A lot of discussion around toxic communities online is limited by assumptions, rather than by actual research. This is a useful summary of some work done by Francesca Tripodi, exploring how viewpoints are pushed out of the debate:

Not only is harassment used by others in a community to silence the kinds of expression they don’t like or don’t agree with, but many fail to consider how avoidance also works as a form of self-censorship because of the fear of harassment to begin with. While legislation surrounding online harassment has improved, those frequently harassed are often left with few resources to combat the problem and start avoiding the spaces altogether.

This vicious circle between harassment and avoidance is powerful, and problematic. This is where decent community management — the exact skillset that the big social platforms are desperately trying to avoid deploying — comes into play.

Revealing visualisation in the piece:

And, of course, the reasons for that are complex — but Wikipedia has known about this for years and still hasn't resolved it.