Dynamic time warp versus vector autoregression models for network analyses of psychological processes

Area of Science:

• Psychology
• Network Science
• Computational Psychiatry

Background:

• Psychopathological disorders are increasingly viewed as complex dynamic systems represented by interconnected symptom networks.
• Multilevel Vector AutoRegression (mlVAR) models are commonly used for these network analyses.
• However, psychological data often violates the assumptions of mlVAR.

Purpose of the Study:

• To evaluate the applications, advantages, and disadvantages of Dynamic Time Warping (DTW) and mlVAR for analyzing temporal relationships in psychological data.
• To compare the robustness of DTW and mlVAR under conditions that violate mlVAR assumptions.

Main Methods:

• Utilized Ecological Momentary Assessment data from 376 participants in the Netherlands Study of Depression and Anxiety.
• Constructed item networks using both mlVAR and DTW techniques on 20 mood and physical condition items.
• Employed simulated data to test network reliability under various assumption violations, including lagged relationships and collider variables.

Main Results:

• Simulated data analysis indicated that mlVAR networks are more prone to spurious connections than DTW networks.
• DTW demonstrated greater reliability in constructing networks when mlVAR assumptions were violated.
• mlVAR is effective for causal inference when assumptions hold, whereas DTW is robust for examining co-occurrence and synchrony.

Conclusions:

• Dynamic Time Warping (DTW) presents a robust alternative to Multilevel Vector AutoRegression (mlVAR) for analyzing complex temporal dynamics in psychological data, particularly when data violates standard model assumptions.
• DTW provides a reliable method for assessing synchrony and lagged connections in real-world psychological assessments.
• The findings suggest DTW is a valuable tool for understanding the dynamic interplay of symptoms in psychopathology.