ABSTRACT
It is well known that experimental designs based on randomization are powerful in terms of statistical analysis and inference. However, the estimation of treatment effects can be biased even with successful randomization unless everyone complies with the given treatment. Noncompliance is not only an obstacle to fair statistical comparison between the treatment group and the control group, but also a major threat to obtaining power to detect intervention effects (Jo, 2000c). Depending on how noncompliance is dealt with in the estimation of treatment effects, different conclusions may be reached about the effect of the same intervention trial.
