|
Abstract:
|
Mediation analysis detects the indirect effects via a third variable (mediator) on the pathway between exposures and outcomes. Multiple mediators usually form a chain of indirect effects called serial mediations. In complex surveys that often use stratified multistage sampling, the conventional mediation analysis approaches assuming simple random sampling are incorrect. This study introduced five new estimators to address this issue: Balanced repeated replication (BRR), Fay's method, Jackknife repeated replication (JRR), bootstrap, and Taylor series linear approximation (TSL). The preliminary Monte Carlo simulations showed that the conventional estimation had severely inflated Type-I error rates and inadequate confidence-interval coverage. The proposed estimators maintained the Type-I error control across most conditions. JRR and TSL demonstrated the highest power, followed by bootstrap, BRR, and Fay's method. Fay's method and bootstrap provided the highest confidence-interval coverage rate, followed by BRR, JRR, and TSL. The proposed estimators are applied to National Health Interview Surveys to detect the serial mediations between smoking and health status among cancer survivors.
|
