Table 1.
Aspect | SEM approach | PS approach |
---|---|---|
Key modeling difference | Focus on observed mediator status | Focus on potential mediator status |
Key strength | Compatibility with the mediational process theory | Clarity in underlying assumptions necessary for causal interpretation |
Frequent context | Very general, theory-based mediation | Treatment noncompliance |
Fixed assumptions for causal interpretation | Ignorability of mediator status | None |
Frequently used assumptions for causal interpretation | Ignorability, constant effect, linearity | Monotonicity, exclusion restriction |
Key method of checking validity of causal effect estimates | Test the fit of the model to the data | Test sensitivity of estimates to violating underlying assumptions |
Note. SEM = structural equation modeling; PS = principal stratification.