Randomised controlled or clinical trials (RCTs) are generally viewed as the most reliable method to draw causal inference as to the effects of a treatment, as they should guarantee that the individuals being compared differ only in terms of their exposure to the treatment of interest. This 'gold standard' result however hinges on the requirement that the randomisation device determines the random allocation of individuals to the treatment without affecting any other element of the causal model. This 'no randomisation bias' assumption is generally untestable but if violated would undermine the causal inference emerging from an RCT, both in terms of its internal validity and in terms of its relevance for policy purposes. This paper offers a concise review of how the medical literature identifies and deals with such issues.
clinical trials social experiments design of experiments randomisation bias sample selection causal inference treatment effects external validity generalizability