I Won’t Do What You Tell Me: Noncompliance, Encouragement, and Preserving Causality in Experiments

Experimental psychologists rely on random assignment to infer causality; but what happens when participants simply do not do what is asked of them in the condition they were randomly assigned to? Noncompliance—when participants fail to take the treatment they’re assigned or take one they weren’t—breaks randomization and threatens causal inference. Here a conceptual and intuitive overview of the problems with noncompliance, differential attrition, and broken randomization are provided before introducing an Intention-to-Treat (ITT) analysis, a conservative but causally valid approach, followed by more flexible tools, such as instrumental variables and encouragement designs. The goal is not to turn every experimentalist into an econometrician, but rather to clarify, in non-statistician language, the logic that preserves causality even when people behave as, well, people.