233 – Section on Health Policy Statistics Student Paper Competition Winners
Causal Mediation Analysis in Multi-Level Nonrandomized Exposure and Multi-Component Mediator Case
Cheng Zheng
University of Washington
Xiao-Hua Zhou (Andrew)
University of Washington
Mediation analysis is an important issue in social and behavirol sciences as it helps to understand why a behavioral intervention works. To yield a causal interpretation the most common approach (e.g., Baron and Kenny, 1986), as discussed by (Imai, Keele, and Tingley 2010), need an often-unrealistic assumption of "sequential ignorability". Rank preserving model (RPM; Ten Have et al., 2007) is proposed to relax this assumption. However, RPM is restricted to the case with binary intervention and single mediator. Also, it needs the strong "rank preserve" assumption. We propose a new model that can handle multi-level intervention and multi-component mediator with weaker assumption. Also, our model has the ability to handle correlated data and missing data. And our method can be used in many different research questions, such as treatment compliance.
