Exploratory Data Mining with Structural Equation Model Trees

Structural Equation Model Trees (SEM Trees) combine Structural Equation Models (SEM) and decision trees. SEM Trees are tree structures that partition a dataset recursively into subsets with significantly different sets of parameter estimates. The method allows the detection of heterogeneity observed in covariates and thereby offers the possibility to automatically discover non-linear influences of covariates on model parameters in a hierarchical fashion. The methodology allows an exploratory approach to SEM by providing a data-driven but hypothesis-constrained exploration of the model space. We summarize the methodology, show applications on empirical data, and discuss Hybrid SEM Trees, an extension of SEM Trees that allows the finding of subgroups that differ with respect to model parameters and model specification.