ABSTRACT
This chapter describes modern causal inference approaches to evaluating mediating variables. These approaches specify the criteria for causal relations that clarify the limitations of mediation models and suggest additional methods to identify mediating processes. Several modern approaches to assessing causal mediation relations use instrumental variables. Several approaches to demonstrating causal relations are briefly described, followed by a description of the Rubin causal model, one of the most widely used models to interpret causal relations. Holland’s causal model for the encouragement design uses a detailed notational system, consisting of two sets along with the three variables in the single mediator model. Principal stratification is a most promising approach to causal relations in mediation models. Various unit-level causal effects are never directly observable because of the fundamental problem of causal inference, but they may be used to define causal parameters that can be estimated or measured with data.
