Quantifying social media predictors of violence during the 6 January US Capitol insurrection using Granger causality

Protests involving brute force are growing in number and are viewed as a likely source of increased collective violence in industrialized nations. Yet, our scientific understanding of how violent protests are related to a leader's social media communications during protests remains nascent. Here, we analyse new data from the 6 January 'march on the US Capitol' to quantify the links between leadership, social media and levels of violence. Using data on thousands of live footage videos, Trump's tweets and rally speech, other rally speeches and #StopTheSteal tweets, we apply Granger regression methods to analyse the links between former President Trump's tweets, #StopTheSteal tweets, rally speeches and the severity and duration of outbreaks of violence and weapons use during the riot. We find that Trump's tweets predict bursts in rioters' levels and duration of violence and weapons use. Trump's tweets also predict changes in the volume and sentiments of #StopTheSteal tweets, which in turn explain additional variance in levels of violence and weapons use over the course of the riot. Our findings reveal new patterns of behaviour that link an authority figure's online behaviour during a protest and the shift from peaceful protesting to violence.