Ian White
In randomised trials with departures from allocated treatment, it may be desirable to estimate the causal effect of treatment. Intention-to-treat analysis provides an unbiased comparison of treatment policies as implemented and does not estimate the causal effect of treatment. Per-protocol analysis is the most commonly used alternative in clinical trials, but it is subject to selection bias. Randomisation-based methods of causal inference avoid both these problems. They are not widely used, perhaps because of their complexity and unfamiliarity to many statisticians. I will give some examples where randomisation-based causal inference has given useful conclusions.
