Bayesian estimation and comparison of idiographic network models.

Area of Science:

• Psychological network analysis
• Time series modeling
• Statistical modeling

Background:

• Idiographic network models analyze person-specific associations from time series data.
• Traditional graphical vector autoregressive (GVAR) models use LASSO regularization, which can be unstable with small datasets typical in psychology.
• This instability risks misinterpreting random variation as genuine individual differences (heterogeneity).

Purpose of the Study:

• To evaluate a Bayesian alternative for fitting GVAR models, accounting for estimation uncertainty.
• To develop and assess a novel statistical test for the reliability of differences between estimated networks.
• To compare Bayesian and LASSO GVAR approaches and validate the new test.

Main Methods:

• Simulation studies comparing Bayesian and LASSO GVAR performance under various conditions.
• Development of a novel statistical test for network differences, implemented in the R package 'tsnet'.
• Application of Bayesian GVAR estimation and the novel test to empirical data on daily clinical symptoms.

Main Results:

• LASSO estimation performed well, but Bayesian GVAR without edge selection showed advantages for dense networks.
• The novel statistical test demonstrated conservative properties and good control of false-positive rates.
• Bayesian GVAR modeling effectively assessed estimation uncertainty, crucial for understanding interindividual differences in intraindividual dynamics.

Conclusions:

• Bayesian GVAR modeling provides a robust framework for analyzing individual differences in time series data.
• The novel statistical test acts as a safeguard against erroneous conclusions of heterogeneity.
• This approach enhances the reliability of findings in psychological network research.