Abstract:
|
Direct and indirect effects decompose the effect of a treatment into the part mediated by a covariate (the mediator) and the part that is not. A causal interpretation of mediated effects usually relies on "cross-worlds" counterfactuals, which can never be observed because they involve quantities under two different treatments where only one treatment is given to any particular patient. Lok (Statistics in Medicine 2016) proposed "organic" direct and indirect effects, a generalization of natural direct and indirect effects which instead of cross-worlds counterfactuals are based on "organic" interventions. I will present an extension of organic direct and indirect effects to situations with post-treatment common causes of the mediator and the outcome. In this situation, organic direct and indirect effects are no longer a generalization of natural direct and indirect effects. I will compare organic direct and indirect effects in this situation with three approaches proposed in VanderWeele et al (2014). I will show that in contrast to natural direct and indirect effects, organic direct and indirect effects are identifiable when there are common causes of the mediator and the outcome.
|