For example, @sgonzalezbailon.bsky.social et al. use news-story level data with a different approach, because they have access to an incredibly large Facebook dataset: doi.org/10.1126/scie...

But, as you suggest, you could separate out classes of stories (by editorial v. ordinary story, by topic, by author, etc.) and then you'd need less data. Would just require a separate step to classify each story by class. You thus need the news story text, although the headline + URL might do.

The article's empirical section examines news organizations as a whole for pragmatic reasons. In principle, one could get ideology estimates at the story-level with, say, stories with a news organization prior. Would just require a lot of data to get reasonable precision of the estimates.

So even in the face of equivalent rates of violence/toxicity, the effects on women/men politicians can be different: the effects of negative behaviors in any workplace are going to be experienced much more intensely if the reasons behind it are understood as sexist, racist, anti-immigrant, etc.

Yes, our paper complements @sandrahkansson.bsky.social & nazita.bsky.social bc we get at a potential mechanism why women are more likely to want to leave politics from threats/toxicity: politicians infer perpetrators' motives differently depending on who is attacked (and who the attacker is)