7
$\begingroup$

In the Stats.StackExchange community, there's one tag called 'Time-Series'.

About 16.2% of CV questions are related to Time-Series, so I would think that only a negligible part of these would not get answered.

To my surprise, about 42.7% have no answers. (I just used the numbers that show on the right side of the site)

Is this an idiosyncrasy of our community?

I wish I could help people on this tag, but I cannot, since I'm trying to learn the subject myself.

Is there a way the community can solve this?

| |
$\endgroup$
  • 3
    $\begingroup$ One thing you could do to help whenever you have some time - even on questions you can't quite do yourself - is help identify duplicates or near duplicates. Flag the ones that are essentially the same question as one previously asked and answered, especially if it's accepted or has a couple of upvotes. Link to any that you think may help but don't exactly answer that question. $\endgroup$ – Glen_b Apr 1 '16 at 18:21
  • 12
    $\begingroup$ Sure it's an idiosyncrasy: what other community would routinely field questions on time series in the first place? A quick scan of the bottom third of these questions (in terms of votes) shows a variety of complicated, highly technical, and frequently poorly posed questions. It seems to be a popular subject: many people have time series data and want to make forecasts but appear to have little or no statistical background. It's a difficult mix of conceptually accessible questions being asked out of lack of knowledge but requiring extensive knowledge to be answered well. $\endgroup$ – whuber Apr 1 '16 at 18:28
  • $\begingroup$ @Glen_b I'll try to do that. ;) $\endgroup$ – An old man in the sea. Apr 1 '16 at 19:20
  • $\begingroup$ @whuber You're right, but I wonder if the same could not be said of some other tags here. Also, even if the OP 'seems' not to have the necessary statistical foundations, I do not think this should hinder other users from posting an answer, since the questions are viewable by everyone in the community. And as such, they are a 'public good'. $\endgroup$ – An old man in the sea. Apr 1 '16 at 19:30
  • 5
    $\begingroup$ I did not mean to suggest that people ignorant of a statistical technique ought not to ask questions! I was only pointing out that a significant portion of the posts tagged with time-series tend to be exceptional in some ways that make it unusually difficult (and unrewarding) to answer them. The problems I noted create a tendency for questions to be vague, ambiguous, and incomplete. They require extensive interaction to get clarification. But unilaterally posting answers to such questions, before they have been clarified, is not a good idea. $\endgroup$ – whuber Apr 1 '16 at 21:07
  • 2
    $\begingroup$ "About 16.2% of CV questions are related to Time-Series, so I would think that only a negligible part of these would not get answered." - I don't quite grasp the logic here. Why would a tag that is applied to 16.2% of questions correlate with a higher (or lower) probability of getting an answer? Could you explain this a bit? (And: in the light of our overall low answer rate, is 42.7% significantly lower than for other tags?) $\endgroup$ – Stephan Kolassa Apr 4 '16 at 13:16
  • $\begingroup$ @StephanKolassa Usually this type of communities are tracked in some statistics, right? for example, in area51. If one tag, which composes a great part of the asked questions, is being neglected, then it can be used as evidence that there's a lot of room for improvement. $\endgroup$ – An old man in the sea. Apr 7 '16 at 14:51
  • $\begingroup$ @whuber could the following question be described as you've had? stats.stackexchange.com/questions/201062/… $\endgroup$ – An old man in the sea. Apr 7 '16 at 14:54
  • $\begingroup$ But (especially considering that we have an overall low answer rate, which we regularly discuss here on Meta), to intelligently discuss whether a tag is neglected, we would need to compare its answer rate to the answer rate of other frequent tags. Do you have any data on that? $\endgroup$ – Stephan Kolassa Apr 7 '16 at 14:57
  • $\begingroup$ @StephanKolassa Tag R has 35%, tag Regression has 37%. I don't have much time to dedicate to this discussion. If you would like to do a deeper study on this, you're welcome $\endgroup$ – An old man in the sea. Apr 7 '16 at 23:25
  • $\begingroup$ Perhaps an autoregressive model for each tag to forecast changes in answer rate over time? $\endgroup$ – Sideshow Bob Apr 8 '16 at 8:20
  • 4
    $\begingroup$ whuber listed the problems of time-series tag quite clearly. I would add that time series field requires a certain specialist knowledge, which does not generalise well in other fields of statistics. Thus there are less people who can answer time series questions. Your own question illustrates this problem quite succintly. It is a basic textbook question, but it misses important definitions, so it is hard to answer without knowing context of this question. $\endgroup$ – mpiktas Apr 8 '16 at 12:48
  • 1
    $\begingroup$ @mpiktas An excellent point. I regularly see some of our most prolific answerers on CV - extremely knowledgeable, highly qualified, professional statisticians - comment that they can't help very much because time series isn't their thing. I suspect, as a distinctive sub-community of expertise, it's similar to Bayesian and ML. A large proportion of the Bayesian questions we get, though, are standard textbook stuff, and get eaten for breakfast by the knowledgeable folk here. ... $\endgroup$ – Silverfish Apr 10 '16 at 11:21
  • $\begingroup$ ... If we had an à la mode influx of "I've just collected masses of data, which I'm not going to show you, for some poorly specified objective. Bayes is cool right now, so show me how to do Bayes on it?" type questions then I imagine the answer rate on Bayesian questions would drop dramatically. Come to think of it, I wonder whether our answer rate on Machine Learning shows signs of a similar problem. $\endgroup$ – Silverfish Apr 10 '16 at 11:22
  • 2
    $\begingroup$ I can only strongly agree with @whuber (fwiw). Given my own background, answering questions on cointegration, unit roots and the like should be my biggest relative strength, and yet I hardly ever do so as these "questions" are often phrased very poorly. Indeed, they often reflect a deeper lack of understanding of statistics so that properly answering them would take quite long. Also, an above-average part of such questions seems to be by drive-by users, which additionally reduces the incentive to work on an answer. $\endgroup$ – Christoph Hanck Apr 14 '16 at 6:04

You must log in to answer this question.

Browse other questions tagged .