jump to main area
:::
A- A A+

Seminars

Causal Mediation Analysis Without Sequential Ignorability

  • 2013-12-24 (Tue.), 10:30 AM
  • Recreation Hall, 2F, Institute of Statistical Science
  • Prof. Xiao-Hua (Andrew) Zhou
  • Dept. of Biostatistics, Univ. of Washington and Director of Biostatistics Unit, VA Seattle Medical Center

Abstract

Mediation analysis is an important tool in social and behavioral sciences as it helps to understand why a behavioral intervention works. The commonly used approach given by Baron and Kenny requires the strong assumption “sequential ignorability” to yield causal interpretation. Ten have proposed a rank preserving model to relax this assumption. However, the RPM is restricted to the case with binary intervention and single mediator. Also, it needs another strong assumption, “rank preserving”. We proposed a new method that could relax this assumption and handle a multi-level intervention and a multi-component mediator. As an estimating equation based method, our method can also handle correlated data with the generalized estimating equation and handle missing data with an inverse probability weighting. Finally our method can also be used in many other research settings, which have a similar model as mediation analysis such as treatment compliance, post randomized treatment component analysis. For the proposed causal mediation model, we first showed identifiability for the parameters in the model. We then proposed a semi-parametric method for estimating the model parameters and derived the asymptotic results for the proposed estimators. Simulation showed the good performance of the proposed estimators in finite sample sizes. Finally we applied the proposed method to one real-world clinical study on the college student drinking study.?

Update:2024-12-02 22:26
scroll to top