Sensing Psychological Well-being Using Social Media Language: Prediction Model Development Study

Area of Science:

• Computational Social Science
• Psychological Measurement
• Digital Health

Background:

• Assessing psychological well-being (PWB) at scale is challenging.
• Social media offers a nonintrusive method for monitoring user PWB.
• This study explores social media language as a PWB predictor.

Purpose of the Study:

• To investigate the predictive accuracy of social media language for psychological well-being.
• To develop and validate a model for estimating PWB from online linguistic expressions.
• To assess the model's reliability and validity using established psychometric criteria.

Main Methods:

• Recruited 1427 participants, assessing their PWB across 6 dimensions.
• Collected social media posts and extracted linguistic features using 6 psychological lexicons.
• Developed a multiobjective prediction model and evaluated its discriminant, convergent, and criterion validity, alongside split-half reliability.

Main Results:

• The linguistic prediction model demonstrated good criterion validity, with correlation coefficients between predicted and actual PWB scores ranging from 0.49 to 0.54 (P<.001).
• The model showed excellent convergent validity but less satisfactory discriminant validity.
• Good split-half reliability was observed across all PWB dimensions (0.65–0.85; P<.001).

Conclusions:

• The study confirms the availability and stability of a linguistic prediction model for PWB.
• Social media language effectively predicts psychological well-being, validating its use in nonprofessional settings.
• Findings support the use of social media data for large-scale mental health assessment and self-testing.