In the past 40 years, statisticians have been developing a principled framework for making inferences about causal relationships and estimating "causal'' effects. Scientists are typically interested not only in estimating the effect of a cause on an outcome, but also in saying something about how the cause produces the effect through one or more pathways. Thus, statisticians have extended the basic framework for causal inference to the topic of mediation. In this talk, I will introduce the potential outcomes notation and use this to define causal estimands and give conditions under which observed data can be used to identify these estimands. Then, I will extend this to the case where there is a mediating variable. Various estimands are then defined and conditions under which observed data can be used to identify these will be given. I will focus on the case where both the mediator and the outcome are continuous. Other cases and how these are handled will also be discussed briefly.

海报