Chapter 2: Multilevel Mixture Models

Multilevel statistical models allow researchers to evaluate the effects of individuals’ shared environment on an individual’s outcome of interest. Finite mixture models allow the researchers to question the homogeneity of the population and to classify individuals into smaller and more homogeneous latent subpopulations. Structural equation models allow the researchers to explore relations between observed variables and latent constructs. As researchers get more and more experience with these techniques they will inevitably want to use them within a unified framework that will enable them to combine all these ideas into a comprehensive statistical model that addresses all features present in the data. In this chapter we will describe a general statistical model that incorporates multilevel models, finite mixture models, and structural equation models into a very general and flexible modeling framework. The basis of this methodology was first implemented in Mplus Version 3 (Muthén & Muthén, 2004), while the complete modeling framework described in this chapter became available in Mplus 4.2 (Muthén & Muthén, 2006). Overall, the topic of multilevel mixture models is relatively new, although a number of articles have discussed similar frameworks and applications (see, e.g., Asparouhov, 2006; Bijmolt, Paas, & Vermunt, 2004; Vermunt, 2003; Vermunt & Magidson, 2005).