Causal Inference Lecture Notes: Randomized Experiments with Noncompliance

We have studied how to analyze randomized experiments using the randomization-based methods without invoking parametric assumptions. Next, we consider how this approach can be extended to the situations where some experimental subjects do not comply with the randomized treatment assignment. Such noncompliance (or imperfect compliance) occurs frequently in social science field experiments where researchers, for ethical and practical reasons, cannot force experimental subjects to take the assigned treatment. Moreover, researchers often use the encouragement design to conduct a controversial experiment. In this design, randomly selected individuals are encouraged to receive the treatment. Thus, some may take the treatment without the encouragement, and others may not do so even when encouraged.