12
$\begingroup$

Of course in context of R-versus-Clojure, I felt it needs external discussion.

So, I believe that there are two possible sorts of such questions:

  1. An algorithmic question like, "How to implement X?" or "My code is not working!"
  2. A question about tool selection and usage like, "I'm looking for X; what software should I use?" or "How to do X in Y?"

IMO the first case should definitely migrate to SO; the second should get its chance to exist here.
I can't accept the argument that if something is a programming language, then the question about it is automatically off-topic. So would we accept Minitab versus Statistica because those are GUI based, and not SAS versus R because one can do programming there?

$\endgroup$
22
$\begingroup$

Once you get out of your first couple of stats classes (and maybe even before then), you're often doing stats with software.

I would assume that a lot of people who will come here (especially professionals, not so much students or academics) are probably looking for an answer like, "how do I do such and such in SAS or R"? Especially if this is going to be "statistical analysis" and not "statistical theory".

I work in a place with a few "stats guys" and a lot of programmers, and questions like that always come to the statisticians (it's always stuff like "where are the p_values in this output" or "how do I transform this variable so I can use it in a logistic regression procedure")

I would expect that stats-software questions to outnumber pure stats questions as this site progresses (might be wrong).

And, I think this is a better site for that type of thing than stack overflow. Hopefully, we'll see good software answers, with enough of the stats thrown in to give the "asker" a little bit of foundation.

I agree that Colin's example should go here, and I hope that it's answered with a primary focus on the stats, and secondary focus on algorithmic efficiency.

To me, any SAS, R, S+, SPSS, Stata, etc. question is perfectly valid here.

$\endgroup$
10
$\begingroup$

Personally, I think we should be fairly flexible with (statistical) programming questions.

I agree that the first case is more SO, but when it's obviously statistical - as with this question - then I'm not too sure. I think that people may come to this site, but would not necessary venture to the SO site. For example, what tags would the example question get at SO? Who monitors those tags?

Edit following comment from mbq

I suppose I come from the more computational end of statistics, so I would see this as a statistical algorithm. What we need to consider is where the questioner is most likely to get the best answer?

For example, if you use MCMC you may need to keep a running estimate of the mean and variance, i.e. n isn't fixed. This is just an algorithm, but I suspect that people in this site may have more experience in implementing it.

$\endgroup$
  • $\begingroup$ I disagree about the question you mentioned; it is purely about optimization of an algorithm, so it would get tags: algorithm, optimization, maybe smoothing or curve-smoothing. In general on SO there are questions about totally weird and specific algorithms and they get their answers. $\endgroup$ – user88 Jul 20 '10 at 11:24
  • $\begingroup$ Ou, I see there is a problem about the definition of an algorithm. If algorithm=method (like How to adopt MCMC to problem X?), I agree that it is OK. If algorithm>>implementation (like I have written MCMC in bash and it is working for HOURS, how to speed it up?) this is for SO. $\endgroup$ – user88 Jul 20 '10 at 14:57
  • 3
    $\begingroup$ mbq, but the question Colin mentioned is not purely about optimization. That's just a metric to find a better algorithm. In fact, if you see see the answers, they're talking about how approximate solutions can be achieved on a limited sample of data seen at any time within a probabilistic bound. There are pointers to reservoir sampling, which is a neat application of probability. I think this is highly interesting from a statistical viewpoint. $\endgroup$ – ars Jul 21 '10 at 16:35
  • 2
    $\begingroup$ I just want to clarify something, since I both lurk on SO and participated in the median finding question. The kinds of algorithms questions entertained on SO are the ones where people are interested in specific programming answers. More conceptual (or theoretical) answers are not necessarily encouraged. WHereas here, more rigorous comp.stats questions/answers might be within scope - the only other home would be the theoryCS SE site (in commit stage) or MO. $\endgroup$ – Suresh Venkatasubramanian Jul 22 '10 at 20:22
4
$\begingroup$

It seems that the (almost universal) opinion in answers here was that the second kind of question should not be closed or migrated from CV ... but such questions are being routinely closed/migrated at the moment (many more than I am comfortable with). I'm voting against many of the closures, but usually to no avail.

Indeed it seems just recently like almost any mention of a package or a function in a package is treated as grounds for closure. I think it's high time for us to revisit this question, now.

$\endgroup$
  • $\begingroup$ Yes, there is a recent spike of software questions which are migrated to SO, but in my opinion they were better suited for SO, i.e. little to nothing of statistical content. The problem is however that if worded correctly they might have stayed on here. $\endgroup$ – mpiktas Mar 21 '14 at 7:41
  • 4
    $\begingroup$ Perhaps you could link to some examples $\endgroup$ – Scortchi - Reinstate Monica Mar 21 '14 at 13:13
  • 2
    $\begingroup$ There does seem to be an increase in the number of threads popping up in the close queue recently (not just off-topic). Many I agree w/, but there have been some that I thought were ambiguous & where the OP could have gotten the benefit of the doubt (at least for a little while) or at least gotten a comment to help them. If you want to review our close policies & discuss this, it may be better to open a new thread, rather than resurrect this one. $\endgroup$ – gung - Reinstate Monica Mar 21 '14 at 14:51
  • 2
    $\begingroup$ We can agree that this is an area of difficulty. But the thread at meta.stats.stackexchange.com/questions/1744/… (Sept 2013) is much more recent than this. Also, although that thread and some other discussions unsurprisingly showed some disagreement, the net result was absolutely no change in the policy as laid out in the help center. This implies a need for a statistical core to any question here, regardless of whether it is about statistical software. (Please) find my bug, review my code, or tell me the code are all unacceptable otherwise. $\endgroup$ – Nick Cox Mar 21 '14 at 18:14
  • 1
    $\begingroup$ @Scortchi Sorry, I just realized I hadn't come back and responded to this. Fair question. Here is a recent example of what I see as having an underlying statistical question with a software element that I think falls well inside what would be treated as on topic from the tenor of the answers here, yet it was migrated so quickly (in the space of about an hour from posting) that I didn't even get to see it (in spite of only being asleep a few hours). Such posts are pretty common lately. $\endgroup$ – Glen_b Mar 25 '14 at 22:38
  • $\begingroup$ @Glen_b: Unfortunately that's been deleted from SO. $\endgroup$ – Scortchi - Reinstate Monica Mar 27 '14 at 17:10
4
$\begingroup$

So if there were no Stack Overflow, then would the answer to the OP's question be different? My sense is yes.

And yet the fact that SO already exists shouldn't in any way affect our charter. I have been an SO member for over a year but if I were not, and I came the stats site, asked a question and had it closed / migrated to SO, I doubt I would follow it. Let's make this site a one-stop shop.

Finally, I agree with @Baltimark, why try and separate statistics from its principal technique? You can discuss descriptive stats without code, but what about topics like simulation or data visualization? Plus code is a very reliable way to distinguish good answers from bad ones.

$\endgroup$
3
$\begingroup$

I just think that questions need to be in reference to data analysis in some way or other. Asking "which is better: R or Clojure" with no mention of what kind of data analysis is not appropriate.

$\endgroup$
  • 1
    $\begingroup$ Yes, that's an example of a weak question. I would worry that if people come down too harshly on computer questions, the site will become, "what's a UMVU estimator and how do you find it"? I think a lot of stats questions are going to be tied to analysis and computer questions. $\endgroup$ – Baltimark Jul 20 '10 at 13:29
  • 1
    $\begingroup$ Absolutely...we (as the community) just need to educate these people on the right way to ask on-topic questions. Programming questions are on-topic so long as they're related to the site subject area. $\endgroup$ – Shane Jul 20 '10 at 13:55
  • $\begingroup$ I generally agree, but still I believe that the intended use was mentioned in the particular R-Clojure question. $\endgroup$ – user88 Jul 20 '10 at 15:03
  • 1
    $\begingroup$ Absolutely...I'm making a general comment. The particular example was fine (although I had recommended changing the title to be more explicit). $\endgroup$ – Shane Jul 20 '10 at 15:21

You must log in to answer this question.