Affect and Personality: Ramifications of Modeling (Non-)Directionality in Dynamic Network Models.

Area of Science:

• Psychology
• Network Science
• Statistics

Background:

• Dynamic network models are increasingly used to analyze longitudinal data, capturing complex temporal relationships.
• These models often extend or adapt standard vector autoregressive (VAR) frameworks.
• A lack of direct comparison between popular dynamic network approaches hinders methodological understanding.

Purpose of the Study:

• To compare three prominent dynamic network models: GIMME, uSEM, and LASSO gVAR.
• To evaluate their differences in modeling assumptions, estimation, and statistical properties.
• To assess their implications for research in affect and personality.

Main Methods:

• A Monte Carlo simulation was employed to examine the statistical properties of each model.
• The study systematically compared GIMME, uSEM, and LASSO gVAR.
• The comparison focused on modeling assumptions, estimation procedures, and simulation-based properties.

Main Results:

• All three dynamic network approaches yielded group-level results partially supporting affect and personality theories.
• Significant heterogeneity was observed in individual-level results across the different approaches and participants.
• Discrepancies in findings were identified and analyzed.

Conclusions:

• While group-level analyses show consensus, individual-level network dynamics vary considerably between GIMME, uSEM, and LASSO gVAR.
• Researchers should be aware of the distinct assumptions and properties of each method when analyzing longitudinal data.
• The choice of dynamic network model can significantly impact findings, particularly at the individual level.