Dynamic Fit Index Cutoffs for Time Series Network Models.

Area of Science:

• Statistics
• Psychometrics
• Network Analysis

Background:

• The dynamic fit index (DFI) is a simulation-based method for determining model fit index cutoffs.
• Existing methods may not adequately detect model misspecification in time series network models.

Purpose of the Study:

• Extend the dynamic fit index (DFI) to time series analysis.
• Develop improved methods for deriving fit index cutoffs to detect omitted paths in time series network models.
• Address limitations of the original DFI with small effect or sample sizes.

Main Methods:

• Simulation studies to evaluate DFI cutoffs for time series network models.
• Comparison of DFI cutoffs with established benchmarks (Hu & Bentler).
• Development and evaluation of two alternative DFI approaches (DFI_A and DFI_B) using lenient criteria.

Main Results:

• DFI cutoffs for detecting omitted paths in time series networks are closer to exact fit than traditional benchmarks.
• Cutoff values are influenced by the number of variables, network density, time points, and misspecification type.
• Original DFI fails to identify cutoffs for omitted paths under small effect or sample sizes with strict error rate limits.
• DFI_A and DFI_B provide viable alternatives for deriving cutoffs under more lenient criteria.

Conclusions:

• The extended DFI provides more accurate fit index cutoffs for time series network models.
• DFI_A and DFI_B offer practical solutions for detecting model misspecification, particularly in challenging data conditions.
• These methods enhance the reliability of model evaluation in time series network analysis.