Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 158 Stack Exchange communities.

What is meta?
Here's how it works:
  1. Any Stack Exchange user can ask a question
  2. The community provides support, votes on ideas, and reports bugs
  3. Your voice helps shape the way Stack Exchange operates

Assuming the posts on SO follow a normal distribution of quality, as seems likely, should not one downvote roughly as many questions/answers as one upvotes?

Further: It seems that people here (and I'm suprised by this) see downvotes as "negative" or "punishment", when I see them as marking questions that by any sensible measure don't make sense and answers that are simply technically wrong.

Clarification: Perhaps I have not made myself clear. I am not asking about how to explain observed mass behaviour. I am asking about how an individual should best behave.

share|improve this question
I actually took a bath and thought about your question and then answered :) – Ólafur Waage Jul 3 '09 at 21:45
Most people don't think too well when their head is hot :-) – nb69307 Jul 3 '09 at 21:50
@Neil: So do you believe there are as many technically wrong answers on SO as there are technically right answers? If not, why would expect to give as many downvotes as upvotes for them? I've read, re-read and thought about your question, and I still don't see why you'd expect to see a balance there. – Jon Skeet Jul 3 '09 at 23:04
@Jon Actually, I believe that there are probably a lot more wrong answers. There are after all an infinite number of ways to wrong on a technical subject, and only a few ways to be right. So I'm kind of bending over backwards to be fair. – nb69307 Jul 4 '09 at 19:53
@Neil: Given your comment about C questions in my downvote question, I wonder whether you happen to be seeing the worst of SO. Note that one does not have to have the best answer for it to be correct or helpful (and not inaccurate) in itself. – Jon Skeet Jul 4 '09 at 20:37
@Jon No, I'm not talking about the difference between better and best. For example, for any C++ question that litb & I both provide an answer, his will 90% of the time be better than mine. However, both will very probably be techichally correct, and if they are not, both of us will provide fulsome corrections. I am talking about answers that are flatly wrong. And there are a lot of them. – nb69307 Jul 4 '09 at 20:50
Neil: I was commenting on your claim about probability, although not particularly well, admittedly. I think we can only go on what we've actually seen. Just because there are more ways to be wrong than right doesn't imply that each of those ways is as likely to be posted as correct answers. I haven't seen any evidence to suggest that SO as a whole has as many wrong answers as right ones. That may be true in the C++ and C tags, but it's definitely not my overall experience... nor do I see how your idea of a "normal distribution" would suggest that either. – Jon Skeet Jul 4 '09 at 23:20

Actually, the vast majority of people upvote a lot more than they downvote. There are really two reasons for this. First, people like rewarding good behavior much more than they like punishing bad behavior. Second, the rep loss from a downvote provides some disincentive (as opposed to upvotes being free).

Stu Thompson has created an interesting graph that plots upvotes (Y axis) vs. downvotes (X axis) of high-rep users:

Purty Plot

share|improve this answer
Hmmmm.... I think the graph is mislabelled; I've left a comment on the blog ;-p – Marc Gravell Jul 3 '09 at 21:48
The X-axis is downvotes, Y-axis upvotes. The axes aren't equal - the grid is by 100 for downvotes, but 500 for upvotes. However, I don't see that it's mislabeled..? – Kyle Cronin Jul 3 '09 at 21:53
Jon has cast <3k upvotes; I've cast 4662 upvotes. Plus the dot to the left of tvanfosson looks just a shade bigger to me; I'm thinking the red dot at the top is me... – Marc Gravell Jul 3 '09 at 21:58
@Marc: Sounds about right to me. Interesting to see we're on a very similar ratio though :) (~22 for me and ~18 for you.) – Jon Skeet Jul 3 '09 at 22:01
I'm glad to see I'm so close to Rich B. Perhaps we can finally bond! – nb69307 Jul 3 '09 at 22:02
And can I point out that a quick & possibly incorrect look at the users at around the same rep as me seems to indicate that I actually UPVOTE more than most people, given the how long we have all been on SO. Perhaps a definitive graph of this could be produced by those with time on their hands? – nb69307 Jul 4 '09 at 20:28
@Marc: fixed. Sorry about that. <:o – Stu Thompson Jul 7 '09 at 22:09
@Kyle: scaling the axes allows us to keep more data points on the graph. I really wish I could do better/automated axis labeling and data point labeling within Excel... – Stu Thompson Jul 7 '09 at 22:13
@Niel: As noted, the axes are scaled in the comments. Your voting ratio is effectively to 1:1, as noted in my blog entry Kyle has linked too. (And that is very much why Rich B's data point is so "interesting".) – Stu Thompson Jul 7 '09 at 22:20

It depends on how you view a downvote. I've just written a long post about this but the relevant point is that I only downvote an answer if I think it's wrong or misleading.

Now there aren't as many wrong/misleading answers (in the questions I look at) as there are "moderately helpful" ones - or genuinely excellent ones, even.

The quality may well follow a normal distribution, but not centred around "neutrality of helpfulness."

share|improve this answer

Yes posts should follow a normal distribution of quality (or a minor deviation there of) but your 2nd assumption is wrong due to these factors

  • A downvote is a negative act and it gives off a negative effect for the target and you.

Due to that it effects you as well, that rule alone will severely affect the outcome of any result drawn from this.

  • Based on the descriptions of up and downvotes. An upvote is a helpful answer and a downvote is not a helpful answer.

A person (when given a choice) of no effect on him and a positive effect on a target and a negative effect on both parties. Will be very inclined to choose the neutral/positive choice.

The scenario where he picks the negative/negative choice is not within the normal boundry. So a user is far more likely to give an upvote to a mediocre answer than he is to give a downvote to a bad answer.


Of course this is subjective. And one set of people will look at a rule in a system in one way and others in another. The above is my take on it, from how I perceive the rule and what I've seen others go by.

Even though a downvote does little, it is still a negative tool (albeit an important tool) But it's implications are negative even though it's impacts on the system is a positive one (since it separates the good from the mediocre from the bad)

share|improve this answer

First off, how do you measure quality?

Secondly, answer quality will, over many questions, be normally distributed given the central limit theorem (if anyone is interested in why).

Third, they might be normally distributed but the mean is not at 0 (no quality). The mean will be positive quality so if the mean is "great" do you want to treat "good" answers as "negative" (being lower than the mean) compared to "awesome". This would be particularly disastrous if "good" answers in this scenario had negative net reputation.

Fourth, quality is only one factor (and a vague one at that). On some topics you will find people voting on whether they agree or disagree with the opinion stated. And popularity of opinion is not the same thing as quality.

Fifth, there is a time factor in voting. The closer an answer is given to question the more votes (up and down) it is likely to receive. Non-parametrically speaking, this is probably more related to ranking than time.

Lastly, because there are various factors in play, you'd probably want to model this as some kind of multivariate regression analysis.

share|improve this answer
+1 The third point is what I was going to say. While it's distributed normally, the average quality is not 0. – xmm0 Jul 4 '09 at 18:20

The distribution of quality maybe is normal, but I doubt that. In an excellent community such as SO with professionals I would expect more good then bad answers. I doubt that because I want to do so. If this community was really bad, what would be my point in spending my time here?

As I tend to improve much more on positive feedback, I rather encourage by voting up then discourage by penalty of downvote.

Btw, I am a mathematician and therefore do understand the question.

And don't forget that downvotes cost the downvoter, and therefore many are discouraged to downvote.

So there are humane and psychological reasons for upvotes to outnumber downvotes, rather then statistical ones.

share|improve this answer
I don't see why you believe this. There is no test on becoming a member of SO, so a question/answer is as likely to be posted by an expert as a novice, or a genius as an idiot. That's what the normal distribution implies. The question is, faced with such a question or answer, what to do with it? – nb69307 Jul 3 '09 at 22:19

Vacuously, at least half the answers are at least as bad as the median. Even around half of Jon Skeet's answers are as worse than his medianly good answer. Not every answer is voted on, and the cut off in terms of standard deviations is not equal.

The appropriate law here is Sturgeon's Law (interesting the Wikipedia entry seems to disagree with the OED without any citations - what crud (improved it)). 90% of answers should be marked down. I guess we are used to crud, but I shouldn't think more than 10 of answers deserve an upvote.

(FWIW, my most forceful reason for downvoting is because the answer is dangerous, followed by wrong, misleading, idiotic and mental.)

share|improve this answer
I honestly don't follow your reasoning. Whether an answer is a helpful or not has nothing to do with medians. I'm not sure you can apply that 90% figure to a community that is active in highlighting, weeding out and (sometimes) improving errors... – Marc Gravell Jul 5 '09 at 9:47

If you could quantify the "usefulness," which is being measured in voting, you probably do see normal distribution. However, that does not make the mean usefulness of entire Stackoverflow to be zero mean neutral: neither useful nor harmful. We have self-selecting group of people who is going out of their way to tell people some information. The mean usefulness of the answers and questions are way higher than zero.

To put it in terms of money, say some people make more than they spend while others spend more than they make. If you look at the general population, there's no guarantee that the median population spend exactly they make (hopefully not).

Besides, even if the distribution of the perceived "usefulness" of the questions/answers are distributed in Gaussian way, that does not guarantee symmetric human reaction towards "useful" behavior and "harmful" behavior even discounting the penalty. Geeks are still subject to basic human psychology.

So, no, upvotes and downvotes won't be symmetric.

Edit: How does this have anything to do with how an individual should best behave? If 50% of the people told misleading and/or false information, the society wouldn't function, let alone programming Q & A.

Edit2: Another factor is the asymmetry of knowledge distribution about a given topic. By nature of public voting, the topic of the question/answer first needs to be understood by the electorate. Most people can appreciate funny jokes and cartoons, but not too many people even be interested in some weird hack in Scheme or writing Mozilla plugin using Delphi. That may not reflect "favorite joke" question 1000 times more useful than others; rather it heavily reflects that there are 1000 times more people who gets it.

share|improve this answer
There are roughly 10 to 12 up votes per down vote, for the record. – Stu Thompson Jul 7 '09 at 22:25

should not one downvote roughly as many questions/answers as one upvotes?


share|improve this answer

You must log in to answer this question.