Psychometric network models from time-series and panel data.

Area of Science:

• Network psychometrics
• Statistical modeling
• Psychological science

Background:

• Gaussian graphical models (GGMs) are commonly used but assume no measurement error and cannot differentiate between within- and between-subject effects.
• Cross-sectional data limits the ability to capture temporal dynamics and individual differences over time.

Purpose of the Study:

• To extend Gaussian graphical models (GGMs) to incorporate latent variables and temporal relationships.
• To provide a general framework for analyzing complex network structures in time-series and panel data.

Main Methods:

• Developed a graphical vector-autoregression model for latent variables, termed ts-lvgvar (time-series) and panel-lvgvar (panel data).
• Implemented the methods in the psychonetrics software package.
• Evaluated the models using two empirical examples and two large-scale simulation studies.

Main Results:

• The proposed framework successfully extends GGMs to model latent variable dynamics over time.
• The ts-lvgvar and panel-lvgvar models can distinguish between within-subject and between-subject effects under specific conditions.
• The psychonetrics software facilitates the application of these advanced network models.

Conclusions:

• The developed models offer a more comprehensive approach to network psychometrics, accounting for latent variables and temporal dependencies.
• Interpretation of results necessitates careful consideration of measurement intensity, time intervals, and stationarity assumptions.
• The framework enhances the generalizability and ergodicity of network models in psychological research.