Title: The subgroupSEM R Package for Subgroup Discovery in Structural Equation
Models

Author: Benedikt Langenberg

Abstract:
Structural equation modeling (SEM) is a very popular statistical method in
social and behavioral sciences and detecting groups with distinct parameter
sets is of key importance to applied researchers. SubgroupSEM is a novel
approach for efficient subgroup discovery in structural equation models. The
approach is a combination of the well-established subgroup discovery framework
and SEM. SubgroupSEM can serve as an exploratory technique that helps generate
hypotheses by detecting unique groups with distinct combinations of
covariates. For instance, the approach has been applied in detecting subgroups
of university students with unusual trajectories of dropout intentions. This
presentation introduces the SubgroupSEM R package, which incorporates recent
developments from the subgroup discovery framework and makes them available
for researchers working with SEM. The package is designed to be used with any
type of SEM (e.g., growth curve models or mediation models), and only requires
(1) a data set, (2) a set of covariates, and (3) a function that estimates the
SEM and calculates some measure of interestingness (e.g., magnitude of point
estimates or test statistics). The package offers different search strategies
that can be used to explore the covariate space (e.g., depth-first,
best-first, or beam search). The presentation concludes with a demonstration
of the package using real empirical data.