Abstract:
|
The strong internal validity of randomized controlled trials (RCTs) means these experiments yield an unbiased study sample average treatment effect (SATE). However, an RCT sample is often not a random sample of the target population the treatment is intended for. Hence, in the presence of heterogeneous treatment effects, SATEs may not equal the target population average treatment effect (PATE). Novel transportability methods have been developed to extend causal effects from RCTs to the different target populations. These methods rely on the correct specification of effect modifiers. However, it is difficult to know whether this assumption holds until more information is collected in the target population. Most transportability work has been disconnected from RCT trial designs. In this paper, we hybridize transportability methods with a prospective RCT design to design more efficient RCTs when the transportability assumptions hold using empirical Bayes. After comparing a handful of estimation methods, we explore how even slight violations of the effect modifiers assumption affect PATEs when using traditional transportability methods as well as our novel hybrid design.
|