According to the data, FBUs (future banned users, aka trolls) are significantly more readable in their writing than NBUs (never banned users).
That's certainly surprising.
Also of note is that FBUs tend to reply more than NBUs, which actually meshes with some features of HN -- the inability to reply to a nested comment from within the nest (one has to click on the comment itself to reply -- though this feature has always been somewhat ambiguous and not really deterrent of impulsive replies; I don't even notice it anymore).
Also, I think HN deflates the score of posts with many comments, a strategy supported by this line in the paper: "Prior work also identified post frequency as a signal of a low quality discussion (Diakopoulos and Naaman 2011)."
An absolutely fascinating paper. What are the consequences, from an IP standpoint, of taking this paper and implementing features based on its conclusions?