Causal inference in randomized experiments with mediational processes.

Area of Science:

• Statistics
• Causal Inference
• Biostatistics

Background:

• Structural Equation Modeling (SEM) and Principal Stratification (PS) are established methods for analyzing intermediate outcomes in randomized experiments.
• These approaches have historically developed independently, limiting potential synergistic insights.

Purpose of the Study:

• To bridge the gap between SEM and PS methodologies in causal inference.
• To introduce a Cross-Model Translation (CMT) approach for parameter exchange between SEM and PS models.

Main Methods:

• Developed the Cross-Model Translation (CMT) approach to translate parameter estimates between PS and SEM models.
• Utilized Monte Carlo simulations to examine the relationship between PS and SEM under specific identifying assumptions.

Main Results:

• Parameter translation between PS and SEM is feasible based on their conceptual links, even without specific identifying assumptions.
• Simulations clarified the interplay between the two approaches under various identifying assumptions.

Conclusions:

• The choice of identifying assumptions is more critical for practical causal inference than the specific modeling framework (SEM or PS).
• The CMT approach offers a unified framework for jointly considering SEM and PS, highlighting shared inferential challenges.