22
$\begingroup$

Has anybody got the information on the distribution of the time interval between posting of a question and the posting of the first answer to that question (answer need not be accepted)?

I believe the logs may contain the data to calculate the distribution. Could the MSE site make these available to us please?

This would help us to know the chances of getting any answer to a question and then reformat it or move it to another of SE sites such as MO or seek help in some other way.

Edit: There could be many other reasons to know this information for the sake of research if not anything else. Also more information can be given such as distributions based on the time of the day the question was posted as the time does matter since most experts seems to be from few areas of the world, or by tags etc.

$\endgroup$

migrated from math.stackexchange.com Dec 19 '13 at 19:28

This question came from our site for people studying math at any level and professionals in related fields.

  • 4
    $\begingroup$ I do not think the stated purpose makes any sense. The average time it takes for a question to get an answer will not help determine if the question would be better on another SE site, and will not mean much when deciding whether to give up on getting an answer (that would require the question to be "average"). $\endgroup$ – Tobias Kildetoft Dec 19 '13 at 19:46
  • $\begingroup$ @TobiasKildetoft The question is not asking for the average time. It is asking for the distribution. Distribution will help. Even the average and a few other higher moments can give some information. $\endgroup$ – triomphe Dec 19 '13 at 19:47
  • 1
    $\begingroup$ The distribution will help more, but will still not mean much. Anyway, that would only apply to the "give up" option, while the possibility of the question being more suited for another site still has nothing to do with this. $\endgroup$ – Tobias Kildetoft Dec 19 '13 at 19:50
  • $\begingroup$ Otherwise a question can be marked as duplicate. Some question overlap interests of multiple sites so the author can make a better decision of migrating the question if he has this information and thus expose it to another group of experts who do not frequent MSE. But that is only one purpose. It is in general interesting to know this distribution even without any particular use. $\endgroup$ – triomphe Dec 19 '13 at 19:56
  • $\begingroup$ This should be possible using the Data Explorer, I've done this for smaller sites in the past. My method doesn't work for MSE though as it runs into the limit of 50000 results for one query. But someone with better SQL skills could certainly create such a histogram purely in SQL and avoid this limit. $\endgroup$ – Mad Scientist Dec 19 '13 at 21:05
  • $\begingroup$ It is very important. I have experienced many times that the first answer is posted in less than 1 minute. It is suspicious and in these cases we could vote to delete the answer and/or the question. (Sorry. It is just my personal opinion.) $\endgroup$ – vesszabo Dec 27 '13 at 15:07
  • $\begingroup$ @vesszabo Many questions are just so straightforward that the Fastest Guns of Math.SE can indeed bang out a one-line answer right away. But if you suspect foul play, e.g., a user gaming a reputation system by posting answers to own questions from another account, the appropriate course of action is to raise a free-form flag on the user's answer and explain your suspicion to the moderator. Voting to delete content in cases of suspected fraud is not appropriate; it either punishes the innocent or removes the evidence, leaving the guilty party free to try again. $\endgroup$ – Post No Bulls Dec 28 '13 at 19:12
  • 1
    $\begingroup$ @vesszabo Also keep in mind that that posts can be edited within 5-minute grace period without the timestamp being changed. Thus, a long answer that appears to have been posted immediately could actually be a one-line hint that the author expanded within those five minutes. $\endgroup$ – Post No Bulls Dec 28 '13 at 19:14
  • $\begingroup$ @PostNoBills Thanks for your explanation. Everything is clear. $\endgroup$ – vesszabo Dec 28 '13 at 20:28
31
$\begingroup$

I considered the distribution of (question)-(1st answer) time interval among the questions that received at least one answer. There are about 186600 of such questions in the SEDE database, approximately 87% of all questions. (Thus, multiplying the percentages below by 0.87 will give the percentage among all existing questions.) Executive summary of results:

  • 30% of questions were answered in 10 minutes or less
  • 50% -- in 23 minutes
  • 69% -- in one hour
  • 75% -- in 1.5 hours
  • 90% -- in 15 hours
  • 95% -- in 3.5 days
  • 97% -- in 15 days
  • 99% -- in 6 months

Among answered questions, the median time to answer is 23 minutes, as stated above. Among all existing questions, it is about 32 minutes.

The longest it ever took for an answer to be posted is over 38 months; the current record holder is Covering for connected and locally path connected spaces (+1 CW points to dfeuer for posting that compilation-answer).

The distribution is heavy-tailed enough that I did not consider computing the mean, etc. The complete set of percentiles is given below.

I used the simple query Question-answer time interval running it four times to get around the 50000 line restriction. (At first run, enter 0 for minId; note the Id of the last question retrieved; download the results; repeat with the last question Id as parameter. Then merge data in your spreadsheet software.)

I don't feel like slicing the data by tags, time of day, day of week, month of year, phases of Moon, etc. The predictive power of this data is rather weak. The site is growing; its dynamics are different during school semesters and during holidays; some prolific answerers join, others quit or go on hiatus. Questions vary by difficulty and the level of specialization, which is something that tags do not capture. But if someone feels like slicing the data -- go ahead and fork the query.

A question does not have to get lots of views to get answered: it needs to be seen by 1 person with the sufficient expertise, interest in the question, and time available. There is no way to predict when or if such a person will see the question. Choosing proper tags and informative title, writing a clear question in easy-to-read format, setting a bounty if needed -- all of these contribute to success more than pondering statistical tea leaves.

%   time in minutes
min 0
1   2
2   2
3   2
4   3
5   3
6   3
7   4
8   4
9   4
10  4
11  5
12  5
13  5
14  5
15  6
16  6
17  6
18  6
19  7
20  7
21  7
22  8
23  8
24  8
25  9
26  9
27  9
28  10
29  10
30  10
31  11
32  11
33  12
34  12
35  13
36  13
37  13
38  14
39  15
40  15
41  16
42  16
43  17
44  18
45  18
46  19
47  20
48  21
49  22
50  23
51  24
52  25
53  26
54  27
55  28
56  30
57  31
58  32
59  34
60  36
61  38
62  40
63  42
64  45
65  47
66  50
67  53
68  56
69  60
70  64
71  69
72  74
73  80
74  86
75  93
76  102
77  112
78  124
79  137
80  152
81  171
82  194
83  223
84  260
85  306
86  367
87  445
88  549
89  683
90  860
91  1105
92  1422
93  1999
94  3057
95  4986
96  9378
97  21380
98  70302
99  252745
max 1698418
$\endgroup$
  • 2
    $\begingroup$ This is great work, never new there is a place to get this data. Thanks. $\endgroup$ – triomphe Dec 20 '13 at 2:00
  • 8
    $\begingroup$ I would've bet an ice cream cone you were the one who posted this quite a while before I hit the bottom and saw you. Pray tell: what makes you so good at and interested in this sort of data gathering and statistical work? $\endgroup$ – dfeuer Dec 20 '13 at 3:24
4
$\begingroup$

Just for fun, I followed @PostNoBills method and ran this query, calculating the time of question posted till the creation time of the accepted answer.

It took a median of 29 minutes. The record holder is this question.

Note: Some outliers such as this question have negative time interval due to rare cases of merging. There are 4 questions with negative interval while another 108 have 0 interval.

123,431 accepted answers out of 217,050 questions, acceptance rate 56.9%
The first draft of all accepted answers took a median of 29 minutes to compose.

%   Minutes
------------
1   2    (Woah, so fast!)
2   2
3   3
4   3
5   4
6   4
7   4
8   5
9   5
10  5
11  6
12  6
13  6
14  7
15  7
16  7
17  8
18  8
19  8
20  9
21  9
22  9
23  10
24  10
25  11
26  11
27  12
28  12
29  12
30  13
31  13
32  14
33  15
34  15
35  16
36  16
37  17
38  18
39  18
40  19
41  20
42  21
43  22
44  22
45  23
46  24
47  25
48  27
49  28
50  29   (Median)
51  30   (30 minute mark)
52  32
53  33
54  35
55  36
56  38
57  40
58  42
59  44
60  47
61  49
62  52
63  55
64  58
65  61   (1 hour mark)
66  65
67  69
68  73
69  78
70  83
71  89
72  95
73  102
74  111
75  121
76  131
77  143
78  158
79  175
80  194
81  218
82  246
83  279
84  321
85  368
86  434
87  511
88  614
89  737
90  896
91  1104
92  1366 (1 day mark)
93  1728
94  2560
95  3709
96  5775
97  10255 (1 week mark)
98  22601
99  92499
100 1624255
$\endgroup$
  • 2
    $\begingroup$ It's interesting that above the 90% mark, the values in your table are smaller than in mine. Which I guess can be explained: the answers posted long time after the question are less likely to be accepted, possibly because the OP moved on. Even though 1 day is not such an awfully long time... $\endgroup$ – Post No Bulls Dec 23 '13 at 17:21

You must log in to answer this question.

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