Abstract: Educational researchers are commonly interested in examining and statistically modeling mediation and moderation effects in experiments to see why and for whom interventions work. However, even experimental study designs may overlook that mediator or moderator variables are commonly not randomized. This poses threats to the causal interpretability and potential to inform educational practice of many studies involving mediator or moderator variables. We outline these issues in an example with simulated data and summarize potential methodological solutions.
Fahrbach et al. (Fri,) studied this question.