Statistics Colloquium by Ryan Cox '20, Wednesday, October 2

Time-Varying Mediation Models: Inference for the Mediation Effect Using a Simulation-Based Approach by Ryan Cox ’20, Wednesday, October 2, 1 :10 – 1:50 pm, Stetson Court Classroom 105, Statistics Colloquium

Abstract:  Mediation models are a commonly used approach to study the causal relationship between an independent and dependent variable using a third variable called a mediator.  In traditional mediation models, the independent variable affects the mediator, which in turn influences the dependent variable.  This indirect relationship is called the mediation effect.  While the traditional mediation model only applies to static data, many real-life settings record data repeatedly over time.  To address this problem, we extended the mediation model with binary outcome to the time-varying setting.  In this talk I will introduce both the traditional and modified models, discuss the method behind the mediation effect estimation, and show a simulation-based approach to constructing point wise confidence bands for the mediation effect.  I will also show preliminary results from a smoking cessation study.