As a Bayesian, you should count not a user’s downvote, but P(downvote | user, facts about the post).
If user X downvotes half of all posts, each downvote is 1 bit of evidence.
If user X downvotes one out of 16 posts, each downvote is 4 bits of evidence.
The tricky part is how you combine facts about the post with the prior over all posts in cases where user X hasn’t voted on many of user Y’s posts. What if user X downvotes 1 comment in 50, and they’ve only voted on one of Y’s comments before, and down-voted it? I could talk about how to do that correctly, but in the case of mass downvotes this is by definition not the case.
So, the site should report not sum of downvotes and upvotes, but evidence for and evidence against the utility of a post. Users could then choose to mass-downvote legitimately, knowing that would mean that each of their downvotes would count for less.
(I would be cautious about incorporating the poster’s prior! I think voters already incorporate that in their voting.)
To work properly, this system should have the voting options “upvote, meh, downvote”, so that we can use P(downvote | voted) rather than P(downvote | viewed) or (downvotes / (upvotes+downvotes)). The latter could motivate people to vote nearly everything up so that their downvotes would have more weight. The votes of a person who votes seldom give more evidence than the votes of a person who votes on comment.
(At the very least it would help to display “+3/-2” instead of “1 point”. Yes, I know you can compute that by hovering over the score to get % positive. In which case there’s not much reason not to just display +/- votes all the time!)
As a Bayesian, you should count not a user’s downvote, but P(downvote | user, facts about the post). If user X downvotes half of all posts, each downvote is 1 bit of evidence. If user X downvotes one out of 16 posts, each downvote is 4 bits of evidence.
The tricky part is how you combine facts about the post with the prior over all posts in cases where user X hasn’t voted on many of user Y’s posts. What if user X downvotes 1 comment in 50, and they’ve only voted on one of Y’s comments before, and down-voted it? I could talk about how to do that correctly, but in the case of mass downvotes this is by definition not the case.
So, the site should report not sum of downvotes and upvotes, but evidence for and evidence against the utility of a post. Users could then choose to mass-downvote legitimately, knowing that would mean that each of their downvotes would count for less.
(I would be cautious about incorporating the poster’s prior! I think voters already incorporate that in their voting.)
To work properly, this system should have the voting options “upvote, meh, downvote”, so that we can use P(downvote | voted) rather than P(downvote | viewed) or (downvotes / (upvotes+downvotes)). The latter could motivate people to vote nearly everything up so that their downvotes would have more weight. The votes of a person who votes seldom give more evidence than the votes of a person who votes on comment.
(At the very least it would help to display “+3/-2” instead of “1 point”. Yes, I know you can compute that by hovering over the score to get % positive. In which case there’s not much reason not to just display +/- votes all the time!)