I think the other factor is there are 'types' of people that will comment online, and also only conditions in which some people will comment online, so we get a weird filter.

It's easy to think that you'll get a general cross-section of the population posting, and that they're relatively stable when commenting. But we can't see when people are drunk or under the influence of things, who the bots/psyops bad actors are, or the ages etc. Makes for a frothy mix. But I agree, much healthier to just avoid it and just interact in the real world where possible.