This study seems to have failed to compensate for the most basic of confounding variables. For instance their control group was specifically chosen not to be a

This study seems to have failed to compensate for the most basic of confounding variables. For instance their control group was specifically chosen not to be a control, in the nominal sense, but other groups that were likely to be banned or otherwise reprimanded. Why? They showed that the so-called control group and the treatment groups had a slight decline in overall posting activity, but failed to consider the trends for the site at large. What was the overall trend, per user, at the time? Many of their control groups were also banned, some within the window of time they sampled!! The authors also seem to be failing to account for the fact that the entire site began to begin automated censorship of posts with undesirable keywords or from users that had posted in undesirable subs. Many of these posts can not be recovered even from the API which directly taints their data. This also resulted in two further confounding issues. A very non-zero number of users that stayed on decided to post under new names and use language less likely to be tagged by automated censors. I think the important question is what effect this censorship has had on trends of such beliefs, rather than whether the certainly specifically censored words continue to appear. The answer to the former is interesting and the entire question of censorship. The answer to the latter is rather self evident and has little to do with whether censorship is effective, which is itself separate from the question of whether it's desirable.