Causal Inference: Efficacy and Mechanism Evaluation

In randomized trials, the primary analysis is usually based on an intention-to-treat approach which answers the question “What is the effect of offering treatment?” There are many other questions that investigators could pose such as “Does this treatment work if it is received?” “What factors make the treatment work better?” and “How does the treatment work?” These questions require alternative analysis approaches based on statistical methods drawn from the causal inference literature, including instrumental variables and causal mediation analysis. This chapter will define relevant causal estimands and describe methods that can be used to estimate them, their underlying assumptions, and the estimation procedures. The methods will be illustrated using examples drawn from the literature.