Research project supported by NIH uses AI to predict how much trust individuals and communities have in U.S. counties.
Stony Brook, NY, Oct 12, 2024 - In the United States, public confidence in institutions has been declining over the past several decades. It is easy to imagine how these low levels of trust might lead to negative social, political, and economic outcomes, especially in the heat of the 2024 presidential elections. This is why measuring trust consistently and at scale is crucial.
Earlier this year, researchers at Stony Brook University joined hands with experts from the University of Pennsylvania, Stanford University, Johns Hopkins University, and the University of Melbourne, to use AI and machine learning for studying social media posts and estimating trust within U.S. counties. Their goal was to use these learnings to gain new insights into community-level health, well-being, and political ideology.Their paper, titled, ‘Quantifying generalized trust in individuals and counties using language,’ was published after careful review in Frontiers in June 2024.
The researchers started by asking participants to complete the Big-Five Personality Questionnaire and moved on to studying what kind of language trusting people use.
High distrust vs High trust
Jason Jeffery Jones, Associate Professor in the Department of Sociology at Stony Brook University said, “We studied the words people were using, across age and genders, and naturally found that people using more positive language, like ‘amazing,’ ‘cheers,’ and ‘party’ found it easier to trust others than those saying ‘tired of,’ ‘hate,’ and ‘wrong.’ We also gleaned specific insights from the data — for instance, younger distrusting individuals noted loneliness ('alone'), confusion ('don't understand'), and feeling bad ('pain', 'worst', 'tired'). Younger trusting individuals expressed more wonder (‘can’t believe’) and positive anticipation (‘looking forward to’).”
The team looked at social media posts of over 154,000 Facebook users — who directly consented to taking the personality survey and sharing their Facebook language data — and used it to train their AI language model.
Then, the researchers applied this new AI model to about 6 million geotagged Twitter users, and looked at the Gallup–Sharecare Wellbeing Index to validate their results.
Levels of trust by U.S. counties. Red indicates higher levels of distrust; blue indicates higher levels of trust (white counties did not have sufficient Twitter language information available).
Mohammadzaman Zamani, a Computer Science Ph.D student at Stony Brook University, shared the study’s insights, “Our analysis revealed that trust in counties was associated with more education, higher income, deep religiousness, and greater population density. More trusting regions also showed a greater percentage of individuals living in committed, stable relationships, while less trusting counties had a greater percentage of separated individuals.”
It’s also interesting to note that trust is low in areas with ethnic, economic, and political diversity. Moreover, trust is strongly linked to the health and well-being of individuals in a county.
Then, the researchers examined each county’s trust in relation to its political factors.
According to Associate Professor H. Andrew Schwartz, of The Department of Computer Science at Stony Brook University, the relationship between people’s trust and their party preference shifted between the 2012 and 2016 presidential elections, and the Republican vote share moved away from the positive toward the negative. But come 2016, the county trust levels and the votes gained by the 2016 Republican candidate Donald Trump were greater than the previous Republican candidates.”
Trump, the study suggests, was more supported more in counties that used language indicative of high distrust.
Schwartz adds, “Two possible hypotheses follow from this. First, communities lower in generalized trust prefer populist candidates. Second, decreasing levels of generalized trust lead to lower vote shares for incumbent candidates.”
“These predictions are speculative, and based on one surprising election result, but the real-time and local nature of the social media data could allow for future evaluations to be tested in a principled way—by predicting which type of candidate will be unexpectedly popular both locally and nationally.”
Their research suggests a method to consistently and persistently analyze trust in large populations, so we can identify the key aspects of trust at the community, county, and country levels.
Communications Assistant
Ankita Nagpal