Why Choose One Level of Analysis? and Other Issues in Multilevel Research

Several issues related to multilevel research and measurement theory are discussed. It was emphasized that questions regarding choice of level of measurement and unit of analysis are less important than the choice of an appropriate statistical model. It was demonstrated that even with an "error-free" environmental measure not all persons should view the setting in the same manner. The split-sample correlation was clearly shown to have undesirable statistical properties, and the use of standard methods was stressed. Investigators were cautioned to carefully choose a coefficient and were shown that this choice is sometimes tricky. Finally, the analysis of multilevel data using multilevel methods was recommended.