Network Trees: A Method for Recursively Partitioning Covariance Structures

Area of Science:

• Psychology
• Quantitative Psychology
• Psychometrics

Background:

• Correlation-based network approaches, such as psychometric networks, are widely used in psychology.
• Existing methods may not fully capture how network structures vary across different subgroups within a sample.

Purpose of the Study:

• To propose a novel approach for detecting significant differences in correlation or covariance matrix structures.
• To enable the estimation of psychometric networks or other correlation-based models from recursively split samples.

Main Methods:

• Adaptation of model-based recursive partitioning and conditional inference tree methods for covariate-based splitting.
• Recursive splitting of samples based on identified covariates to isolate subgroups with potentially different network structures.
• Estimation of psychometric networks and factor models from the resulting data splits.

Main Results:

• The proposed recursive partitioning approach effectively identifies covariate-driven differences in network structures.
• Simulation studies demonstrate the empirical power of the method under various conditions.
• The approach is validated using real-world data from personality and clinical psychology research.

Conclusions:

• This method provides a powerful tool for exploring heterogeneity in psychometric network structures.
• It enhances the understanding of how individual differences (covariates) influence psychological constructs.
• The approach offers a valuable extension for advanced network and correlation-based modeling in psychological science.