In this article we focus on clinical trials in which the
compliance is measured with random errors, and develop an
error-in-variables model for the analysis of the clinical trials.
With this model, we separate the efficacy of prescribed treatment
from that of the compliance. With additional information
correlated with compliance, we prove that the model is
identifiable, and get estimators for the parameters of interest,
including the parameter reflecting the efficacy of the treatment.
Furthermore, we extend the model to stratified populations, and
discuss the asymptotic properties of the estimators.