23

I notice the traffic of TEX.SX is heavy, for example, from 4:00 PM to 12:00 AM UTC. So, is it a good idea to wait for that time to ask a question to get enough attention/support from plenty of visitors? or the difference is not that much?

Update

I think this graph gives a better idea that the time factor is quite important, which makes sense because high percentage of LaTeX experts are living in this timezone, and, intuitively, they have more free time to answer questions after their workdays in the evening.

P.S. I know the question quality is important to draw attention/support, but here, I am asking about the weight of the timing factor.

10

I used this query to determine the amount of asked and unanswered questions grouped by the hour of asking the question.

This is what the data looks like when plotted with gnuplot:

Plot of question asked/answered by time of day

There percentage of answered questions fluctuates between 57.78% and 62.88%. From the graph it looks like questions asked between 5 and 7 are less likely to be answered and i'm sure one can find a correlation that explains this(More people asking than people answering questions online?), but i doubt that there is a statistical significance.

Here is another plot of the data, with the amount of questions asked in an hour on the x-axis and the percent of questions answered in an hour on the y axis for the 24 hours of the day.

Another plot of answer-percentage against question asked in the same hour

I also ran a query to determine the average wait time, but i do not see any connection to the time of asking. (This query only measures the time between asking the question and the "accepted" answer being posted, so it also only includes questions with an accepted answer)

I think i also came up with a query to determine the average time it takes for a first answer to be posted (ignoring unanswered questions). But i'm unsure how to interpret it. It looks like you can get a first answer a lot quicker, if you ask it at 4, taking an average of more than 8 days. In comparison to asking at 7, when it takes an average of 12 days for a first answer to be posted.

I'm not an expert on statistics or the stackexchange-data-api so take this with a grain of salt. The data provided might be misleading or not suitable for answering the question at all. Feel free to point out any mistakes.


The script used for plotting:

set datafile separator ","
f = "QueryResults.csv"
set xlabel "Hour of asking"
set ylabel "Questions"
set y2label "Percent Answered"
set y2tics
set ytics nomirror
set xtics (0,2,4,6,8,10,12,14,16,18,20,22,24)
set format y2 '%2.0f%%'
plot f using 1:2 with linesp title "Question Asked", \
f using 1:3 with linesp title "Question Answered", \
f using 1:($4*100) with linesp axis x1y2 title "Percent Answered"

and

set ylabel "Percent Answered"
set xlabel "Questions Asked"
set format y '%2.0f%%'
set  xtics auto
unset y2label
plot f using 2:($4 * 100) with linesp notitle, '' using 2:($4 * 100):1  with labels point  pt 7 offset char 0.5,0.5 notitle
  • 1
    very nice. i assume that this maps to "universal time", so 5 is equivalent to midnight (0) on the u.s. east coast. so this activity distribution makes sense. if i were being really fussy, i'd hope to see the hours by 4 or 6 (avoiding the unsightly gap on the right), but since i don't know how to do that, methinks i do protest too much. – barbara beeton Aug 30 '16 at 17:17
  • They are pretty interesting graphs, but I am confused a little bit. Can you tell me what the x-axes represent in both graphs? Are they the time in some timezone? ( I asked this question before seeing the comment above :). It would be nice if it is confirmed that this time is UTC. – Diaa Aug 30 '16 at 17:20
  • 3
    Very nice! But I can't resist, I have to comment. right y axis is a very common data munging ;) Zooming is not allowed on percent unless there is a rigorous reason for it. You show it is flat and then check whether there is a pattern in the wiggle not the other way around. This is a major and common problem in data discipline. – percusse Aug 30 '16 at 17:26
  • 1
    Midnight apparently is the happy hour for asking questions because of its high answering percentage and the decay of questions posting rate. – Diaa Aug 30 '16 at 17:43
  • I added another plot that shows the questions-answered as X and the Percent answered as Y. Also fixed the weird tickmarks. I don't really want to show the full 0-100% range on the second y-axis, because then you cannot really see a change, but i agree: the change looks bigger than it is and i tried to point that out in my text. – MaPePeR Aug 30 '16 at 20:34
  • just to confirm ... no matter how many answers a question gets, it is counted only as "one answered question", so the number of answers doesn't equal the number of questions answered. it's probably a safe assumption that most answers are to "current" questions, but presumably some answers are to older unanswered questions; i don't think you mentioned average delay from the time a question is asked to the time the first answer is posted (over the population of answered questions only). – barbara beeton Aug 30 '16 at 20:45
  • Yes, no matter how many answers a question gets: it should be only counted as one. I now added a query to determine the average amount of time it takes for a first answer to be posted (even if its not accepted). And it looks like there might be a little correlation to the time of asking. – MaPePeR Aug 31 '16 at 7:59
  • I don't really want to show the full 0-100% range on the second y-axis, because then you cannot really see a change, Because there is none. If an answer can be answered, it will be ansered, sooner or later. How fast a question can be answered depends on the quality of the question and the experience of potential helpers. – Johannes_B Aug 31 '16 at 8:22
  • To sum it up: If your question is clear, you have to expect the clock to tick a bit, so the experts of every time zone can see it. One of them will be able to answer. – Johannes_B Aug 31 '16 at 8:23
  • I want to [awesomeness of eventual answer] vs. [time of day at which question is posted] ;-) – SamB Aug 31 '16 at 22:51
  • 1
    Another explaination for the minimal (non significant) less accepted answers for questions asked early in the morning might be the lesser quality of questions asked just after waking up. ;) Measuring "awesomeness of answer" is hard, but one could use the amount of upvotes as an indicator. But i will leave this as an exercise for someone else. :) – MaPePeR Sep 1 '16 at 6:37
  • tl,dr version please? – nilon Sep 2 '16 at 17:42
28

It might seem better to ask in the high time but actually it doesn't matter. Because:

  1. You ask in the high time, more people are online (mostly CET day time). More people can look at it. You have high chance of getting an answer.

  2. You ask in the other hours, but because there are not many people online for the same reason, there not many questions either. Hence, in morning of the same people of the previous bunch, your question is still on the list.

So in terms of getting an answer it doesn't matter. In terms of waiting time for an answer can move a few hours but that's still negligible.

In my opinion it won't matter. This is only valid for our TeX-SX network though. It might be different for other networks.

Also users tend to look at the unanswered lists and sweep through anyways so it will be seen as you can judge from the smallness of the number of Tumbleweed badge.

  • 2
    I would add: Make sure you ask a good, well formulated question and it won't matter when you post it. – yo' Aug 24 '16 at 17:44
  • 4
    @yo' As long as it isn't too hard ;). – cfr Aug 24 '16 at 18:45
  • 9
    I really would like to see a statistic-based answer to this question. Like a plot of "percent of questions answered" against "daytime of asking the question". Or "Time until first answer posted". Does the API allow for Statistics like that? – MaPePeR Aug 25 '16 at 7:57
  • @MaPePeR Me too. There is the Data Stackexchange stuff but has a strange SQL syntax – percusse Aug 25 '16 at 8:20
  • 1
    @MaPePeR there could easily be confounders in such an experiment because the kinds of questions that are asked at different times might be different. Similarly, the quality of answers that are provided at different times might be different. The first problem can be fixed by taking questions and randomly delaying their posting. The second problem is harder to fix. – Neil G Aug 30 '16 at 16:40
  • @NeilG the time-correlation between your comment and my answer is huge, though. You pretty much posted your comment the moment i published my results. – MaPePeR Aug 30 '16 at 16:46

You must log in to answer this question.

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