Multilevel models and unbiased tests for group based interventions: Examples from the safer choices study

For many large-scale behavioral interventions, random assignment to interve ntion condition occurs at the group level. Data analytic models that ignore potential nonindependence of observations provide inefficient parameter es timates and often produce biased test statistics. For studies in which indi viduals are randomized by groups to treatment condition, multilevel models (MLMs) provide a flexible approach to statistically evaluating program effe cts. This article presents an explanation of the need for MLM's for such ne sted designs and uses data from the Safer Choices study to illustrate the a pplication of MLMs for both continuous and dichotomous outcomes. When desig ning studies, researchers who are considering group-randomized intervention s should also consider the features of the multilevel analytic models they might employ.