Hacker News Reader

Comment by pg | Hacker News Reader

Don't worry, it's just an experiment.

Since this may cause people to care about comment averages, here's how they're calculated:

    ; Ignore the most recent 5 comments since they may still be gaining votes.
    ; Also ignore the highest-scoring comment, since possibly a fluff outlier.

    (def comment-score (user)
      (aif (check (nthcdr 5 (comments user 50)) [len> _ 10])
           (avg (cdr (sort > (map !score (rem !deleted it)))))
           nil))

Comment averages are updated asynchronously.

View on HN

@pg, i think the average is a bad way of doing it. it discourages people from commenting on low-traffic, or old threads.

the game becomes that of other threaded comment systems where you get more up votes on heavily traffic-ed threads. it should push the long tail of comments further out, and have fewer threads with high traffic. someone trying to game the system might look for threads with say 10-15 points and then make a comment, hoping they get above the average. it's like investing in DOW companies during an economic upswing.

your karma actually should not be purely based on uptick, but should represent some amount of risk you took, and then measure that against how well the community responded to it.

so commenting on old or low-traffic threads could still be of value.

I think your 'just an experiment' mentality works well to avoid a lot of premature criticism. Also helps you mentally detach from taking things too personally.

I am going to experiment with labeling things as experiments.

I don't think the criticism is premature. The has been visible as long as I've been here.

And it definitely discourages commenting in low-traffic threads when the average isn't normalized for viewers.

I hope you realize that this is an actual experiment and that the language usage isn't just being used as a way to diffuse criticism.

By that theory you can also ignore the lowest-scoring karma (if negative karmas are being added), as some might have had bad day or unlucky thrashing.

If you continue along that line, you end up using the median comment score... which might incidentally be an interresting experiment.

Some thoughts, assuming most of everyone's comments are ignored (success being a stronger function of practice instead of inborn genius) and that commenters with high means are probably being pulled up by a number of outliers.

   1. Median obviously is less suspect to a small number of very
      popular comments which probably describes many users pretty
      well. It's the "betting value" for a game where loss is 
      linear with error.
   2. I'm not sure whether median would be higher or lower for 
      most. Poisson distribution suggests higher medians, but 
      many probably have a bunch of 1-karma comments to weight 
      the origin.
   3. A HUGE psychological difference is that median would be 
      discreteness. Instead of seeming to be a smooth value 
      related (probably naively) to community engagement and 
      improvement it'll look more like a set of clubs. 0-medkarma 
      could be a rabble-rouser, 3-medkarma is consistently 
      meaningful. 10-medkarma means maybe you should run for office.

Your third point is compelling.

Psychologically, it diverts attention away from a specific number, and toward a general concept of "goodness", segregated by club. Then, rather then concentrating on getting the next karma point, users try to enter certain club, and each club has a general identity to it (with the top being those who provide thoughtful, intelligent responses, and typically only speak when they have something to say).

Creating different clubs (creating classes of users) isn't arguably bad so long as there is easy mobility between classes based on merit.

The natural question is how and whether to indicate the "class" or "club" a user is in. Make it public? Indicate it in comments? PG tried a red dot next to high-karma users some time ago, then scrapped it; maybe there's some other way to reward users based on club achieved.

This is exactly how I feel, but am not sure that's a developed enough thought yet. I think discrete 'club' separation is a powerful force for social organization, especially if there is a tangible benefit for inclusion. I also think that median comment score probably isn't completely tracking the kind of behavior that should earn entrance into such clubs.

Both thoughts depend on figuring out more completely what club membership means; it's something I can't flesh out well despite having a feeling it might be beneficial.

Back when he capped downvoting to -4, I would definitely disagree with you, because that prevented serious negative outliners.

What if the average rounds everything under some threshold (say -10) up to that threshold? So your total karma can get totally trashed by one heavily-downvoted comment, but there's a limit to how badly your average gets hit.

But really, I'd guess that 99% of comments that go below -10 are clearly intentionally trolling, so it's probably not an issue.

That's not true - bad jokes also get you a bunch of downvotes. Downvoting is how you emote a groan on here.

By the way, these "experiments" are pretty freaking cool: the fact that they could be done by just typing stuff into a REPL on the server isn't something I had ever considered before.

I think typing new function definitions into your live site seems very appropriate for a place called Hacker News.

Huh? I thought that was a snippet from one of his source files.

As part of your experiments, have you also changed the downvote threshold, or modified it to be based on average karma, not total karma? I ask because I notice that I've lost the ability to downvote - not that I'm especially irritated, as I'm of the opinion that downvotes are generally not an especially good thing, as they can encourage negative behaviors, but just wondering.

http://news.ycombinator.com/item?id=1853529

Downvote threshold now 500. Glad I got to the old threshold (201) in time to see how it worked.

Think my possible upvote demographic for this comment is the set of all Hacker News users between 201 and 500 karma.

I'm dissapointed its been increased. I try and only speak when I have something especially interesting to say. But I regularly vote and in my opinion, vote sensibly. Its a shame to be penalised for quietness.

That's very likely in keeping with another recent thread here, entitled "Tell HN: Uptick in mean-spirited comments" ( http://news.ycombinator.com/item?id=1852560 ). One of the threadlets suggested possibilities like eliminating downvotes altogether and replacing them with the option to flag comments as inappropriate, the idea being that "no up" is a good enough disincentive for most purposes, and that downvoting should be considered a drastic measure. I think I've used the privilege twice, and I'm still not entirely sure that one of those downvotes was truly righteous.

I just noticed that, too. Presumably there is some new, average-based threshold for downvoting in this rep grind system?

For long-time users, isn't this number pretty much going to be static?

I'd love to have a running average of my last 10-20 comments rather than ALL of them.

Mine varies by a couple of points over the course of days, generally. The occasional +50 comment can bump things up, as can refraining from making a bunch of +1 or +2 comments.

Yeah, we have to build up an intuition as to how the new metrics are useful to us as commenters and a community... and in the short term I think most of us won't be affected by it at all.

Reply on news.ycombinator.com