Using Raw VAR Regression Coefficients to Build Networks can be Misleading

Area of Science:

• Behavioral Sciences
• Network Analysis
• Time Series Analysis

Background:

• Behavioral science research often examines complex causal relationships among variables over time.
• Vector autoregressive (VAR) models, specifically VAR(1), are used to visualize these dynamic relations as networks.
• Raw VAR(1) regression coefficients can be misleading due to scale sensitivity and focusing only on direct effects.

Purpose of the Study:

• To address the limitations of using raw VAR(1) regression coefficients for network visualization.
• To propose alternative VAR(1)-based measures that offer better comparability and a more comprehensive view of variable interrelations.
• To enhance the accuracy and interpretability of network analyses in behavioral science.

Main Methods:

• Utilizing lag-one vector autoregressive (VAR(1)) models to analyze time-series data.
• Proposing the use of standardized VAR(1) regression coefficients to improve comparability.
• Introducing relative importance metrics to incorporate direct, shared, and indirect effects into network representations.

Main Results:

• Standardized VAR(1) coefficients enhance comparability across variables with different scales and variances.
• Relative importance metrics provide a more complete picture of variable influence by including indirect effects.
• The proposed methods mitigate misleading network characteristics arising from raw coefficient usage.

Conclusions:

• Standardized coefficients and relative importance metrics offer superior alternatives to raw VAR(1) coefficients for network analysis.
• These enhanced methods yield more accurate and interpretable network visualizations in behavioral science.
• The findings advocate for adopting these improved techniques for robust causal interplay analysis over time.