I'm a bit skeptical of the "power imbalances" approach. I don't think it becomes fine to say "off with the politician's head" just because the speaker is powerless. Additionally, if "power imbalances" are a core factor in your decision-making, that incentivizes people to argue that they are powerless, and "oppression olympics" discussions of this kind are hard to adjudicate. (For example, arguing for your own powerlessness works better if you already have some power. Impoverished people in developing countries have very little power, and often lack the time, money, and language skills necessary to communicate with us.)
If I was writing a terms & conditions for a site like Facebook, I would try to distinguish between speech that aims to inform and speech that aims to intimidate. In general, I think the bar for censoring the former should be a lot higher. But it's not a hard rule, e.g. I wouldn't censor intimidating speech like "if you don't stop I'm gonna call the police".
I'm a lot more willing to censor speech directed at children, relative to speech directed at adults. Censoring speech directed at children doesn't run into the problems I discussed in my comment. (E.g. the circular dependency problem doesn't apply, because adults can still argue about what's OK to say to a child.)
If I was writing a terms & conditions for a site like Facebook, I would try to distinguish between speech that aims to inform and speech that aims to intimidate. In general, I think the bar for censoring the former should be a lot higher. But it's not a hard rule, e.g. I wouldn't censor intimidating speech like "if you don't stop I'm gonna call the police".
I'm a lot more willing to censor speech directed at children, relative to speech directed at adults. Censoring speech directed at children doesn't run into the problems I discussed in my comment. (E.g. the circular dependency problem doesn't apply, because adults can still argue about what's OK to say to a child.)