COVID-19: adverse population sentiment and place-based associations with socioeconomic and demographic factors

During the COVID-19 pandemic, increased adverse sentiment such as, fear, panic, anxiety was observed among the public in the United States of America (USA) apart from physical suffering and death. Authorities may find guidance for anticipation and explanation of such secondary threats by analyzing population sentiment on social media. We performed sentiment analysis (SA) using georeferenced tweets in the contiguous USA during the first two waves of COVID-19 (01 November 2019–15 September 2020). We classified the tweets into “adverse” and “non-adverse” sentiment and computed daily counts for both classes at the county-level. Utilizing clustering and Bayesian regression approaches, we analyzed the place-based demographic and socioeconomic covariates of sentiment. We detected 12 clusters that exhibited elevated adverse sentiment and discovered that higher unemployment, male population, and poverty was associated with increased odds of adverse sentiment in Tweets. Conversely, counties with higher COVID-19 case rates, rurality, and elderly population were associated with reduced odds. Pandemic preparedness, response and mitigation measures may benefit from knowledge of the geography of adverse sentiment. Combining spatial clustering and regression benefits the understanding COVID-19, as well as epidemiology in general.