change results from comment thread interactions, but this is only a possible interpretation that can be neither ruled out nor proven. Evidence for such change cannot be found within individual comment threads. Again similar to Observation 1, if the Facebook algorithm creates a bubble effect where hate speech anchor posts are more visible to people who agree with them, then anyone who attempts counter-speech under such posts is more likely to be swamped by attacks. This may make people reluctant to engage. Hate speech anchors posts may be less likely to be visible to people who might respond in the comments due to the Facebook algorithm. However, neutral anchors, such as news stories about controversial topics, may be more likely to be posts encountered by people of differing views, i.e. both the people who like to post hate speech and the people who respond with counter-speech. Implications The effect of counter-speech in comment threads needs to be carefully evaluated, for its impacts on people including both the person who originally uttered the hate speech, members of the target population, and others who see the hate speech and the counter speech. On the other hand, if the interpretation of Observation 2 is true that there are people reading the comment threads closely, different people could be affected by the counter-speech, such as: members of the target population, feeling good because someone opposed abuse on their behalf; and members of the majority, who could feel emboldened to speak out later, if they see counter-speech, or more afraid to speak out, if they do not see any counter-speech. But these effects cannot be confirmed by this study from online data alone. What the material we analyzed did show is that the arguments online have a tendency to escalate, sometimes with the counter-speech becoming hate-speech itself. This underlines that any attempt to do direct response to hate-speech with counter-speech should be done carefully. It could also have side effects for the counter-speaker, who may end up traumatized by the hate-speech they are dealing with. It may also inadvertently promote the visibility of negative contents by increasing engagement with the original post and driving up its value in the calculations of Facebook’s algorithm. 3.4 – Views on how to manage hate speech online Observations Interview and focus-group discussions provided some insight into how people regard the fact that speech targeting them is common and how they manage this fact. Almost all of the interviewees said such speech is best handled by not responding at all. Four reasons were identified by different interviewees. Some believed that the persistence of hate speech on social media means their response would not make any significant change and thus it is pointless to 12