As Hox (2002) points out, educational researchers model complex
relationships between latent and/or observed variables using multilevel
SEM models. In addition to facilitating the examination of direct, indirect,
and reciprocal effects in a single analysis, as in HLM multilevel SEM has
the added benefit of examining relationships between outcome variables at
lower levels of analysis (i.e., at the individual level) and explanatory variables at higher hierarchical levels (i.e., at the group level). However, unlike HLM,
SEM allows variables to be modeled as outcomes and predictors in separate
but interrelated equations (Hox, 2002; Kline, 1998).