Mapping Cognitive-Behavioral Networks: A Graphical Lasso Approach to Dependence Structure Visualization

This article presents a comprehensive review of statistical methods for estimating graphical model structures across both Gaussian and non-Gaussian distributions. We examine Gaussian Graphical Models (GGMs), highlighting their limitations and introducing a robust implementation of the Graphical Lasso (Glasso) to address these challenges. The proposed methodology can be extended to count data, including multivariate binary, multinomial, and Poisson distributions, with robust estimation techniques developed for each case. Our comparative analysis demonstrates that GLasso-based partial correlation matrices produce sparser and more interpretable dependency structures compared to Maximum Likelihood Estimation (MLE) based partial correlation matrices. To validate our approach, we analyze two real-world Cognitive-Behavioral datasets. The first application visualizes symptom network structures in PostTraumatic Stress Disorder (PTSD), while the second examines dependency patterns among key factors in student emotional monitoring. The results consistently show that GLasso outperforms the MLE in visualizing the dependence structure, offering both computational efficiency and improved interpretability. These findings provide practical insights for researchers analyzing complex dependency structures, particularly in psychological and behavioral studies where interpretable graphical models are essential. The robust methodologies presented here advance current approaches to graphical model estimation while maintaining statistical reliability across diverse data types.