32

From time immemorial, prominent placement for accepted answers has been baked into Stack Exchange. The reason is that, by accepting an answer, the OP is saying "I did that and it worked for me; I don't need any more help". That signal is valuable. So we pin that answer to the top of the list, so viewers will see it first. (Except for self-answers, because of concerns over gaming the system.)

There have been lots of meta posts about whether this pinning is actually a good idea when there are other, higher-voted answers. No really, lots. For example:

This cluster of problems has been troubling bluefeet ♦ and probably others on the team. Discussions of moving the accepted answer are controversial, even if we provide ways to jump to it.

All of this raises what, to me, seems a more basic question that we ought to answer first: what does it actually mean, today, and in practice rather than in theory, when somebody accepts an answer? 'Cause that's the signal that we'd be disrupting if we change how pinning works.

I'm dubious about the theory, in two ways:

  • On many sites that weren't even a glimmer in Joel's eye when he launched Stack Overflow, answers aren't actually testable or verifiable. I think that sometimes people pick answers that worked for them in some fuzzy way, and other times they pick answers that they like for their own reasons. Subjective-leaning sites can be especially prone to this (as previously noted).

  • On the parts of the network I frequent, people still answer questions that already have accepted answers. (I had been under the impression that this doesn't much happen on SO, but I was wrong. More on that in a bit.) So the fact that the original asker has been helped might not mean much to the community, which is supposed to address the question for future readers too.

I don't know how to systematically investigate my first point, but we can science the second one. I used this SEDE query (h/t msh210) to answer the question: how often do questions with accepted answers get new answers? (The query reports separately on positively-scored new answers, but this isn't a big difference on most of the sites I tested.) Deleted answers don't count in this query.

I ran the query on the top five sites by number of questions and a smattering of other non-tiny sites. I was surprised by what I found:

  • Approximately half of the questions with accepted answers got newer answers on: Stack Overflow, Mathematics, Ask Ubuntu, English Language & Usage, Emacs, and Chemistry.

  • Approximately one-third of the questions with accepted answers got newer answers on: Super User, Server Fault, Programmers, Mi Yodeya, and Skeptics.

  • I didn't find any in my sample with drastically higher or lower proportions.

That's a lot of answering for questions that the OP said he's done with.

So, to restate the question:

What signal do we intend acceptance to deliver? What are people who see the pinned answer with the green check mark expected to do with that information? How should it modify behavior?


Here's the data I collected:

Site            Qs w/accept     w/newer A       A upvoted
SO              5891756         2860654         2277478
Super User      131019          47618           36618
Server Fault    107697          34887           27483
Mathematics     279096          169396          150686
Ask Ubuntu      70594           35399           27346
English         34726           17995           15557
Programmers     27608           9788            8740
Mi Yodeya       6342            2532            2362
Chemistry       5197            2529            2331
Skeptics        3116            945             871
Emacs           2787            1586            1450
  • Adding an answer may not mean a lot. For example superuser.com/questions/618252/… got a new answer today. And it's not because the accepted answer wasn't testable or verifiable. – Kate Gregory Nov 5 '15 at 22:15
  • 1
    @KateGregory other times, new answers do add value. I'm not even trying to measure that yet, though, beyond the "upvoted" part of the query; I'm starting by asking about user behavior. That person presumably thought he was adding useful information even though the OP no longer needed it. (Also, will that answer still be there in the next SEDE refresh?) – Monica Cellio Nov 5 '15 at 22:25
  • 4
    Another note... some answers are "popular" but don't really address the question very well. These answers often get tons of upvotes but in some occasions, the accepted answer is actually better than the highest-voted answer. As an example: ell.stackexchange.com/questions/65647/is-whom-a-deprecated-word/… (I'm certain there are better examples but this was one that I'm familiar with) – Catija Nov 6 '15 at 4:37
  • In other cases when a new answer comes along the OP changes his accepted answer... – DavidPostill Nov 6 '15 at 13:52
  • @Catija sure, I've seen that a fair bit on The Workplace. But I've also seen a fair bit of the OP accepting the answer that most matches his own opinion (which is, often, "I was wronged!"), rather than the one that tells him how to fix his problem. So it goes both ways. – Monica Cellio Nov 6 '15 at 14:04
  • Perhaps we should allow for multiple accepted answers, pinned in order of votes. – Double AA Nov 6 '15 at 14:18
12

There's a number of things going on here.

The accept mark has always been an extra bit of signal from the question author. In most applications, it generally means this worked best for me, or led me to discover something else that worked. It's designed to be a beacon for future visitors, because the question stops belonging solely to the question author the second that they submit it. That's a key thing.

While the OP is free to accept an answer, which does sort of tacitly imply a sense of finality - there's positively no reason that others should hold off on posting additional answers that cover additional ground, or cover things that have been previously explained in a much better way. The question author might be done, but we try to optimize for everyone else that has the same or similar problem. If you have better information than other answers provide, post it.

What do you do with it? Well, that's up to you. It's a beacon, kind of like a hobo sign - it means someone once appreciated something and left a mark. It's up to you to check out the other answers, and pay particular attention to more recent answers, if they exist, that might contain newer information more relevant to your scenario.

The fact that it influences the sort order, and that old answers don't always age particularly well for a number of reasons (which is what we were looking at) is another matter entirely; the accept mark just further complicates that because losing that tiny bit of extra signal would be a bad idea. Even really smelly accepted answers are often educational, in a "do as I didn't" sort of sense.

We could come up with a number of schemes where the sort order changes based on signal that we get from votes, but that would mean piling even more debt on a heap of complexity we're trying to assuage in the anonymous / new user experience. We'd end up making things a little better for people that have a good understanding of how things work, but even more confusing to those that don't.

Anyway, I digress - the signal itself hasn't changed much over time, it's just the fact that it doesn't age particularly well that's been coming to light a bit more often as of late. Ask Bluefeet what "complicated" means and she might pull out a sharpie and draw a check mark on you :)

  • 4
    It's not as if it's not already slightly complicated... I've been here for 10 months, actively, and it took me about 8 months to realize that self answers that are accepted do not get the top spot by default. – Catija Nov 6 '15 at 7:18
  • One way to keep it simple w.r.t. aging: green check mark says its asker thought it was useful recently, and holds it at the top to encourage others to look at it first; amber check mark says asker thought it was useful a while back, and lets it take its chances in the order with all answers to date. Full proposal at meta.stackexchange.com/questions/268666/… – PolyGeo Nov 6 '15 at 7:32
  • 2
    @PolyGeo But that would only be useful for some sites... Yes, with computer systems updates, many answers may become out of date eventually,... but an increasing number of the SE sites are centered around answers that are factual and simply don't go out of date at all. A system like that would be wise to only implement on sites that want it. – Catija Nov 6 '15 at 7:37
  • @Catija My views on this are formed mostly from Genealogy & Family History and Geographic Information Systems which I think represents one of each in the dichotomy that you see. Aging may be an old answer going out of date or it may be a new answer being better due to any number of reasons - sometimes just the same facts written more clearly. Besides, if it is not useful on some sites just set the time period to 0 or infinity to achieve accept being nothing special or super special. – PolyGeo Nov 6 '15 at 7:42
  • 3
    @PolyGeo Yeah... but what if there aren't newer/better answers? What if the accepted answer is one of two answers and has all of the votes? Why age out the accept check in cases where it's unnecessary? It seems rather arbitrary to, essentially, un-accept every question that's over 12 months old... or whatever date range you recommend. – Catija Nov 6 '15 at 7:47
  • 1
    @Catija It's not aged out, it just has to earn its top spot after a period - it remains accepted, If there aren't newer/better answers then there is absolutely no difference in outcome, answer still has a check mark next top it, answerer still has 15 points and answer would still be at top but on its merit. – PolyGeo Nov 6 '15 at 7:50
  • @PolyGeo I don't really see how it fixes anything... the only questions that are helped by this are questions that have almost no upvotes on the accepted answer. I have questions that I've answered that have outright wrong answers with lots of votes... but those (accepted) answers have 10+ votes and my new, correct, answer can't compete, so the bad answer will always keep the top spot unless you somehow take into account recent (up and down) voting activity and not simply the raw vote count. – Catija Nov 6 '15 at 7:57
  • 3
    @Catija I wonder if those outright wrong answers at the top with 10+ votes were beneficiaries of spending years in top slot anointed with a green check mark attracting donkey votes. My proposal won't be an instant fix to all possible situations but its implementation should prevent some situations like that you just described from occurring. – PolyGeo Nov 6 '15 at 8:04
  • 2
    Tim, I've been starting to come around to the idea that the accepted answer should never be pinned (but there should be an easy way to jump to it). I agree that even more byzantine systems with moving answers over time, changing the check-mark color, having score-based or author-based rules for sorting the accepted answer, removing the check-mark, allowing the community to change the check-mark (particularly bad), and other such complexities make things worse, not better. And thanks for the warning; I'll beware of Bluefeet bearing Sharpie. :-) – Monica Cellio Nov 6 '15 at 14:10
  • @MonicaCellio We worked on something like that. The sort order was natural (by votes, by default) and if there was an accepted answer, there would be a tab so you could sort by it. That sounded really good to me when we first hashed it out on a call and then the work that would need to go into what 'accepted' means, in the absence of the intuition someone could draw from it being above everything else .. well that made the new user curve substantially higher if we didn't seriously clutter the design of the sort tabs. It's truly a very hard problem, we're still looking at it. – Tim Post Nov 6 '15 at 17:50
  • 1
    @TimPost thanks for letting me know what work you've already done on this. Elsewhere on meta I proposed adding another tab to the answer sorts, with "votes" being strictly vote order (& the default) and "accepted" showing accepted first (regardless of score/author) and then the rest in vote order. But instead of another tab, which it sounds like you tried, what happens if you just sort by votes and mark the accepted answer (wherever it falls) and probably add a link from the bottom of the question to jump to the accepted one? Does that confuse new users? Curious about failure mode if so. – Monica Cellio Nov 6 '15 at 18:10
19

To start with, in my experience this reason given in one of the posts referenced is just wrong:

The asker has decided it works

In my opinion and going on my own experience (mostly on SO), OP doesn't always test the solution provided. And if they did, they often pick the first, highest-voted or shortest answer (usually all three of them). Often that solution just gives a quick fix with not much more information. And that is what OP wants: just a solution, not the solution he wanted to have a few years later on (most users don't care).

That means that the better, more explaining and eventually late answer doesn't get any attention of OP, but it does from the community who recognizes the better answer from their experience. To me, the accepted answer just means 15 reputation, nothing more. It is often just meaningless.

What is should mean? I am not sure. Accepting seems the wrong word if you ask me. It was the answer OP likes best at the time he was interested in it. Most of them don't bother later when other answers come in.

In the ideal world, OP takes ownership of the question and maintains it, but that is far from the current practice.

If I am looking for an answer, I usually trust the community better than OP, so I start with the highest-voted answer, not the accepted one.

4

When accepting an answer, what I am saying is not always "I did that and it worked for me; I don't need any more help".

Oftentimes, it is:

I accept that this answer has done as much to help me as I think is fair to expect, given the question that I asked, so I think you deserve to be rewarded

I suspect this applies to many answers that I see accepted by others, and is irrespective of whether the question asked is good or bad.

I think that, in practice, accepting is used to reward not just the result, but also the effort to assist.

The instant 15 points is part of that reward, but what can gain many more reputation points for a good answer, is occupying that top slot for a period of time, and that is currently set to forever. The top answer is sometimes the only one I have time to read, and if it sounds reasonable then I'll upvote it.

You must log in to answer this question.

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