A multivariate statistical model for emotion dynamics.

Area of Science:

• Psychology
• Computational Social Science
• Affective Science

Background:

• Emotion dynamics research typically analyzes individual features separately, limiting the understanding of their interrelationships.
• Existing methods struggle with short time series and missing data common in intensive longitudinal emotion studies.

Purpose of the Study:

• To propose a novel vector autoregressive Bayesian dynamic model for analyzing elementary emotion dynamic features.
• To enable the simultaneous study of multiple emotion properties and their relationships, even with limited and incomplete longitudinal data.

Main Methods:

• Developed a vector autoregressive Bayesian dynamic model incorporating six elementary emotion properties: within-person variability, innovation variability, inertia, granularity, cross-lag regression, and average intensity.
• The model accommodates univariate and multivariate time series, short data lengths, missing values, external variables, and non-Gaussian data.

Main Results:

• The proposed model effectively integrates and quantifies six key emotion dynamic properties within a single framework.
• Demonstrated the model's applicability on a dataset of 50 participants reporting on 3 emotions over one week using experience sampling, highlighting its utility with real-world data.

Conclusions:

• The Bayesian dynamic model offers a powerful, flexible tool for advancing emotion dynamics research by enabling comprehensive analysis of interconnected emotional features.
• This approach overcomes limitations of traditional methods, facilitating deeper insights into the complex interplay of emotions in intensive longitudinal studies.