediators using a potential outcome framework.
• To compare the performance of three inference methods for testing natural indirect effects.
• To provide a reliable methodological framework for researchers.

Main Methods:

• Monte Carlo simulations were employed to compare inference methods.
• Three methods were assessed: non-parametric bootstrapping, parametric resampling, and Bayesian estimation.
• Nominal mediation models were utilized within a potential outcome framework.

Main Results:

• Nominal mediation models produced accurate estimates across various conditions.
• Maximum likelihood and Bayesian approaches demonstrated strong performance.
• All three inference methods showed acceptable Type I error control and adequate statistical power in larger samples.

Conclusions:

• Researchers can reliably estimate and test nominal mediation effects using bootstrapping, parametric resampling, or Bayesian estimation.
• The study provides a robust methodological framework for causal pathways with categorical variables.
• Findings support the utility of these methods in social, behavioral, and health sciences.