What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 134 Stack Exchange communities.

Two years ago I asked an adorable question: is 180,000 unanswered questions too many?

Now there are 704,563 questions with no up-voted answers and counting. I've had the feeling that this flood was starting to overwhelm the site, but I went looking for some concrete data.

Percentage of open questions with answers, by quarter:

Answered Question Percentage

This is probably the most telling. The percentage with any answer is dropping steadily, while the percentage with a good answer (Score > 0) is in freefall. Voters don't seem very impressed with all the new questions, either:

Question Votes by Quarter

Question Votes

Question quality by quarter

Question Quality

This also suggests an increase in bad questions (score < 0), but just as importantly, 60% now have a score of 0. If the flood is too much even for simple actions like voting which almost everyone can do, what chance do editors and moderators have of keeping up? I'd also be interested in seeing the percentage of users with close-vote privileges over time, as well as the ratio of those users to new questions over time. I don't think SEDE has the required data though.

Eventually the site will simply cease to work if these trends continue. Community moderation will be affecting such a small percentage of questions that it might as well not be happening at all.

I can see three avenues of attacking the problem: increasing the number of (good) moderators, increasing the amount of moderating individuals can do, and building more flood gates. I'm hoping to trigger some brainstorming and draw attention to any other proposals that attempt to address parts of this problem.

Related Reading to help your brainstorming:

Assorted other things to ponder:

  • How can we reduce the feeling of futility when trying to moderate such a large and growing backlog?
  • If we were to hypothetically "raise the bar"/narrow the scope of the site, what would we exclude? How would we draw that line in a non-arbitrary way?
share|improve this question
Don't older questions and answers have a bias though? With time on the site, old questions could find answers, and posts could accumulate votes. Time is a factor, surely. –  Martijn Pieters Aug 11 '12 at 9:42
@MartijnPieters only considering answers that came within two weeks made surprisingly little difference: 0.5%-2% lower. –  Brad Mace Aug 11 '12 at 10:27
Thanks for verifying, it always pays to double-check assumptions. :-) –  Martijn Pieters Aug 11 '12 at 11:59
This analysis also needs to take the automatic deletion of zero-score unanswered questions after 365 days into account. Else this will make the old days look better than they actually were. –  Mad Scientist Aug 11 '12 at 12:28
@Mad that's a good point but I may have to leave that to someone else. tsql is not my home turf. If we compare 2011 Qs 2&3 (one year ago), that drop is larger than most, but we can still see the trend on both sides of the divide. –  Brad Mace Aug 11 '12 at 18:39
@YannisRizos: Clearly you spend way too much time on Meta reading adorable questions. –  Robert Harvey Aug 11 '12 at 19:12
The SO conceptual model & implementation of a Q&A site is flawed and not scaling well but the grey beards aren't interested and simply double down with their rigid views. That's fine - but it is not attracting new people who would be interested in actually building a knowledge base. A pity since there are so many talented people here with good intentions. Unfortunately their tasks (as @RobertHarvey noted in a "Summer of Love" blog comment) are primarily janitorial. Cheers + "thanks in advance" ;-) –  skinnyTOD Aug 12 '12 at 1:45
@skinnyTOD - "The SO conceptual model & implementation of a Q&A site is flawed and not scaling well" - OK, provide examples of other sites that have done better with the kind of posting traffic this site is now receiving. Every other forum, mailing list, or newsgroup I've ever participated in fell apart long before it reached the volume we're seeing now. I don't see the rigidity that you describe, as many of the policies and tools of the site have changed significantly since 2008. Simply look at what we accept for questions now vs. what we did then for one example. –  Brad Larson Aug 12 '12 at 3:11
@BradLarson - That's the wrong end of the stick: not a matter of "what other sites do it better." It is what SO is doing. And sorry, but the rigidity is rampant - see the earnest suggestions for improvements on meta with resulting downvotes (+ snark). An example of what is broken: I see your name on lots of edits (btw- I voted for you as moderator), doing trivial copy editing tasks when I know you could answer the question. Is SO a community generated knowledge base or more about grading student papers? Not enough room to say more but it saddens me to see all the pointless busywork. –  skinnyTOD Aug 12 '12 at 3:24
@skinnyTOD - My point was that we are in uncharted territory, but the fundamental Q&A approach and the mechanisms around it are why we've been able to get this far. I agree that we'll need to think of new ways to address the scale we're at, but I'm not as pessimistic about their reception by the community. The new /review system is one such attempt at dealing with the scale, and the community has had significant input in that. In regards to edits, I don't mind taking a little time to make things a little clearer. Sometimes that can have as great an impact as providing an answer. –  Brad Larson Aug 12 '12 at 3:42

6 Answers 6

Updated to include the last two quarters of 2012.

I think what's most interesting about the update is how the numbers have changed over time. For instance, when I originally answered this question quarter 1, 2012, was complete. There were 409,490 undeleted, open, questions that quarter of which 85.61% were answered and 70.71% answered "well". Those numbers today are 405,131 questions, 87.47% answered and 74.30% answered "well", which implies that as time goes by the Stack Overflow community is doing something about the older answered questions, just not very much. The increase in the number of questions with a score of 1 or more also indicates that SO users are upvoting (viewing/using?) older questions, which is good.

The extremely noticeable consistency between the current and previous results is the fact that all these downward trends have continued, things are getting worse. On current trends, quarter 3 2013 will have less than 50% of questions where an answer is either upvoted or accepted. If the trend in answering/deleting older questions continues then it'll be rescued later but the same will happen again, permanently, around quarter 1 2014.

This is a "problem" that needs to be fixed, somehow.

I know this isn't an answer but it's far too long for a comment. There's a problem with your SQL, caused by the LEFT OUTER JOIN; you're counting questions multiple times so the problem is both slightly better and slightly worse than what you think.

I also disagree with your definition of a "good" answer. By definition an accepted answer is "good" as it has helped the OP (unless of course they've been pushed into accepting it by loads of comments but that's another matter). I've excluded Community owned questions and answers as I don't really think they are relevant (and it helps the query to work!).

Your first set of results on the number of open questions by quarter now returns the following:

Year Quarter Questions Answered GoodAnswer 
---- ------- --------- -------- ---------- 
2008 3       17508     99.99%   99.83%     
2008 4       38790     99.87%   98.11%     
2009 1       53441     99.73%   96.23%     
2009 2       75339     99.56%   94.24%     
2009 3       98426     99.22%   92.49%     
2009 4       113136    99.02%   91.72%     
2010 1       142909    98.62%   90.76%     
2010 2       159213    98.01%   88.50%     
2010 3       187222    97.56%   87.32%     
2010 4       205903    97.41%   86.87%     
2011 1       267895    96.98%   85.83%     
2011 2       298140    96.16%   84.30%     
2011 3       312239    95.30%   82.52%     
2011 4       316593    94.46%   81.55%     
2012 1       405131    87.47%   74.30%     
2012 2       431420    85.41%   71.38%     
2012 3       452507    83.44%   68.07%     
2012 4       461118    80.62%   64.02% 

As you can see the percentage of answered questions is a little worse than you thought, but the percentage of questions with a "good" answer is a little better. Personally I think the telling point here is not necessarily the number of questions but where the differences lie between my results and your own. For Q4 2008 you have 122,616 and I've got 39,557 questions, which implies that every question received over 3 answers. For Q2 2012 the difference is minimal, on average questions receiving about 1.2 answers.

I've also run this for questions with a score >= 0:

2008 3       17454     99.99%   99.84%     
2008 4       38618     99.87%   98.13%     
2009 1       53133     99.73%   96.24%     
2009 2       74795     99.55%   94.26%     
2009 3       97569     99.22%   92.52%     
2009 4       111776    99.01%   91.77%     
2010 1       141491    98.60%   90.82%     
2010 2       157757    98.00%   88.54%     
2010 3       185404    97.54%   87.34%     
2010 4       203733    97.38%   86.91%     
2011 1       265103    96.95%   85.88%     
2011 2       293660    96.11%   84.35%     
2011 3       305355    95.20%   82.55%     
2011 4       308130    94.31%   81.54%     
2012 1       392376    87.07%   74.14%     
2012 2       417341    84.92%   71.18%     
2012 3       436875    82.85%   67.81%     
2012 4       443642    80.31%   64.00%  

I think the surprise here is how little difference it makes. It reflects well on Stack Overflow that no matter if the question is not as good as it could be you are just as likely to get a "good" answer. Obviously, closed questions would skew this massively and as a number of "poor" questions get closed not too much can be read into this.

Lastly, here the same query is for questions with a score >= 1:

Year Quarter Questions Answered GoodAnswer 
---- ------- --------- -------- ---------- 
2008 3       15797     99.99%   99.88%     
2008 4       32945     99.90%   98.87%     
2009 1       41381     99.74%   98.18%     
2009 2       53922     99.63%   97.38%     
2009 3       65645     99.53%   96.71%     
2009 4       69626     99.35%   95.97%     
2010 1       99493     99.02%   94.66%     
2010 2       104134    98.50%   93.04%     
2010 3       116514    98.10%   92.16%     
2010 4       122588    98.00%   91.82%     
2011 1       152765    97.79%   91.31%     
2011 2       169152    96.96%   89.90%     
2011 3       167683    95.92%   88.18%     
2011 4       165138    94.92%   86.80%     
2012 1       183482    93.92%   85.34%     
2012 2       180243    92.93%   83.66%     
2012 3       173618    91.26%   80.88%     
2012 4       188466    87.52%   74.33%   

As you can see the number of answered questions and the number of questions answered "well" significantly improves though the same drop-off is observable.

My own conclusion from these statistics is that a finesse to the system to remove unanswered questions, or whatever it might be, is not what is required. The number of answered "good" questions at over 91% is, in my opinion, a pretty high number.

What seems to be needed is an increase in the number of people who answer questions. Whilst Stack Overflow has had an ever increasing number of people asking questions there hasn't been a commensurate increase in the number of people answering them.

I ran a little query to test this hypothesis:

Year Quarter Questioning Answering 
---- ------- ----------- --------- 
2008 3       6411        9007      
2008 4       10724       13276     
2009 1       13713       15995     
2009 2       18929       21276     
2009 3       24474       25631     
2009 4       37388       36124     
2010 1       47454       41411     
2010 2       56895       46868     
2010 3       66340       52398     
2010 4       73162       59578     
2011 1       95347       74931     
2011 2       108828      78798     
2011 3       118150      85501     
2011 4       117156      89839     
2012 1       151337      107792    
2012 2       167394      116139    
2012 3       182767      125379    
2012 4       201461      133558 

As you can see in the "early" days the number of users answering questions was more than the number of people asking them. This has now been completely reversed and the questioners are in the ascendant.

What the solution is, I'm not entirely sure. What seems certain though is that Stack Overflow needs to find a way of converting question askers into question answerers. Without flooding the place with crap answers.

share|improve this answer
Nice work fixing my science. Your point about needing more answerers is a good one but we'll still need some way to help moderators keep up. –  Brad Mace Aug 11 '12 at 18:44
I strongly disagree with your statement that the measure of a good answer is acceptance by the OP. The only thing the check mark indicates is "fixed my problem", and sometimes, it doesn't even indicate that. Looking at other viewer's opinions as expressed through up/down-votes (while granting that they too can be misused) is a much better way to find quality answers. Indeed, that's one of the fundamentals of the site's functionality -- crowd vetting of answers. –  Josh Caswell Aug 11 '12 at 18:55
What seems to be needed is an increase in the number of people who answer questions. I'm not sure I can agree with that; I think the overwhelming factor is more important than the number of people. Too many duplicates are being asked. I think a better site search/suggestions could help the problem more than more people looking at an already daunting stream of questions –  Ben Brocka Aug 11 '12 at 19:02
@Josh That had occurred to me too. Possibly if an answer is accepted with score 0, this indicates that it's Too Localized and isn't useful to anyone else. It's a pretty hard thing to verify though. –  Brad Mace Aug 11 '12 at 19:59
The other problem is the criteria for a 'good' answer. I didn't see it in your answer, but an accepted answer with no upvotes can be just as good as an accepted answer with one upvote. Or even a question with answers and no upvotes at all. The fact that questions are getting answered at all is important. I'd be more interested in questions that have lingered with no answers for days. –  staticx Aug 14 '12 at 18:17
@0A0D, it's in the second paragraph... (and all the queries!) I count accepted answers but to be honest it doesn't make much difference. Look at my percentages compared to the OPs. –  ben is uǝq backwards Aug 14 '12 at 18:36
"which implies that as time goes by the Stack Overflow community is doing something about the older (un?)answered question" Actually, this is probably due to the increased inflow of new questions, together with the greater number of people answering them. This decreases the proportion of unanswered questions, without actually making a dent in the quantity. –  Asad Feb 10 '13 at 20:46
Why @asad? That sentence was discussing a static, past, period of time. There can't have been any new questions during that period, though they could have been answered by new answerers. –  ben is uǝq backwards Feb 10 '13 at 21:10
@benisuǝqbackwards Oh, I might have misunderstood then. Do 87.4 and 74.3 percentages apply to the same batch of questions? –  Asad Feb 10 '13 at 21:22
Undeleted? Did you mean nondeleted? –  Cole Johnson Jul 28 '13 at 17:16
I meant not-deleted yes @Cole... –  ben is uǝq backwards Jul 28 '13 at 17:19

Yes, this is a radical suggestion. Don't panic. We're just brainstorming.

A pressure relief valve for the backlog:

One possible option would be to expire more questions automatically, such as:

  • ignore the view count (as Pekka says, they don't get any better by just from being viewed)
  • questions with score <= 1 and no answers
  • questions with score < 1 and answers that are all score < 1.
  • Tie score required to remain on site to a question's age. For example, deleting unanswered questions whose age in months is greater than their score

This could significantly reduce the backlog of questions that need voting, answering, editing, and closing.

There's a risk of losing some wheat with all this chaff, but perhaps the extremely high traffic would make this acceptable and/or necessary. In this case if someone is still interested in a the question they could ask it again with no penalty. Or perhaps they just click something on the original question (still accessible from their profile) indicating that they still care about it. Knowing that question scores really matter might also encourage more voting.

share|improve this answer
Can I suggest that the questions with negative answers be deleted instead of zero score answers? It is possible some good answers never get any vote and I wouldn't want that to happen. But, a bad or a wrong answer has a decent chance of accruing a downvote, making your suggestion a little bit on the sensible side. –  jokerdino Aug 11 '12 at 10:19
Editing the question could serve to indicate they still care about it. –  Bill the Lizard Aug 11 '12 at 12:18
As long as it doesn't affect stackoverflow.com/badges/95/reversal or TumbleWeed. –  Jeremy Thompson Aug 12 '12 at 2:00
@jokerdino Some bad questions have good answers and I read recently on Meta that these should not be deleted. –  Remou Aug 12 '12 at 20:19
In the more obscure areas, single vote, or even zero vote questions and answers need not be bad, just not much frequented. Some may even be valuable as the only answer to some obscure point. –  Remou Aug 12 '12 at 20:19
This would affect the Tumbleweed badge. If you change it from a score of <= 1 to -1 and remove the time span month clause, I'll recind my downvote... –  Cole Johnson Jul 28 '13 at 17:20

Increase the vote limit

Let users do more voting, possibly tied to reputation.

I don't regularly run out of votes, but that's largely because I know they're limited, so I save them for the really good and the really bad. This leaves a lot of stuff in the middle with no indication of quality.

Currently 46% of questions and 37% of answers have a score of zero, ignoring closed questions.

share|improve this answer
I used to run out of votes (last year), but don't do that anymore. Could it be that the amount of not-bad-but-not-very-interesting questions has increased? That would explain the high number of zero votes. –  Bo Persson Aug 11 '12 at 9:58
@Bo I'm pretty sure that is exactly what's happening. The really bad stuff gets filtered out; it's the "meh" mediocre stuff that feels like the majority nowadays –  Pëkka Aug 12 '12 at 20:13
30-40 votes a day is a lot on typical SE site, but it's very little on StackOVerflow, with so much questions asked each day. –  РСТȢѸФХѾЦЧШЩЪЫЬѢѤЮѦѪѨѬѠѺѮѰѲѴ Jul 28 '13 at 16:56
@ŁukaszLech agreed. It should be increased. –  Cole Johnson Jul 28 '13 at 17:17

Prior review

Require posts from new users to be moderated before they show up on the site, until they reach 50 rep or so. Do the same for users whose last post was closed or deleted.

This would help the people who are answering questions as they won't have to wade through so much junk to find the decent questions. Plus they won't be stinking up the website at large.

The incentives for askers would also be improved, as they'll have more reason to fix their questions if they can't just shoot out a garbage one and get and answer anyway.

share|improve this answer
Who would pre-moderate those posts? The same people who are going to moderate them as soon as they're posted anyway. What would that accomplish, other than hiding the question from other low-rep users for a bit? –  Kevin Aug 12 '12 at 2:37
@Kevin edited... –  Brad Mace Aug 12 '12 at 2:39
But the ones who answer the most questions have the most rep and would therefore be the ones doing the vetting in the first place. –  Kevin Aug 12 '12 at 2:43
@Kevin, yes but only two would have to look at it, versus dozens that will typically see it under the current system. –  Brad Mace Aug 12 '12 at 2:45


Create sub-portals for popular languages and platforms, similar to facebook.stackoverflow.com. This would be akin to a View in a database where you would see only a subset of the site, whether browsing, searching, reviewing, handling flags, etc. For example, a Java portal could include not just , but also , ,, etc.

Allowing people to focus on a specific area of interest divides the problem into more manageable chunks, and cleaning up their particular slice feels like a more achievable goal, just as Facebook.SO's 16000 unanswered questions seems a lot less daunting than the 700,000 on the full site.

Likewise for reviewing, seeing only a slice of the 40.5k low quality posts and 58.7k close votes would make it seem a lot less futile.

share|improve this answer
We already have sub-portals. They're called Tag Wikis. For a good example of a properly used Tag Wiki, see c++ or php. –  Robert Harvey Aug 12 '12 at 19:36
What I'm suggesting here isn't anything like tag wikis. I've edited to try to make it clearer. –  Brad Mace Aug 12 '12 at 19:44
Because this idea worked so well with facebook. –  mikeTheLiar Oct 24 '13 at 17:58
Brad: it's totally unclear how this is any different from browsing a specific tag or set of tags or tag wiki. All it seems to say to me is that one language portal/master-tag/whatever would correspond to a set of tags, and those would rigidly silo discussion of a concept across languages. But we already allow user to define their own favorite tag lists, and without the partitioning, so this seems to have downsides and none of the upsides? –  smci Apr 11 at 9:35

Implement an invite system

Only allow users to join if they're invited by an existing user, similar to how Gmail worked early on. We could require a user to have, say, 100 rep before they can invite others. This would slow the tide and give us more time to get each new user up to speed on the site's rules and standards.

Current users would hopefully explain a bit about how the site works when they invite someone. Or if they're not following the rules, they won't get enough rep to be able to invite their friends who probably wouldn't follow them either.

share|improve this answer
Isn't this how Careers works? –  Cole Johnson Jul 28 '13 at 17:20
Also, don't be invite only, but they have to ask to join. That would be better. –  Cole Johnson Jul 28 '13 at 17:20

You must log in to answer this question.

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