Predicting Mood Based on the Social Context Measured Through the Experience Sampling Method, Digital Phenotyping, and...

Area of Science:

• Psychology
• Computational Social Science
• Digital Health

Background:

• Social interactions are crucial for well-being and mood.
• Accurately capturing an individual's social context is key to predicting well-being.
• Existing methods like digital phenotyping, ESM, and egocentric networks measure different social aspects.

Purpose of the Study:

• To investigate the predictive accuracy of mood based on social context.
• To assess the utility of combining multiple data sources for mood prediction.
• To address the gap in research combining diverse social context measurement methods.

Main Methods:

• Collected intensive within-person data over 28 days from a student sample (N=11).
• Utilized digital phenotyping, experience sampling method (ESM), and egocentric networks.
• Trained individualized random forest machine learning models with predictors summarized over various time scales.

Main Results:

• Mood prediction accuracy remained low even when combining social interaction data from different methods (R²=0.06 for affect).
• Predictive accuracy varied significantly across individuals, with optimal predictors differing.
• Combining predictors generally improved mood prediction for most participants.

Conclusions:

• Integrating diverse social context data minimally improved mood prediction accuracy.
• Further research with larger, more diverse samples is necessary to enhance clinical utility.
• Current predictive modeling approaches for mood require refinement.