The invariance partial pruning approach to the network comparison in time-series and panel data.

Area of Science:

• Network analysis
• Psychometrics
• Statistical modeling

Background:

• Network models are crucial for understanding dynamic relationships among variables in time-series and panel data.
• Comparing network structures across groups is vital for explaining individual heterogeneity in treatment response.
• Existing methods for comparing individual networks lack the precision to pinpoint specific edge differences.

Purpose of the Study:

• To introduce a novel approach, invariance partial pruning (IVPP), for comparing idiographic networks from time-series and panel data.
• To address the limitations of global tests in identifying precise locations of network heterogeneity.
• To provide an easily applicable method for comparing networks with limited time points per individual.

Main Methods:

• The IVPP approach combines a global network invariance test with partial pruning to identify edge-level heterogeneity.
• Simulations were conducted to evaluate the performance of different statistical criteria (AIC, BIC, LRT) for the invariance test.
• The performance of IVPP was compared against fully constrained and fully unconstrained models.

Main Results:

• The network invariance test using AIC and BIC performed well, though BIC showed insufficient power at small sample sizes.
• The likelihood ratio test demonstrated a tendency for false discovery.
• Partial pruning successfully identified specific edge differences with high sensitivity and specificity.

Conclusions:

• The IVPP approach is a valuable addition to network methodology, enabling precise comparisons of networks across individuals and time.
• It effectively identifies specific edge differences, offering insights into treatment response heterogeneity.
• The IVPP algorithm is implemented in the R package 'IVPP' for practical application.