Score-Guided Structural Equation Model Trees.

Area of Science:

• Psychometrics
• Statistical Modeling
• Computational Statistics

Background:

• Structural Equation Model (SEM) trees are data-driven tools for identifying variables that predict group differences in SEM parameters.
• Traditional SEM tree estimation, primarily using the R package semtree, relies on a computationally intensive likelihood-ratio approach for variable selection.
• This computational demand hinders the efficiency of SEM tree construction.

Purpose of the Study:

• To introduce a novel, computationally efficient method for constructing SEM trees.
• To implement score-guided SEM trees within the existing semtree R package.
• To evaluate the performance of score-guided SEM trees using Monte Carlo simulations.

Main Methods:

• The study proposes guiding SEM tree construction using score-based tests, a method popularized in psychometrics.
• Score-based tests efficiently detect parameter differences by monitoring fluctuations in case-wise likelihood function derivatives.
• This approach avoids the computationally demanding refitting of models required by the traditional likelihood-ratio method.

Main Results:

• Score-guided SEM trees are introduced and implemented in the semtree package.
• Monte Carlo simulations are employed to evaluate the performance of the new method.
• The results are expected to demonstrate the computational advantages of score-based tests over likelihood-ratio tests in SEM tree construction.

Conclusions:

• Score-guided SEM trees offer a computationally efficient alternative for identifying predictors of group differences in SEM parameters.
• The proposed method enhances the practicality and speed of SEM tree analysis.
• This advancement facilitates more accessible and efficient data-driven subgroup analysis in structural equation modeling.