> Twitter is structured in such a way that direct censorship can't be differentiated between
Twitter provides data to researchers, at least in some situations, and may provide data about these events.
> Differentiating between "left" and "right" political posts is not possible from an objective, programmatic standpoint.
AFAIK, there are well-known methods in political science for defining left and right.
> AFAIK, there are well-known methods in political science for defining left and right.
Small problem: political science isn't computer science.
Small problem: if the data they provide to the public can't be trusted, then why would the data provided to scientists be any better?
The parent comment is valid. hackuser isn't asking for a double-blind study with p=0.05 and n=[all Tweets ever produced] ; she or he is asking for any evidence whatsoever to substantiate the claim.
Consider:
- Twitter is structured in such a way that direct censorship can't be differentiated between somebody's finger on a button and technical difficulties. (Did this hashtag disappear from autocomplete because someone removed it, or because the backend index is messed up?)
- Differentiating between "left" and "right" political posts is not possible from an objective, programmatic standpoint.
Political identity doesn't work either. Someone who's an unabashed Republican tweeting about, say, abortion, there's a very high likelihood that whatever they say about it is negative regardless of the content of the message, but how do you get a computer to understand that without a lot of false positives?
- Nobody from the companies is ever going to admit to hiding/removing content for political reasons. (Barring whistleblowers of questionable veracity, c.f. Facebook)
That only leaves generalities and user reports to work from, which have their own reliability problems. We're talking sociology/political science at the end of the day, and it's opinions all the way down, with very little hard data.