ABSTRACT
Many data sets are suitable for multilevel mediation modeling. The mediation model can be extended to multilevel analysis, but it is considerably more complicated and requires a new iterative estimation strategy. There are several additional multilevel mediation models that are possible because of the possibility of measures at multiple levels of analysis. A measure of the extent to which observations in the same group tend to respond in the same way is the intraclass correlation (ICC). Given the problem of a nonzero ICC on standard errors, researchers have typically used three options before multilevel models were introduced. The use of a multilevel mediation model is described and illustrated with a hypothetical study of the effects of two different types of group therapy on depression. The MIXED code to estimate the independent variable effect on the mediator is analogous to the program effect on the dependent variable.
