Hold on --- the article says that the top questions were asked by people with a lot of reputation. Isn't it the case that the top questions bring the asker a lot of reputation points, which means it's not necessarily the case that their question is a "top" question because of their preexisting reputation?
Well, there's really something else going on - I'm sure the phenomenon is there but I think they're asking the wrong question.
Stackoverflow tends to get a mix of questions that would require an experienced person to give a great deal of thought to, or things that require specialized knowledge ("good questions") and things that look very much like someones first year CS homework or questions coming out of pure laziness. In the worst cases the posted could have solved these questions by just checking public API docs but didn't bother. Often the worst questions are anonymous or accounts with single-digit reputation, display a fair amount of ignorance, and are syntactically mangled. The posters will generally not bother to take the trouble to upvote, respond, or thank for the replies.
I wonder if this phenomenon would disappear as a trend if they filtered out the "clown" questions (lets say, questions by people who posted less than 3 or 4 questions and no answers, or questions by anyone who had questions downvoted as homework).
This is a very interesting article, however, IMHO it has several major problems:
1. The author doesn't understand, or is too lazy to recognize, that "correlation does not imply causation". It's hard to read sentences like it seems like the length of the title has a minor influence on the quality of the question and not shudder.
2. There's no interpretation. At all. It's just queries, and doesn't provide any real insight into how to write better questions, get better answers, or get more upvotes. No hypothesis or model.
3. It's missing meaningful statistical measures, such as standard deviations, error bars, etc. For example, is this even statistically significant: The average title length of the top questions was about 5% shorter than that of standard titles (47 characters vs. 50) ?
4. Biased samples.
So my takeaway is that it was fun and interesting to read, but not very meaningful.
I'm kind of worried that they checked so many types of things (64) on small outlier samples (top 300):
> We ran eight different queries and sorted the questions according to: [8 criteria]. Then we compared the top 300 questions in each section to 300 questions that received an average score on the parameter we focused on.
Would results of that be statistically significant, if you picked the apparently-best 5 resulting from that process, or would they just be noise?
(Isn't ten thousand questions not even two days' worth of questions?)
7 comments
[ 3.7 ms ] story [ 34.4 ms ] threadStackoverflow tends to get a mix of questions that would require an experienced person to give a great deal of thought to, or things that require specialized knowledge ("good questions") and things that look very much like someones first year CS homework or questions coming out of pure laziness. In the worst cases the posted could have solved these questions by just checking public API docs but didn't bother. Often the worst questions are anonymous or accounts with single-digit reputation, display a fair amount of ignorance, and are syntactically mangled. The posters will generally not bother to take the trouble to upvote, respond, or thank for the replies.
I wonder if this phenomenon would disappear as a trend if they filtered out the "clown" questions (lets say, questions by people who posted less than 3 or 4 questions and no answers, or questions by anyone who had questions downvoted as homework).
1. The author doesn't understand, or is too lazy to recognize, that "correlation does not imply causation". It's hard to read sentences like it seems like the length of the title has a minor influence on the quality of the question and not shudder.
2. There's no interpretation. At all. It's just queries, and doesn't provide any real insight into how to write better questions, get better answers, or get more upvotes. No hypothesis or model.
3. It's missing meaningful statistical measures, such as standard deviations, error bars, etc. For example, is this even statistically significant: The average title length of the top questions was about 5% shorter than that of standard titles (47 characters vs. 50) ?
4. Biased samples.
So my takeaway is that it was fun and interesting to read, but not very meaningful.
> We ran eight different queries and sorted the questions according to: [8 criteria]. Then we compared the top 300 questions in each section to 300 questions that received an average score on the parameter we focused on.
Would results of that be statistically significant, if you picked the apparently-best 5 resulting from that process, or would they just be noise?
(Isn't ten thousand questions not even two days' worth of questions?)