Lies, Damned Lies, and Social Media Following Extreme Events

With the increased use of social media in crisis communication following extreme events, it is important to understand how the public distinguishes between true and false information. A U.S. adult sample (N = 588) was presented 20 actual social media posts following a natural disaster or soft-target terrorist attack in the United States. In this study, social media posts are conceptualized as truth signals with varying strengths, either above or below each individual's threshold for believing the post is true. Optimally, thresholds should be contingent on the (incentivized) error penalties and base-rate of true posts, both of which were manipulated. Separate receiver operating characteristic (ROC) analyses indicate that participants performed slightly better than chance for natural disasters and moderately better than chance for terror attacks. While the pooled thresholds are ordinally consistent with the base-rate and error penalty manipulations, they are underadjusted compared to the optimal thresholds. After accounting for demographic and cognitive variables, the base-rate manipulation significantly predicted sensitivity, specificity, and true response rates in the expected direction for both content domains, while the error penalty manipulation had no significant effect in either domain. Self-identified political conservatives performed worse at classifying false content as false for natural disasters, but better for terror attacks.