Dear experts,
essentially, I am dealing with a moderator variable in which the observations are not independent. Here is the situation:
My independent variable is dichotomous and is simply “treatment yes/no”. The treatment group has received a professional development training that aims to reduce burnout symptoms, while the control group has received no such training. As the dependent variable, I (obviously) measure burnout symptoms. However, we believe that the effect of the training on burnout is moderated by how a person rates their work team on two specific dimensions. We measured this team rating with a questionnaire, and our sample contains workers nested in teams, and often multiple workers from the same team. This means that, for some of our participants, their team ratings are not independent, as they are rating the same team. We calculated ICCs for the team ratings and found that about 23% (first team dimension) and 37% (second team dimension) of the overall variance is indeed variance at the team level, confirming that the observations are not fully independent. If I want to enter these ratings into an analysis as a moderator, do I need to take this dependency into account, and if so, how?
If you need any more information for a meaningful answer, please ask and I will reply as soon as I can.