What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 130 Stack Exchange communities.

I understand the idea behind showing a users "accept rate" when they post a question, which is to encourage people to accept answers to their questions but it seems like it can/will encourage unwanted behavior as well. This unwanted behavior would be:

  1. Accepting an answer that isn't correct just to increase their accept rate score.
  2. Not answering questions asked by those with low accept rate scores.

In an ideal world, these behaviors wouldn't exist but there is no such thing as an ideal world.

share|improve this question
2  
"2. Not answering questions asked by those with low accept rate scores." Is this undesirable? I thought that was the point of having the "accept rate" scores: to encourage users who give recognition to the people who help them. –  Justin Morgan Feb 3 '11 at 23:10

12 Answers 12

The accept rate has a lot of problems I can think up right away:

  • It interacts badly with special cases (no answers, all negative-voted answers, self answers).
  • The value it displays is too precise. As if a person accepting 82% of answers in the past means there is an exactly 82% chance they will accept an answer for this question.
  • It isn't reliable (users can game the metric). If you want a high accept rate, accept a random response to all your old unsuitably-answered questions.
  • It doesn't match users' expectations. An accept rate of 70% or higher is great, but is that is what a user perceives 70% to mean? On a test, 70% is a pretty mediocre score.

A better metric would be for Stack Overflow to estimate what a 'reasonable' user should do for each question. Questions with a lot of answers or a high voted answer should definitely have one of them accepted. Questions with no activity are less likely to have an accepted answer. Then you combine this information to classify the user into much fuzzier boxes like "over-accepter", "normal", "under-accepter".

share|improve this answer
4  
I really like your thinking, I would say if a question has no upvoted answers there is a reason the poster hasn't accepted an answer, so perhaps those questions should be left off the calculation –  Breadtruck Aug 24 '09 at 2:45
2  
And just because say 4 people upvoted a answer doesn't make it the right answer though. Take this question meta.stackexchange.com/questions/16032 , I will not accept this because I don't agree with the answer. In the end I think this new feature will help more than hurt the process. –  Breadtruck Aug 24 '09 at 2:56
2  
Any metric for estimating the quality of a user is going to ignore edge cases, be flawed, be easy to manipulate, and be subjective. The only way to find out if a user is a "good" or a "bad" user is to browse through their questions and answers. The Accepted answer rate is an estimation designed to make it easier to pick out potential problems. Someone who has a 0% accept rate will get looked at, and when we see that they've asked 80 questions and gotten good answers on them, we'll know that this is a problem. Very rarely is a bad acceptance rate (like 0% or 10%) misplaced. –  Super Long Names are Hilarious Nov 14 '09 at 0:35
    
agreed. it's probably human nature for many of us to see that we have a score of 79% and want to make it 80%. (or maybe it was just me). –  Steve Bennett Dec 12 '10 at 1:28

Point #2 is sort of a feature, and one of the reasons we implemented this. Answerers have a right to know if the asker has a sub-10% accept rate before participating in the question.

Now, let's put on our "game theory" hats. How could one improve his or her accept rate percentage, by any means necessary?

  • accept an answer, obviously -- a net positive
  • delete the question (only works if no significant upvoting and the question has few answers, otherwise not possible) -- probably a net positive
  • convert question to wiki -- neutral, as this also prevents you from ever earning any more reputation on that question.
  • answer your own question, wait 2 days, and accept -- neutral to negative

As for point #1 -- if users accept an answer that is obviously wrong or incorrect, that makes them look bad, and will be reflected in the voting and comments. Since we operate on what is effectively a prestige system, I don't think this will be a serious issue.

Remember that any accepted answer rate of 70% or higher is great, and I'd consider anything at 50%+ plenty sufficient to indicate that the user is reasonably engaged. It's only a problem when the rate gets to 30% or lower, and for the real problem users, it's closer to 0%.

share|improve this answer
    
>and for the real problem users, it's closer to 0%. This will be any and all new users when asking their first few questions. Imaging someone just starting out and running in to problems. They post 4+ questions to try and understand different aspects of what they are doing (or doing wrong) and immediately have a 0% accept rate as soon as one of those questions gets a single answer. –  Scott Dorman Aug 23 '09 at 21:42
    
@Scott: They all have to have a single answer. Also, it will encourage them to accept the answers, since they will be able to see that 0%. If somebody has < 1000 rep and a 0% accepted answer rate, it's reasonable to assume that they're new, and you can encourage them in comments to accept answers. This helps bring people along to what is the encouraged behavior of questioners. How does that hurt at all? –  Eric Aug 23 '09 at 23:18
    
@Jeff - Maybe you can select a range of rates that correspond to "good", "marginal", and "bad" accept rates (say, >70%, >50%, 0-50%, whatever). Then color the accept rate text as green, yellow, or red; a visual indicator so people understand that they aren't necessarily striving to be 100%... just "green." It also emphasizes the "warning" nature of having a red percentage to both the asker and the answerer. –  Robert Cartaino Aug 24 '09 at 1:19
2  
already deployed last night my friend. already there. –  Jeff Atwood Aug 24 '09 at 2:45
1  
@Jeff My issue is that I have asked several questions where a few answers have been given, but haven't answered the question. The answers have no upvotes but SO still complains that I haven't accepted an answer and brings down my accept rate for not picking a incorrect answer. I don't know how this would be solved, but I have a problem with this incentive structure. –  CSharperWithJava Oct 27 '09 at 16:12
    
@Jeff, I don't think the green colouring helps - because I've never seen any colour that wasn't green, so it wasn't meaningful to me. Suggestion: >70: Display "Accept rate: good" >50: Display "Accept rate: ok" <50: Display "Accept rate: x%" (in red) with mouseover text explaining what this means. –  Steve Bennett Dec 12 '10 at 1:29
    
@steve just beacause you haven't seen it, doesn't mean we aren't doing it. I see red and orange accept rates on a regular basis. –  Jeff Atwood Dec 12 '10 at 2:35
    
I'm facing a problem in that I ask difficult but meaningful questions on Stack Overflow and nobody answers them. I'd be more than glad to accept a good answer. Why do unanswered questions count toward a user's accept rate? –  Daniel Mar 20 '11 at 15:09

Since the accept rate was added to the display, I've had several accepted answers on old questions. While I don't consider my own answers to generally be harmful, I'm still somewhat worried by it. Take this one for example. Useless tongue-in-cheek answer because I figured the question would be closed anyway, suddenly accepted yesterday.

It makes me wonder what else people are choosing to accept, just to reach 100% accept rate. I myself felt the urge to accept an answer on a question, even though the answer did not help solve my problem. I thought about deleting the question altogether for a moment as well :P

Sure, it is sometimes annoying if you take a lot of time to answer a question to the best of your ability to find that nobody else is taking the time to attempt to understand the question. Not a single point from upvotes, and the poster seems to completely forget about the question as well. Still, I'm beginning to think that I'd prefer having no points over this artificially boosted accept rate.

I'm wondering how the accept rate is calculated anyway. I have asked 7 questions on Stack Overflow. One has no answers at all, and one has only a single answer that wasn't very helpful. I've accepted an answer on the other 5. My accept rate is 80%. Neither 5/6 nor 5/7 is very close to 80%. I've accepted my own answer on 1 question, but that still doesn't explain the 80%?

share|improve this answer
    
+1 - I think this exactly illustrates my points/concerns. –  Scott Dorman Aug 24 '09 at 2:26
    
Keep in mind that the rates are bone cached and won't update until hours later. How it's calculated: blog.stackoverflow.com/2009/08/new-question-asker-features –  random Aug 24 '09 at 2:33
    
I totally agree, but at the same time I have looked at various questions where there were good answers and the question asker had not selected any. –  Breadtruck Aug 24 '09 at 2:34
    
Ah, right.. didn't think of community wiki. That brings it down another question => 4/5. –  Thorarin Aug 24 '09 at 3:06

Valid points here. It appears that this feature was, in part, implemented in response to my question here so I feel slightly responsible for it's presence.

It's not what I was asking for, and while I can see that some people think it's useful I have my doubts along the same lines as Scott.

My idea was to have it only on the users profile where it could be seen in relation to the number of answers provided as obviously 100% of 5 answers is somewhat different to 50% of 100 answers (or indeed questions).

Given one of the comments on my question I can see Scott's second point being particularly pertinent.

share|improve this answer
2  
Well, rest easy knowing that if anything goes wrong, we're all blaming you. –  Eric Aug 23 '09 at 21:12
    
@Eric - Cheers! –  ChrisF Aug 23 '09 at 21:13
    
I now understand why we have the Destroy option under the user tools :) –  Diago Aug 23 '09 at 21:17

Everything you do will be another avenue of unwanted behaviour.

The real question is, does the bad behaviour outweigh the good behaviour that it promotes?

share|improve this answer

Due to the fact that none of the answers are good enough I have so far deleted one question and answered 2 of my questions my self, so I could accept answers.

However I have also revisited some old questions and accepted a good answer.

But I have had to choose between two equally good answers on some questions just so I could accept one answer.

share|improve this answer

Point one will be eliminated by the votes in the long run. I do agree that certain people will be less inclined to answer a question by someone with a low acceptance rate, but it will also act as a motivator for people to accept answers which is a bigger issue at the moment.

There will always be pro's and con's with the changes intruduced into the Trilogy. It is the nature of communities.

If the system does get heavily abused, Jeff and the team will most likely put tools in place to allow the moderators to manage these abuses.

share|improve this answer

Ugh, I wondered what the huge influx of accepted answers today was. I will reserve judgement to see if it's useful or not, but in the short term, it is going to result in a huge influx of rep. to answerers.

share|improve this answer
    
Considering that this was an issue where answers weren't being accepted, isn't this a good thing? –  Eric Aug 24 '09 at 1:08
1  
It made me go back and review my unaccepted answers just to make sure that I hadn't missed any that should have been. –  Breadtruck Aug 24 '09 at 2:31

The highlighting is wrong for the accept rate. I've answered 4/5 questions, the one unanswered one where there is honestly no satisfactory answer given. It's actually a pretty good accept rate, but for some reason, the accept rate is not green. I think the question is of more use with no accepted answer (indicating to people browsing that the issue is still unresolved), but the system is basically prodding me to accept an answer for every last question that I ask.

Jeff mentioned before that a 70% accept rate was "excellent" but I think the actual cutoff point for highlighting is > 80%.

I think that highlighting should only be done when you are sure that the accept rate is harmful: say, under 20 or 30%.

share|improve this answer

I've found that having the acceptance-rate reminds me to go accept answers to questions that I was holding off on, until more answers had come in. Every now and then I notice my acceptance-rate, and then I go catch up.

share|improve this answer

I think that the really problematic edge cases are not questions but users. Many questions will tend to average out over all users and just influence the average accept rate, which will just change the definition of a "good accept rate", and not really causing a problem unless someone is adamant that it must. be. 70%..

But then you will also end up with people who tend to do lots of research before asking questions, and consequently end up asking almost, or perhaps entirely, exclusively tough questions that there may not even be answers to.

And then it gets trickier: People will post answers to these questions that are best-guesses, but wrong, and people will upvote them because they are the closest thing to an answer for the question. This means that even a proposed scheme to base the percentage of acceptable questions on the number of upvotes to the answers of these questions won't necessarily work.

So the situation is that there's an unclear definition of a 'correct answer': does this mean "the definitive answer to the question" or just "the closest thing that we have to a correct answer"? Until this is clearly defined and agreed on by both the poster and the voters (for all possible values of poster and voter BP) it's very difficult to quantitatively assess whether someone is responsibly accepting correct answers.

A different, and perhaps more effective, tactic would be to provide a way for people flag a given answer as the correct one, and to show the percentage of those that exist for a given user. This is not foolproof either: for example, people may misunderstand the user's original question.

It seems appropriate, though perhaps impractically long-winded, to allow up- and down-voting on {the flagging of an answer as inappropriately non-accepted}. The question at that point becomes one of determining what the relevant level of precision is.

[/braindump]

share|improve this answer

Seems like getting 2 rep just isn't such a powerful motivator such that folks are going to keep a steady 100% accept rate. Then again, when your rep is as low as mine...

Is there a way to flag/change accepted answers that are clearly (harder than I make it sound) not very "answery"?

In any case, this seems like the worst solution, except for all the others. Smiley face.

share|improve this answer

You must log in to answer this question.

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