How do you interpret your “% positive”?
I just noticed that if I hover my mouse over the big green dot with my total karma, it says, “81% positive”. Presumably 81% of the votes on my posts and/or comments have been positive.
I checked out the % positive for everyone on the all-time top 15 list:
Eliezer_Yudkowsky (223304) 94%
Yvain (68331) 97%
lukeprog (52586) 92%
Alicorn (34512) 86%
Kaj_Sotala (30919) 94%
wedrifid (26242) 83%
gwern (25567) 92%
PhilGoetz (22008) 81%
Wei_Dai (19049) 94%
AnnaSalamon (18625) 97%
Vladimir_Nesov (18232) 86%
cousin_it (17073) 90%
NancyLebovitz (14436) 92%
orthonormal (12781) 94%
Konkvistador (11887) 87%
Average = 90.6%, Standard deviation = 4.93%
So I’m 1.95 standard deviations below average for the top 15. Not only am I at the bottom of the list, we would expect me to be at the bottom of the list of the top 39 users. (Assuming these numbers are representative of the top 39 LessWrong users, which is dubious, and that LessWrong users are “normal”, which sounds even more dubious, 97.44% of them have a higher upvote/downvote ratio than me.) I’ve gotten about 6744 down-votes, a bit more than Alicorn’s 6711, but still second to Eliezer’s 15225.
How should I interpret this? I could say that I’m the most-controversial poster on the top 15 list, and be proud of that. But if I’d had the highest %positive score, I’m sure I’d be proud of that, too. As long as I’m extreme in some way. Or if I were closest to the average, I suppose I could also be proud of that.
Before checking, would you guess that the top 15 have higher, or lower, % positive scores than most users?
I think it matters why you’re controversial. When wedrifid is controversial, it seems like it’s for different reasons than when you’re controversial, and it might be worthwhile to contemplate that. It also seems worthwhile to break apart what sort of posts and comments have led to the most downvotes.
For example, my total is 91% upvotes. My book reviews have gotten 100% upvotes, as well as my recent comments on controversial topics. The lowest one of my DA sequence posts got 92% upvotes. On the low side, my MLP fanfic got 70% upvotes, and my attempt to recast a Biblical story as a Bayesian parable got 67% upvotes, and a link supporting adversarial debating got 60% upvotes. I don’t have the time to investigate my comments, but I imagine there would be systematic differences between types.
One of the things I’ve noticed is that some people get downvoted for bike shed reasons; if you’re making an opaque technical point, your comment’s score will sometimes be more indicative of its politeness (which everyone feels competent at assessing) rather than your comment’s correctness (which only a few feel competent at assessing).
Most users measured how? I think it would be most meaningful to look at people weighted by karma- I imagine there are lots of people who post a few times, get downvoted a bunch, and then never come back, who you wouldn’t want to be a significant part of the analysis.
Regardless of that effect, we know that the top posters have the highest number of gross upvotes (modulo a person or two). That strikes me as strong reason to suspect they have higher percentages of upvotes- but it would be interesting to look at the upvote and downvote numbers for the top ~100 posters and see if the rankings change significantly when you move from sorting by net upvotes to sorting by gross upvotes.
On a similar note, the established voting pattern in a thread that you join matters a lot—people establish ‘sides’. If you are willing to rebut particular arguments from the side that is already winning socially, expect downvotes regardless of merit (even if the comment on net ends up positive there will still be some downvotes according to existing trend.)
The articles of mine that get the most downvotes are the ones like this that are provably correct. The more rigorous and logical a post is, the more down-votes it gets, because people will down-vote it if they think they detect a single error in it, and a sizable percentage of readers will misunderstand the argument.
The thing where people down-vote links because they think there’s a community consensus that we should down-vote links—I don’t like that. If a link is useful and relevant, I want to see it. I’m not here to give out brownie points for original contributions; I’m here to learn.
That is similar to the bike shed observation I noted.
One of the things I have noticed about your posting style is that you don’t seem to go out of your way to prevent or reverse misunderstandings. Simply telling someone that they are wrong, and telling them to work it out on their own, is not particularly persuasive, and you should not expect it to either be persuasive or earn you status.
Stating negative claims about others carefully and politely does much to improve others’ reaction to them. In particular, the many contrarians on LW dramatically increase the value of carefulness.
This is especially amusing in light of the recent reverse situation between Phil and Eliezer in a recent rerun.
Can you give a link or a title? I don’t remember what it was.
http://lesswrong.com/lw/5v/church_vs_taskforce/44x
(I didn’t vote on those posts, though I did comment already)
But I do go out of my way to respond to comments and correct misunderstandings, or correct my post if they have found an error in it, more than most posters on this site do. And I try to anticipate prevent misunderstandings; it just isn’t possible.
I think there is a perceptible difference between correcting and reversing misunderstandings. Is that the impression you have?
Can you elaborate?
Much of the difference that I see is attitude, but some of it is tactics. The attitude difference is important for status reasons: others may be much more willing to listen to a “let’s figure out what went wrong” than a “listen harder.” The tactics difference is in trying out more angles of approach, as well as trying to figure out how what you said sounds to others, like you’ve done recently. Changing presentation errors is often as useful as changing factual errors.
Would more be helpful, or do you think that’s enough?
Do you think this applies to all posts of this type, or only to yours? Can you think of an example from another normally well-received poster?
Higher % is better. Duh.
Coming from almost anyone else on this site, I would take a comment like this at face value without further thought.
Coming from you, I find I have roughly equal confidence that you mean it at face value, that you mean it ironically to express its opposite, and that you mean it to be taken at face value despite in fact meaning the opposite. For that matter, I have non-negligible (though significantly lower) confidence that you mean it to cause precisely this confusion in long-time readers.
I thought you might appreciate knowing that.
Keeping karma fixed, lower % is certainly more impressive.
Not necessarily, given that the average score of comments is positive. If you want at least 300 net 30-day karma regardless of % positive, you just need to post a lot, but if you want at least 300 net 30-day karma and 95% positive, you have to be particularly thoughtful about what you write.
Or just stick to posting witty things in the quote and MoR threads.
I disagree.
This from the same person who used to go around saying how he didn’t care if his comments were downvoted, and that extremely negative karma comments were frequently as interesting as high karma comments.
Will doesn’t seem to be trolling lately.
Yeah, this puzzled me.
Interesting. Last time Will was posting regularly, he made a big deal about wanting to destroy his own reputation and implied he might be interacting with some kind of basilisks and/or superhuman entities of unknown friendliness.
With an ‘all else being equal’ in there somewhere.
My own karma is 83% positive—it’s not that I’m controversial (I don’t think I am), it’s that sometimes I get too entangled in negative-sum discussions with various aggravating individuals.
I wonder how to measure, like, most controversial plus/given total karma (above some threshold?) (agh I can’t quite articulate the maximand mixture thing) in a semi-sensible way. (Maybe “most controversial in the top 100” is a decent approximation.) ’Cuz I got 7,400 karma, 69% positive, which I’d figure is a contender. (Then again maybe I’d be disqualified since I went out of my way to be downvoted multiple times, which is kinda artificial.)
I would like to see % positive as a function of time on the site.
This is not a popular position on LessWrong. It’s also simplistic to the point of wrong. Even the most hardcore constructionist ought to admit that it is possible to interpret meaning in a particular context.
In short, meaning can be interpreted and constructed.
Please stop that.
In a very boring sense, “I’m wearing socks” is constructed, in that wearing, socks, and maybe even I are arbitrary labels.
Still, you didn’t construct the meaning yourself when you read the sentence. It was constructed by society, and you appealed to that context to interpret what I meant.
You might respond that the social context under-determines the meaning of words. That’s true. But even so, we mostly understand each other in ordinary conversation. At that level, Phil Goetz’s question is meaningful (if not particularly interesting to me).
I interpret my 85% as “huh, I guess I should think twice before submitting a comment more often”. But overdoing this can lead to the groupthink failure mode.
Hm.
I would expect the typical LW user with a karma score above, say, 1000, to have a %positive of approximately 90%.
Among users with karma < 1000 I find myself wanting to distinguish among “new active users”, “lurkers”, and “long-time active users.” I’m not exactly sure how to define these groups operationally. Long-time active users with K<1000 I expect to have %positive of less than 75%. The others I’m less sure about.
The more I think about this, though, the more it gets cluttered in my head with the basic problem of karma correlating with number of comments posted, and therefore with post frequency and tenure. The statistic I’d actually like to see for users is average comment score and average post score.
Anyway, to answer your question… if the average among the top 15 is 90.6%, it seems I expect the top 15 users to have approximately equal %positive scores to most users above 1000.
Isn’t 50% positive 0 net karma?
* blushes *
Yes, of course. I meant “less than 75%”.
Fixed.
Hmm, 77%, must be my MWI skepticism? Or maybe just my general obnoxiousness.
I’m at a steady (does anyone have a %positive which changes much?) 92%, and I interpret it as I’m doing something right with a slight element of maybe I should do more, like figure out some top level posts to write, though actually I don’t know whether they’d affect the quantity of karma more than the %positive.
When I clicked on the title of this post, I was expecting something technical about statistics.
I’m at 62% (+81 total.) I imagine the people with the highest % scores stick to mostly saying stuff that is obviously useful or interesting, though if they get recognisability they might be able to get away with more. It’ll be interesting to go back and see what gets what % in my past comments.
edit: Is there an easy way to find my older posts? I can only go back a few pages if I click my name on the right.
To find your older posts use Wei Dai’s tool.
I’d imagine the people with the highest % are the people who stick to making amusing comments in the MoR threads.
Both of those are factors… as is the simple fact that people who have participated the most have learned more of the kind of things that people here value. (This is the same when joining any social group.)
You’re #1! You’re #1! You’re #1! Your score is even lower than mine. I’ll have to try harder.
You have to decide how you want to interpret it. Maybe you say a higher percentage of stupid things. Maybe you say a higher percentage of controversial things. Maybe your style rubs people the wrong way. Maybe those more likely to rate are biased against you.
(That’s actually a real weakness of the scoring system—potential bias of those who vote more.)
But my real question is what do you hope to accomplish by interpreting the score? If we know that, we might better know how you should interpret it. The score itself is really very limited information that marginally constrains the probabilities of people’s opinions of you.
I think you’re better off looking at your posts, and seeing how you feel about them. Look at the high downvoted posts, and evaluate them. Are you proud of them? Ashamed? As long as I’m proud of the post, I consider downvotes bonus points. I try to avoid giving myself those bonus points when people are just annoyed at my tone. Tone is an unfortunate confounding factor, particularly for me.
The ratio of downvotes tells me how meta-contrarian I’ve been. At least so I thought until recently, now I’m less sure: It could also be someone with a grudge and a few suckpuppets.
Luckily I am 6-7 sigma above normal … in self-aggrandizing (it’s the highest I can go given the number of living humans and while still leaving place for a handful of similarly talented individuals), so I can live with it. I can live with it.
Now that I’ve got that out of my system, to the point: What has been surprising to me is how few readers make use of voting in the first place. At first I thought it’s that many comments just don’t get read a whole lot, until I saw surveys with large (in the dozens) numbers of participants, yet whose vote count was barely touched.
Then I thought that it may be that oftentimes it’s just that a large number of positive and negative votes cancel each other out, giving an illusion of sparse voting participation, but that can be ruled out by inferring the number of votes using the percentage (the solutions are mostly unique with no LCD issues), or by ripping those numbers straight from the html source, as Vladimir Nesov pointed out.
Now I’m uncertain as to the relative importance of some technicality in the voting system contributing to said phenomenon, such as lurkers with no karma not being able to vote much, versus readers just not wanting to actively vote. Maybe reading is often more of an outwardly-passive process for some? I do vote a lot. It gives me fuzzies.
However, it should be safe to say that a karma count of e.g. −1 for a comment that’s probably been read a dozen times doesn’t signal anything much different from a comment at 0, i.e. small karma differences get overinterpreted (“Why did I get downvoted, did I break some norm?”).
When someone does vote, there are so many factors at work, it’s very hard indeed to say which exactly caused the downvote. I realize there probably won’t ever be some anonymous-feedback feature built into the commenting system, such as “hover above the vote count, see some anonymous comments explaining the vote”, but it would make sense.
Abg bayl jbhyq V rkcrpg gur gbc 15 nyy-gvzr gb or jvyqyl aba-ercerfragngvir, V jbhyq or fhecevfrq vs ‘gur gbc a nyy-gvzr’ jnf pybfr gb ercerfragngvir sbe a yrff guna 50% bs npgvir hfref.
Why is this is in rot13?
Why did you rot13 that? You just wasted a minute of my life.
… Because of anchoring bias. Is there a reference to when predictions which are requested should be plaintext?
Oh! Sorry. That makes sense. Is there a lesswrong built-in rot13 decoder? (That’s why it took a minute—I was searching for documentation about that.)
I’m not aware of one, but I personally find rot13.com to be non-onerous. New-tab, copy, paste, close.
Wow. I’m so used to just copying and pasting rot13′d stuff that I didn’t even notice how not-necessary it was.
In both Firefox and Chrome (at the very least) you can find plugins that will rot13 texts by right-clicking and selecting an option in the popup menu. On the whole that’s much more convenient than going to rot13.com or such, and saves time in the long run.
upvotes are mostly noise, downvotes are interesting.
Really?