ABSTRACT
Sometimes this is not the case. In particular, if the data is collected in certain clusters like families, practices, hospital departments, etc., it may happen that these clusters have an influence by their own without being covariates. For example, if we look at an outcome like fat intake in subjects, and we collect data in families, we may have to expect a tendency that the shared cooking and eating habits of a family influence the outcome. So if we know for one member of a family that he or she has a high fat intake (relative to his or her covariate pattern), then we have to expect to some degree that also the other family members tend to have a high fat intake. So the outcomes are no longer independent, and we have a correlation of the outcomes within each cluster, which we cannot explain by the covariate values of the subjects.
