Figure 4

Predictive analysis of hate and counter speech. The relationships between current and future proportions of hate and counter speech for three periods of study. Shown are the results of VAR models for the periods before RG, between RG and RI, and after RI. Red (blue) arrows show statistically reliable effects (** and *** represent \(p<0.01\), and \(p<0.001\) respectively) on hate (counter) speech, and gray show nonreliable effects (\(p>0.05\)). Full arrows show positive relationships, and empty arrows negative relationships. Before RG was established, the effects of hate and counter speech were relatively balanced. After the occurrence of RG and before RI, past hate and counter speech became related to more hate speech going forward. After RI was established, past counter speech now negatively predicted future hate and positively predicted future counter speech