Standard randomization-based tests of sharp null hypotheses in randomi
zed clinical trials, that is, intent-to-treat analyses, are valid with
out extraneous assumptions, but generally can be appropriately powerfu
l only with alternative hypotheses that involve treatment assignment h
aving an effect on outcome. In the context of clinical trials with non
-compliance, other alternative hypotheses can be more natural. In part
icular, when a trial is double-blind, it is often reasonable for the a
lternative hypothesis to exclude any effect of treatment assignment on
outcome for a unit unless the assignment affected which treatment tha
t unit actually received. Bayesian analysis under this alternative 'ex
clusion' hypothesis leads to new estimates of the effect of receipt of
treatment, and to a new randomization-based procedure that has freque
ntist validity yet can be substantially more powerful than the standar
d intent-to-treat procedure. The key idea is to obtain a p-value using
a posterior predictive check distribution, which includes a model for
non-compliance behaviour, although only under the standard sharp null
hypothesis of no effect of assignment (or receipt) of treatment on ou
tcome. It is important to note that these new procedures are distinctl
y different from 'as treated' and 'per protocol' analyses, which are n
ot only badly biased in general, but generally have very low power. (C
) 1998 John Wiley & Sons, Ltd.