11
$\begingroup$

I noticed that multiple questions that ask about R packages are voted to migrate to SO (I also do this). This is consistent with the fact that CV is not about programming. On the other hand, "I am looking for R package that does something" is not really a programming question and I would say that this kind of question is also off-topic on SO. So maybe the unwritten policy of migrating those to SO is not valid? What do you think?

$\endgroup$
  • $\begingroup$ I think it depends. There might also be some other SE sites, which better fit software-focused questions. For example, most recently I've recommended to migrate this question to Data Science (DS) SE site, as, while the question is clearly off-topic on Cross Validated, it fits DS reasonably well. $\endgroup$ – Aleksandr Blekh Feb 28 '15 at 8:32
4
$\begingroup$

Stack Overflow uses a fairly stringent definition of what constitutes a programming question (perhaps just as we have a fairly strict definition of what constitutes a statistical / machine learning question). A question that simply has to do with how to use, say, R is not necessarily a programming question. One good rule of thumb is that programming questions should have a reproducible example / MCVE. A question about programming that doesn't have one should not be migrated until it has been edited to include such. A question that cannot have one (e.g., 'what is the package for _______') should not be migrated, but should simply be closed here as off-topic ("because it is not about statistics, machine learning, data analysis, data mining, or data visualization").

I recognize that this may seem harsh to people who want help and are upset that we are not providing it. However, there is no guarantee that any given question must fit in somewhere on the SE system. Many general R questions can go to the r-help listserv—that's what it's for. In addition, we can point people to our meta.CV thread that lists Internet Support for Statistics Software.

$\endgroup$
  • 2
    $\begingroup$ Yes, but still one could argue that this kind of questions are related to statistics ("how to do statistics with" or "what can help me to do statistics") so the description of the off-topic mark does not fit perfectly to this situation. The real reason is that this kind of questions are about software and not that they are not about statistics. $\endgroup$ – Tim Feb 22 '15 at 22:47
  • 1
    $\begingroup$ That's a reasonable position, @Tim, but I disagree w/ it. Once you are no longer asking about statistics, but about something that may help you to do statistics, where do you draw the line? I use my keyboard & mouse to help me do statistics, after all. My position is questions about statistics / ML / DM / visualization are on-topic here; all others are off-topic. $\endgroup$ – gung - Reinstate Monica Feb 22 '15 at 22:56
  • 1
    $\begingroup$ This is perhaps the most difficult long-term question on Meta and even higher-reputation members rarely agree exactly where the line should be drawn (although I seem usually close to @gung in voting). I've found myself increasingly noting that many questions don't belong on SO any more than they belong on CV. I'm tempted to suggest that almost no community wants many learner questions about software (is there a package/function/command to do X? why do I get <basic error message>?) any more than we want to calculate means for people or check their arithmetic. $\endgroup$ – Nick Cox Feb 23 '15 at 0:11
  • 2
    $\begingroup$ The criterion in the help for determining the topicality of such questions is "requires statistical expertise to understand or answer". I think in at least some cases advice about the appropriate choice really does require exactly that, presumably making it on topic. There's a previous question about interpreting the phrase here (not that I'm claiming my comment is related to answers in that question, just pointing out its existence) $\endgroup$ – Glen_b -Reinstate Monica Feb 23 '15 at 0:22
  • $\begingroup$ @Glen_b, I strongly dislike that statement, which I think is ambiguous & leads to problems. My criterion has always been, what does the OP need explained?. Note that 'why is my ROCR code not giving me the ROC I wanted?' requires "statistical expertise"--no one w/o stat / ML expertise would know what a ROC is, but the Q is clearly a coding Q. I hold that Qs where the OP needs a statistical issue explained belong here, even if they are asking for code, and Qs where the OP needs code explained go to SO, even if no one w/o stat expertise would know what a ROC is. $\endgroup$ – gung - Reinstate Monica Feb 23 '15 at 4:43
  • 1
    $\begingroup$ I can imagine a Q where the OP asks for a package, but what they really need (although they wouldn't know it) is to have some statistical concepts clarified. However, I don't believe I have actually seen such a Q. $\endgroup$ – gung - Reinstate Monica Feb 23 '15 at 4:44
  • $\begingroup$ @NickCox, there are 2 issues: (a) the essential nature of the line, & (b) where it lies relative to any given Q. I believe that most users on CV are more or less in agreement on (a) & certainly that you & I are. (b) OTOH is often a difficult judgment call; I'm not sure that you & I actually disagree there so much as some situations are truly ambiguous. $\endgroup$ – gung - Reinstate Monica Feb 23 '15 at 4:57
  • 2
    $\begingroup$ What about the fact that R tag is by a large margin the most popular tag on Cross Validated (CV)? If questions of how to use R are off-topic at CV, then what could be done? I am not sure how good r-help listserv is, but if people are looking for R help at CV so often, maybe R deserves a separate site on Stack Exchange?.. $\endgroup$ – Richard Hardy Feb 24 '15 at 10:15
  • $\begingroup$ @RichardHardy, this is a contentious issue. There are many R Qs on CV--in fact I've asked 1 & answered ~200! So Qs w/ an R component are not necessarily off-topic. However, there are many fold more on SO (8k vs 82k), & much much more than that on the r-help listserv. I vote to migrate b/c I believe askers will get better & faster As there. The idea of an R specific SE has been raised several times; it didn't go anywhere (see: What happened to the idea to start an R-specific StackExchange site?). $\endgroup$ – gung - Reinstate Monica Feb 24 '15 at 14:28
6
$\begingroup$

I have seen other forums migrating such requests to Software Recommendations, though that forum tends to be a still defining itself and not known for prompt responses or well moderated. Right now as it covers a broad range of topics and not people who are say, experts in R or what it does in this case.

R packages based on this forum's core competency I believe would be appropriate here due to the original core purpose of R and its continued primary use even though it's use is definitely branching out.

Though in general, I would prefer any such requests that 'are migrated'go to Software Recommendations than the bigger black hole of Stack Overflow which is supposed to be programming questions.. not which software package should I use....

I know I have recommended R packages over in Software Recommendations on several questions as well.

$\endgroup$
2
$\begingroup$

I think that this would depend on the question being asked. If someone is looking for a data manipulation package like plyr for example, I would think that would be a SO question.

But if the question is more about how to implement a particular model in R, then surely this is straightforwardly a statistics question? R, as with any programming language, can be used for things other than statistics - I presume this is why general programming questions are referred away to SO. But if the question relates to doing statistics IN R, I would think that this fell within the purview of this site.

Whether there is a particular package that implements the model directly or it has to be implemented by hand probably shouldn't be used to discriminate which questions should go where. Equally, I don't think that model implementation questions should be removed from Cross Validated either, because that would reduce its usefulness substantially to practitioners.

The major factor to consider here is probably more around question quality and whether the person has made an effort to do their own research on the subject. If there is no evidence that they have made an effort, then the question probably contravenes other rules around question quality and can be closed on that basis.

$\endgroup$
  • 2
    $\begingroup$ You have, I think, a good point that lack of research effort is a key signal, but that is a criterion for down-voting, not for exclusion from the site. Otherwise a high fraction of learner questions with the flavour "I know nothing much about statistics, but" would not qualify here. Similarly, the distinction between "data manipulation" (not suitable here, you say) and "model implementation" (suitable here) is in my view a long and smooth gradation. I don't have easy demarcation rules either; in my experience of this site software-related questions raise more difficulties than any others. $\endgroup$ – Nick Cox Mar 5 '15 at 0:40

You must log in to answer this question.

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