ABSTRACT
To examine the impacts of government crisis messaging on social media, we draw on Situation Crisis Communication Theory to classify government messages related to COVID-19 and develop theories about how these messages affect citizen online engagement and offline compliance. We utilize gradient boosting trees to classify tweets of fifty U.S. governors from March to December 2020. To mitigate social desirability bias, we connect social media data with mobility data, which reveals actual compliance with policies. Using double fixed-effect models, we show that governors' informational, instructional, and compassionate messages are consistently associated with increased citizen online engagement with state government. The online engagement, in turn, correlates with compliance with stay-at-home orders and advisories to avoid non-essential travel except in Republican-controlled states. Meanwhile, governors' instructional, compassionate, and praising messages are directly associated with better compliance. However, the direct associations except for compassionate ones disappear in the last four months of 2020.