Modeling Heterogeneity in Indirect Effects: Multilevel Structural Equation Modeling Strategies

Abstract

The heterogeneity implicit in much of social science research can be accommodated by using complex modeling procedures such as SEM or MLM. Ignoring heterogeneity, particularly with regard to nested data structures, can have serious consequences for model estimation and lead to incorrect conclusions about tested hypotheses. In mediation models, the consequences of ignoring nesting can have a substantial impact on the indirect effect. Inflated standard errors and bias in the parameter estimates lead to inaccurate estimates of the indirect effect, as well as reduced power to detect the effect. Using multilevel structural equation modeling (MSEM), data were generated based on a cross-lagged panel model for mediation. By fitting a single-level model to the data, the consequences for the estimation and detection of the indirect effect when heterogeneity is ignored is examined through measures of relative bias, power, and model fit.