Abstract:
|
Mediation analysis typically examines the undirected effect from the intervention to the outcome through a so-called mediator variable. While the traditional mediation model is applies to static data, many real-life settings record information repeatedly over time. We extended the mediation model with binary outcome to accommodate the longitudinal data. In the generalized model, not only the mediator and the outcome variables are allowed to change over time, the mediation effect can also be time-varying. Both the methodology and a simulation-based approach for inference will be introduced. The newly proposed method is also tested on various simulation settings, as well on a smoking cessation study.
|