Geographically aggregated psychological traits from linguistic analysis of Twitter data predict U.S. voter realignment since 2016

The 2016 U.S. Presidential election heralded the beginning of a political realignment in American politics. A key question for understanding this realignment is whether the Republican party’s shift towards right-wing populism was driven by Donald Trump’s candidacy, versus Trump’s political success being driven by dynamics in the electorate that predated his political rise. To address this question, we examined a corpus of Twitter posts written between July 2009 and February 2015, aggregated by U.S. county. The geographic distribution of psychological traits (personality, empathy, and moral foundations) was estimated by applying to the aggregated Twitter data lexica quantifying how strongly individual words predict psychological traits. The aggregate personality measures were then used to predict Donald Trump’s vote share in 2016, 2020, and 2024, as compared to Mitt Romney’s 2012 vote share, while controlling statistically for the 2000-2008 Republican vote share. Low agreeableness predicts support for Trump but not Romney, a novel result relative to other geographically aggregated data but consistent with prior survey findings relating this trait to right-wing populism. Low empathic concern also predicts vote share for Trump but not Romney. Finally, the degree to which tweets tend to reference unfairness and defilement exclusively predicts shifts towards Trump. Our analysis suggests that people in geographic regions that shifted rightward beginning in 2016 were already expressing emotions consistent with Donald Trump’s messaging in their social media postings before his political rise. Our analysis also provides novel evidence for the high value of aggregated social media data in elucidating voter psychology.