Sounds like it would have self reinforcing feedback problems. I've certainly found that placement and traffic in the threads they're posted in is at least as powerful a predictor of my comments' upvotes as my perception of their quality, probably more so.
I think early comments are more likely to attract karma than later comments. Have you noticed other patterns?
You could be right about self-reinforcing feedback-- people would be less likely to comment to articles with low average comment karma.
Well, some threads simply get more traffic than others; any post that generates abundant discussion is more likely to generate highly upvoted discussion, which means threads like Curi's Critical Rationalism threads would appear as having "high quality" discussion when in fact they contained very little worth reading, and may even have been a response to a deliberate trolling.
And of course, it's well known that Eliezer has a karma wake, since any comment that his own comments call attention to will get a large influx of readers. Recently, one of my own comments which had been stable at +2 till a few hours after I posted it shot up to +7 when Eliezer replied to the parent. Other posters like Yvain may generate similar effects.
I would say I rarely downvote, partly because I'm not quite sure how one should view karma; one alternative is that it should reflect some form of very crude statistics of peoples opinions of a post/comment, no matter if the post has -4 or 100 you vote it up if you think it's a good post or vote it down if you think it's bad one. Another alternative is that karma reflects how many points a post/comment should have, that is if a comment has -2 and your preference value is 0 you vote it up.
Could someone help me clarify how one should view karma?
People express both opinions, or more often a mix of the two. In practice if a substantial number are acting according to the later one, that is enough to make it dominant.
If people act according to the simple strategy, and assume others act according to it, one can have disproportionate apparent influence by strategically voting towards a target. The more people that act strategically, the more likely it is votes according to the simple strategy will get cancelled out, even by people who substantially agree.
Because of these incentives, I think it best to mostly vote towards a target.
Thanks, good to know there isn't any outspoken rule of how one ought to view karma in that aspect. But then when are you allowed to downvote? Obviously when someones argument is unsound or when a comment do not actually add something to discussion, but is it "OK" downvote if you just strongly disagree, for example if someone makes a good case for eating babies, that is to say based upon that persons assumptions about the world eating babies is (EDIT: not) a bad thing, lets say "since babies are not conscious it's okay to eat them" (I'm not implying that the statement harbors any truth).
when are you allowed to downvote?
That's a good question, so I upvoted this post.
Then I decided to downvote the first comment you wrote on or after October 15th, because fuck your first comment on or after October 15th! unless in it you advocated eating babies.
Then I read it and decided that advocating pesticide was close enough, because weeds are people too, and undid my downvote.
Karma divided by page views would be a better metric.
That would penalize threads that were so interesting that you would want to read the comments there multiple times, and even revisit them in the future.
Generally it'd be a bad metric if it encouraged you not to revisit the threads you liked because you didn't want to lower their score.
I really don't want the thought in my head, as I browse lesswrong, that my reloading a page is going to affect something. Or choosing to read a thread and then not bothering to comment or upvote.
This is a bit complicated, because LW does not currently track how far down the page people scroll, or how quickly they do so, so there would be a new set of weird incentives around posting in places where readers read the comments more thoroughly.
I think the popular recycling ad slogan "don't mess with karma" applies here.
Losing one's purpose is already very easy on LW, given that the karma score provides an instant emotional feedback. Some get worried or even obsessed with it, losing sight of what brought them here to begin with: learning rationality and being a part of the community. Some even delete their accounts after a negative reaction to their posts.
In short, the karma feedback is already strong enough, make it even stronger would render it less useful.
Dividing by number of comments would penalize posts for being interesting and productive. I consider the number of comments a better indicating of quality than the karma rating.
Inspired by this: what if the link for comments to a post included a comment karma total divided by the number of comments ratio?
This would also be handy for comments with an individual score below threshold.