ABSTRACT
Chapter 12 presents mediation analysis, which addresses questions about the mechanism of a causal effect. For example, does a new treatment act by increasing an intermediate variable, which in turn improves the outcome? It introduces expanded potential outcomes based on a treatment and a mediator, and uses them to partition the total causal effect into either a controlled direct and indirect effects or natural direct and indirect effects. It also defines the proportion explained and the proportion mediated. It specifies assumptions under which different parametric and nonparametric methods validly estimate these causal quantities, including the traditional product and difference methods, and it proves their validity. Examples and R code are also provided.
