Quantifying generalized trust in individuals and counties using language

Area of Science:

• Social Psychology
• Computational Social Science
• Digital Humanities

Background:

• Trust is crucial for societal functions like civic cooperation and economic growth.
• Declining trust in institutions and increasing partisan division in the U.S. necessitate new measurement methods.
• Existing trust assessments can be obtrusive or difficult to scale.

Purpose of the Study:

• To develop and validate a language-based assessment for measuring generalized trust.
• To apply this assessment to large-scale social media data to estimate trust levels across U.S. counties.
• To explore the relationship between estimated trust levels and community well-being.

Main Methods:

• Collected language data from over 16,000 Facebook users and correlated it with self-reported trust scores.
• Developed a language-based trust assessment model.
• Applied the model to over 1.6 billion geotagged tweets (2009-2015) to estimate trust in 2,041 U.S. counties.

Main Results:

• Generalized trust was associated with the use of more affiliative words (e.g., "love", "we", "friends") and fewer angry words (e.g., "hate", "stupid").
• Social word associations were weak, primarily driven by negative associations with "othering" terms.
• Higher trust counties showed better physical health and higher life satisfaction, according to CDC and Gallup data.

Conclusions:

• Language analysis offers a scalable, low-cost, and unobtrusive method for assessing generalized trust.
• Estimated trust levels correlate with measurable community health and well-being indicators.
• This approach can help monitor population-level trust dynamics.