Causal Modeling of Descriptive Social Norms from Twitter and the Physical World on Expressed Attitudes Change: A Case Study of COVID-19 Vaccination

Abstract

The high infection rate of SARS-CoV-2 makes it urgent to promote vaccination among the public. Previous studies found that people tend to follow the behaviors desired in descriptive social norms, which exist in both social media (e.g., Twitter) and physical-world communities. However, it remains unclear whether and to what extent the descriptive social norms from the cyber and physical communities affect people's attitude change. This study, focusing on COVID-19 vaccination, developed a Directed Acyclic Graphs model to investigate the causal effects of the descriptive social norms of (i) Twitterverse and (ii) physical-world communities on people's attitude change as well as the temporal scales of the effects. It used a Long Short-Term Memory classifier to extract expressed attitudes and changes from relevant tweets posted by 843 sample users. We found that a people's attitude change toward the vaccination receives a more significant impact from Twitter-based descriptive social norms over the prior week, whereas the norms in the physical-world communities tend to be less influential but still notable with the time gap between 2 weeks and 1 month. The findings revealed the potential of using online social norm approaches to proactively motivate behavioral changes toward a culture of health.