Using billions of geolocated tweets to prospectively predict domestic violence

EWB Hub Presentation in collaboration with MDAP

Using billions of geolocated tweets to prospectively predict domestic violence

Dr. Khandis Blake.

How people behave and present themselves online is a socially important but historically novel aspect of the human social repertoire. Yet the state of knowledge about how and why people behave as they do online remains embryonic. In particular, the uncivil, sexist, and often abusive behaviours people engage on social media—especially on topics such as sex, gender, sexuality and sexual autonomy—have profound consequences, but are inadequately explained. Knowledge of how online social behaviour covaries with real-world outcomes also remains poorly understood. Using a dataset of billions of geolocated Tweets across multiple years, we examined the relationship between the frequency of misogynistic attitudes expressed on Twitter and domestic and family violence incidents that were reported to the FBI. We tracked the number of misogynistic tweets in over 400 populated areas defined by the U.S. Census Bureau across 47 American states from 2013–14 inclusive. Misogynistic tweets not only correlated with domestic and family violence incidents in those areas, but a cross-lagged model showed that misogynistic tweets at Time 1 positively and prospectively predicted domestic and family violence one year later. Results were robust to several known predictors. Our approach shows that monitoring and geolocating online misogyny may provide a cost-effective means for prospectively predicting outbreaks of violence against women, and may also allow governments to disseminate public health resources to reduce violent outbreaks before they occur.

To attend this zoom presentation, please email ethics-wellbeing@unimelb.edu.au