2
$\begingroup$

Is it true, that long answers to questions attract more upvotes than shorter precise answers?

I have seen many beautiful short answers, but they are neglected and overshadowed by the longer ones.

$\endgroup$
9
  • $\begingroup$ Possibly, but there are some counter examples. You might be able to use the Stack Exchange Data Explorer to find out if there is a relation. $\endgroup$
    – Kyle Kanos
    Apr 13 '17 at 18:38
  • $\begingroup$ I have used it and found similar results $\endgroup$ Apr 13 '17 at 18:40
  • 1
    $\begingroup$ Well do you plan on posting your findings? Or are you wanting others to confirm it for you? $\endgroup$
    – Kyle Kanos
    Apr 13 '17 at 18:41
  • $\begingroup$ I just want others views on this $\endgroup$ Apr 13 '17 at 18:43
  • 1
    $\begingroup$ Well can you post a link to your SEDE query so that someone else can use that as a starting point for their own answer? $\endgroup$
    – Kyle Kanos
    Apr 13 '17 at 18:44
  • $\begingroup$ Ya sure.I will.Just waiting for a few answers. $\endgroup$ Apr 13 '17 at 18:45
  • $\begingroup$ So, you 'know' the answer but want others to check? Seems, well, like key a homework problem... Lets face it - some people are more verbose then others. $\endgroup$
    – Jon Custer
    Apr 13 '17 at 23:54
  • 10
    $\begingroup$ The body of this question asks a very precise data-answerable question, and in the comments you claim to already know the answer. I'm therefore confused as to why you are asking in the first place, or why at the very least you don't share the data analysis you've already done. I can understand wanting to engage with people over nontrivial voting behaviours, but this is not how you do it. Share your data and explain and justify your opinions, and then actually ask for engagement. You say you've "seen" examples of X behaviour? Then link to them. $\endgroup$ Apr 14 '17 at 0:40
  • 1
    $\begingroup$ This is a difficult question to address as it's currently worded. It asks if there is an overall trend, which completely devalues any specific examples of short answers being more highly voted. Plus, there are far too many confounding factors to get a very clear picture from the SEDE. For instance, often times a longer answer means more time, care, and research was put into it, which makes it more likely to be voted up because it is more correct and complete. To find the proper trend, you'd need to only compare posts of equal content value but differing length, which is difficult to determine $\endgroup$
    – Jim
    Apr 17 '17 at 13:00
4
$\begingroup$

I wrote a SEDE query for the task. The results:

enter image description here

What we can see here, is the following:

  • Yes, the average score of the answers grows monotonically with their length (I grouped them in 20% growing groups).
  • The length distribution of the answers follows a very beautyful Boltzmann-distribution with a maximum of around 800-1000 bytes.
  • But, the average score per text length, measured in $\frac{\mathrm{upvote}}{\mathrm{kByte}}$, shows an obviously shrinking tendency, and sets to a roughly constant $\approx 1.2 \frac{\mathrm{upvote}}{\mathrm{kByte}}$ from around 3 kByte answer length.
  • If you post 2 1kB answers, you can count with around 50% more upvotes as if you post a single 2kB one.
  • There is also a very high per-length reward for very short answers (in the <200byte region). But, in my opinion, it is practically impossible to write ten 100byte-length answer in the time as you would write an 1000byte one.
  • Note: a significant part of the work behind answering a question is not typing it in. This part is not measured in this stat.
  • This stat doesn't count with downvotes and with deleted posts. The first would be easy but I didn't count with them. More exactly, this query counts only the voting scores. The second is impossible as the content of the deleted posts are hidden by the SEDE, thus we can't measure their length.
  • This query measures only the vote score of the answers, it can't measure their actual value. We could get a better estimation of it by overweighting the votes of the high-rep users in the average counts, but also the induvidual votes are hidden in the SEDE.
$\endgroup$
0
1
$\begingroup$

I think the answer is partially yes, partially not.

More exactly, we have different secondary effects which can increase the score of the answers. All of these are my subjective experience, I don't have SEDE queries to support them (although I could produce some if needed):

  1. Its general quality on the first spot, correct grammar & spelling
  2. The reputation of the poster (additional psychological effect if he has some golden badges, too)
  3. Including a beautiful picture in the answer
  4. The answer is long
  5. The answer contains many complex formulas

Some of these can be considered contraproductive, some of them not.

Many voters don't read through the whole text, but they vote it up if they think it looks well. These secondary effects are mainly about their votes.

All of these secondary effects are overshadowed by the primary effect of the voting score: write good and useful answers. All of the users of the site with the highest reputation are really good in this, and not in (1), (3)-(5).

$\endgroup$
3
  • $\begingroup$ come to think of it, being really good in (2) might not just be a necessary condition. It might also be a sufficient condition of being one of the users of the site with the highest reputations. I should apply for a research grant to study whether or not there are any other conditions of that status. Might get me a publication $\endgroup$
    – Jim
    Apr 17 '17 at 13:23
  • $\begingroup$ @Jim I wrote another query, it shows the average answer score by the estimated log2 reputation of the poster (at the time of posting the answer). Most answers are coming from users with 2000-4000 reputation, and the average score of the answers grows monotonically, and (very) roughly logarithmically with reputation of the answerer until around 5. $\endgroup$
    – peterh
    Apr 17 '17 at 15:33
  • $\begingroup$ @Jim More exactly, the average answer score is around 0.7*(digits of reputation). It could be made better by a more exact reputation estimator and by filtering out the long-term votes, but these probably wouldn't cause a significant change. $\endgroup$
    – peterh
    Apr 17 '17 at 15:45

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .