Analytic Strategies for Analyzing Multilevel Data

Abstract

Multilevel analyses have become increasingly common in psychological research, although unfortunately, many researchers’ understanding of multilevel analysis has lagged behind this increased interest and use. Many researchers have heard of and are curious about multilevel modeling (MLM), but they are unfamiliar with it, perhaps so unfamiliar that they do not know where to start. This unfamiliarity is probably due in part to the fact that many graduate programs in psychology do not offer (or have not offered) courses in multilevel analysis. This chapter is an attempt to meet this need by familiarizing readers with MLM as it pertains to psychological research broadly defined. In writing this chapter, I had two goals in mind. First, I wanted readers to learn the basics of multilevel analysis. Second, I wanted to increase readers’ awareness of the multilevel perspective so that they might recognize the multilevel features of the data they have collected and would be able to be formulate more clearly research questions that might involve multilevel data. As Kreft and de Leeuw (1998) noted, “Once you know that hierarchies exist you see them everywhere” (p. 1). Conversely, if you do not know how to conceptualize a multilevel data structure and the accompanying analyses, you may not see or recognize hierarchies anywhere. In this chapter, I provide a rationale for MLM: why it is necessary, its advantages over other techniques, and so forth. I describe the basic structure of univariate multilevel analyses: the nature of the models and the types of parameters they can estimate and how to conduct multilevel analyses, including different aspects of analyses such as centering, modeling error, weighted analyses, and categorical independent and dependent measures. I also offer suggestions about how to interpret the results of analyses and how to report results in papers. Finally, although they are in flux, I discuss software options. This chapter is intended as an introduction for those who are not familiar with MLM. When writing this chapter, the only statistical training I assumed readers would have was an understanding of basic ordinary least squares (OLS) regression. Analysts who are familiar with the basics of MLM may find some value in my treatment, but advanced topics are not covered. Other chapters in this handbook cover some of these topics, such as Chapters 18, 19, and 20 of this volume.