Deepening the Divide: Crises Disproportionately Silence Vulnerable Populations on Social Media

Abstract

In the last decade, crisis informatics has sought to produce and use actionable information from social media data. Although substantial progress has been made in discerning how the data can be used, there is a lack of research identifying possible inequities in that use. Previous research has shown that vulnerable populations use social media less in a disaster; however, the extent to which this social media usage disparity is predictable and the magnitude of that disparity has not been explored. This paper compared the covariance in Twitter activity and social vulnerability factors during a steady-state period prehurricane with that during the perturbed state period following Hurricane Harvey’s landfall. These models showed that sociodemographic vulnerability factors better predict Twitter activity during a crisis than do infrastructural damage, that sociodemographic factors negatively influence Twitter activity, and that this phenomenon is strengthened by a crisis. The crisis-specific negative covariance indicates the need for increased consideration of vulnerability factors in social media data–driven management of urban resilience and resource distribution.