What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 122 Stack Exchange communities.

Once the Stack Overflow data dump became available, I put together a data analysis script and created some graphs of SO activity and history. I thought this new meta site would be an appropriate place to make people aware of this.

Some of the graphs might be a little obscure at the moment. Mostly they reflect my curiosity and they're a bit experimental in the sense that I would create a graph and then see whether there was any interesting trend.

Here's one of the interesting graphs:

If you have any ideas for more graphs, suggest them here!

share|improve this question
    
There's a similar question on SO: stackoverflow.com/questions/951056/… –  dbr Jun 28 '09 at 12:02
    
@dbr: I think we should probably move that to meta.so –  Kyle Cronin Jun 29 '09 at 0:27
    
What is on the x axis? –  Sergey Jul 12 '09 at 9:27
1  
The x axis is time (current time on the right). –  Greg Hewgill Jul 12 '09 at 11:19
1  
could you put right axis on the graph's it's much easier to read the current value with them: is.gd/2Cmgn –  Sam Hasler Aug 27 '09 at 18:47
1  
I came looking for this. My impression is that there are now a lot more questions than answers. I have been taking pretty unscientific snapshots of the questions page on SO for a while and yesterday was the first time that none of the questions on the first page had any answers. The questions are becoming increasingly obscure too - perhaps that's why the answers/question ratio is falling, there are fewer people who could possibly know the answer. Or maybe the experts have got bored and moved on and SO is all newb on newb action. In any case I am seeing far less that I can contribute to. –  Simon Apr 22 '10 at 11:29
add comment

8 Answers

The two final stats are interesting: link text alt text

It seems that most don't get a better answer than the first one. That either means that they don't bother to answer again, or that their latter ones are worse than the first one.

Could we have a "Time to second answer" graph?

share|improve this answer
4  
I don't think it shows "that most don't get a better answer than their first one". I would interpret this as showing that the fastest-gun-in-the-west problem makes the first one get chosen as the best one. –  Paul Biggar Sep 26 '09 at 13:55
    
1) People prefer a quick answer. 2) After an acceptable answer is submitted, the questioner has little reason to revisit it. 3) Fewer people will want to answer a question that already has a reasonable answer. 4) The first answer has more time to get votes and few readers will read past the top answer unless it is horrendous. –  user253579 Feb 26 at 3:53
add comment

I'd like to see some bar charts showing various stats for some of the main tags.

i.e. something like:

    C#     | ========
    Java   | =====
    SQL    | ===
    Python | ====

Stats you could display:

  1. Number of questions.
  2. Average number of answers per question.
  3. Average number of votes per question.
  4. Average number of votes per answer.
  5. Average number of total question votes (i.e. add up the votes for the question and the votes for all the answers to that question).
  6. Percentage of questions with accepted answers.
share|improve this answer
add comment

Tag popularity over time would be a nice thing to visualize

share|improve this answer
    
specific tag? There's TONS of tags :( –  Ian Elliott Jul 3 '09 at 3:49
    
The primary Language tags (Java, C#, PHP, ...) –  C. Ross Jul 23 '09 at 17:43
add comment

The label "Answers per question by day" suggests that average values are computed/used.

Perhaps it would be more appropriate to use the median value instead of the average value (as the distribution of answers per question is likely a power-law or exponential and not symmetric around an average value).

Or you could do both and the difference would say something about the distribution.

Using the median could be part of the experimental process.

share|improve this answer
add comment

It would appear the way you're processing data is aggregated, ie, successive plot points are based on prior plot points so you're not seeing it windowed properly.

I'd like to see it graphed in terms of when the question was asked, not the global average, if its possible.

share|improve this answer
    
Not sure I understand what you mean. Each day should be grouped and counted independently. Can you suggest which part of the query to change? –  Greg Hewgill Jun 28 '09 at 11:13
    
If you share the code to how the graph works it could make understanding easier. –  Kent Fredric Jun 28 '09 at 11:16
    
The query used for each graph is shown on the linked page. The code to actually generate the graph isn't terribly interesting, it's just interfacing with the google chart api. –  Greg Hewgill Jun 28 '09 at 11:24
    
Gah, inivislbe link.meta.stackexchange.com/questions/77/change-the-link-color –  Kent Fredric Jun 28 '09 at 11:38
add comment

I'd like to see the percentage of 1st answers that are accepted, although it's not really a graph it's a number I'd be curious to know. Or perhaps something relating an answer's posted time to the original question's posted time and the number of votes they got.

share|improve this answer
    
Paolo - if you post that as a question, someone will be able to get you an answer. –  Brent Ozar Jun 28 '09 at 11:35
1  
While waiting for my query to finish running, I found the answer already: stackoverflow.com/questions/951056/… –  Greg Hewgill Jun 28 '09 at 11:46
    
Thanks, Greg! Interesting stuff. –  Paolo Bergantino Jun 28 '09 at 11:51
    
Greg's link has been migrated to meta.stackexchange.com/questions/7682/… –  Ether Nov 1 '09 at 0:42
add comment

I'd like to see a graph showing the number of views before getting the first answer. Perhaps something like this:

                    Views before first answer

  1-10  | ====
 11-20  | ======
 21-30  | =========
 31-40  | =====
 41-50  | ===
   >50  | =
share|improve this answer
2  
Not possible :( –  Ian Elliott Aug 19 '09 at 3:34
    
Really? Why not? I'm not familiar with the data dump. Is it because the number of views per question is not included in it? –  mfriedman Aug 19 '09 at 3:36
2  
View per question is included, but no indication of when those views were made; they're aggregated. Also answers have no view counts. –  Ian Elliott Aug 19 '09 at 3:57
add comment

The "Vote count histogram" graph for SO has a URL that is over twice the size that the Chart API supports: the URL is about 4909 bytes long at the moment, but the Chart API limits GETs to 2048 character URLs. (POSTs get 8 times the space.)

share|improve this answer
1  
Thanks, I've (finally) fixed this by limiting the histogram to 100 votes. There's no useful information in the very few questions that have more than that number of votes anyway. –  Greg Hewgill Jul 22 '11 at 8:28
add comment

You must log in to answer this question.

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