12
$\begingroup$

If you see the comments to the question on MLE for Naive Bayes in R you will notice that there is confusion about whether R questions should go to Stackoverflow or Stats Analysis. I am of the opinion that for the sake of consistency we should (in cooperation with the Stackoverflow community) adopt one consistent policy. Towards this end, I propose the following policy:

  1. R questions should be asked on Stats Analysis.

  2. Any future R questions on SO should be migrated to Stats Analysis.

The above policy will achieve several objectives.

  1. Posters will not have to post twice.

  2. Cross-posting will not happen thus avoiding splintering the knowledge base for R across two sites.

  3. Less noise in comments both at Stackoverflow and at our site.

What do you think?

Once we reach consensus on this issue perhaps we can ask the same question on meta.SO to ask what they wish to do about this issue.

$\endgroup$
4
  • $\begingroup$ I appreciate the idea of trying to work toward a consensus on this policy, as it's one of our most substantial and contentious issues. I feel like the Stack format might not be well-suited for developing and identifying consensus, however - we don't have any way to tell who is undecided and who just doesn't care. Any thoughts? $\endgroup$
    – Matt Parker
    Commented Oct 28, 2010 at 18:57
  • $\begingroup$ @Matt I am inclined to wait some more time for others to chip in. I have some ideas as to how to approach consensus but I do not want to muddy the water too soon with any modifications to the existing question. $\endgroup$
    – svadali
    Commented Oct 28, 2010 at 19:03
  • $\begingroup$ That's perfectly sensible. $\endgroup$
    – Matt Parker
    Commented Oct 28, 2010 at 19:32
  • $\begingroup$ See this follow-up question to arrive at some consensus reg R questions: Vote on policy for R questions $\endgroup$
    – svadali
    Commented Oct 29, 2010 at 15:57

5 Answers 5

Reset to default
11
$\begingroup$

I strongly disagree with this approach. I think that permitting programming questions here dilutes the actual statistical/machine learning/data viz content of the site. I see three main problems:

  1. It creates entire categories of questions that entire categories of users will not ever care about. I like STATA, I've used STATA, but it is incredibly unlikely that I'll ever see it again - and I definitely don't care about how to get extra digits in my p-values with it. Conversely, I'm continually expanding my repertoire of statistical methods, so I'm always interested in reading the stats questions, even if they're not of immediate relevance - useful perspectives or approaches can come from almost anywhere, it seems.

  2. It makes it tedious for people who are interested in both programming and analysis to sort out which questions they feel like checking out that day. I'm rather partisan about R, but I still don't want to see questions about how to vectorize operation X every time I come here. And what about people who are mostly interested in programming for R and not statistics? They might exist! I think a very relevant example of this is the GIS StackExchange, which is extremely technical and is having an oddly parallel discussion on its meta about how not having so many software questions might be good for the site. Disclaimer: I complain about this same stuff over there, too, so the discussions aren't exactly independent.

  3. Permitting these questions here encourages duplication of past effort. Deciding to move all R questioning here means that we're orphaning > 2,300 questions that are already on StackOverflow (for reference, we've got 993 total questions here). In fact, there are several well-established sources for technical help with R and even specific packages. The same is true for the dozens of other statistical software programs out there. It will take us incredible effort to duplicate that knowledge and lure those users here; wouldn't it make more sense to focus our efforts on the need that we actually do address - Q&A for data analysis itself? This site is awesome because it provides new functionality to the Internet... not because it could potentially reproduce existing functionality.

I think a very easy rule of thumb for identifying where a question will go is to ask: can the question be restated as "How do I analyze X?" If so, it belongs here. If the question is better restated as "How do I program X?", then we (helpfully, knowledgeably, courteously) send it to SO or other appropriate resources.

There will always be questions that fall into the grey zone in between (e.g., algorithms for calculating statistics on streaming data or very large datasets), but just because we can't say definitively if a question belongs here or on SO doesn't mean that we should duplicate the entire functionality of SO.

$\endgroup$
5
  • 1
    $\begingroup$ +10 : I can't see how we can ever solve a question on statsexchange dealing with linking R through RODBC with Excel, or how to interact R and Python. These questions do get asked. OK, Sometimes the question is "how do i analyze X and how do I program it?" but then the answer if most often : read more books until you understand what you're talking about. $\endgroup$
    – Joris Meys
    Commented Oct 28, 2010 at 11:15
  • 1
    $\begingroup$ I recently pointed out on the GIS forum that probably over half of the questions were tagged with something to do with ESRI products. I thought here would not be the same, but I looked and easily the most common tag is R (although we have alot more mix of tags than does the GIS site). stats.stackexchange.com/tags $\endgroup$
    – Andy W
    Commented Oct 29, 2010 at 13:26
  • 4
    $\begingroup$ +1. But it is interesting that even that Stata question (about formatting p-values) led to a potentially useful discussion about why one would want to measure p-values to more than three decimal places. (stats.stackexchange.com/q/4027/919 ) In reviewing view counts it also seems to me that questions explicitly oriented towards one platform actively discourage readers who otherwise might benefit from them. Maybe what we should be doing is encouraging questioners to emphasize their analytical needs and to try to formulate questions without reference to particular software. $\endgroup$
    – whuber Mod
    Commented Oct 30, 2010 at 22:35
  • 1
    $\begingroup$ whuber, I like that idea very much. $\endgroup$
    – Matt Parker
    Commented Nov 1, 2010 at 17:44
  • 2
    $\begingroup$ Might I suggest the excellent 'rule of thumb' described above be applied to math as well as computing-related questions too? Just as some questions are purely about programming, some questions are straightforward math or probability problems with no stated relevance to actually analyzing any data or understanding any particular model or method. The check would be to ask: Is the code or derivation requested in this question essential to (or necessary for communicating about) some aspect of a data analysis, model or method? If the not, the question belongs someplace else. $\endgroup$
    – conjugateprior
    Commented Nov 2, 2010 at 19:42
6
$\begingroup$

I agree wholeheartedly with Matt Parker. Quite a few questions on R have hardly anything to do at all with statistics. There are two things I would like to see happening though:

It happens people ask the same question both here and on SO. That should be discouraged by actively merging those questions to either site, depending on the type of question (main focus on analysis or main focus on programming).

It happens people ask an obvious statistics question on SO. Mods should be able to migrate that one to this site.

I don't think that this can be done as long as stats is in beta, but once it is launched, this should definitely be added to the functionality. And when that is, I suggest strongly to apply the point of view Matt Parker explained.

$\endgroup$
1
  • 1
    $\begingroup$ A clear counterexample to "R questions should be asked on Stats Analysis" would be stackoverflow.com/questions/14208016/… - as Joris says, it really has nothing at all to do with statistics! If anything, it might belong on the GIS site, but at heart it was a programming question. The "main focus" rule would have worked very sensibly. $\endgroup$
    – Silverfish
    Commented Oct 31, 2013 at 2:58
6
$\begingroup$

Why single out R?

R, Stata, SAS, SPSS, Mathematica, etc. (and even Excel ;-) are (merely) tools for understanding and resolving questions of statistical analysis. From this point of view, the problem we are addressing here stems from how the original question is framed. An alternative framing that appears more meaningful IMHO is:

  • Questions about statistical analysis should be posted on Stats Analysis.

  • Questions not about statistical analysis (but, presumably, about some aspect of programming) should be posted on SO or a programming-related site.

Isn't that the status quo?

Please, let's not make draconian statements involving "must", "have to", and forced migration. If policies (characterizing what kinds of questions are best posted where) need to be clarified, we can discuss appropriate changes, but let's trust users to follow those policies. Doing so, after all, is in their own interest.

$\endgroup$
3
  • 2
    $\begingroup$ I am not sure that your way of framing is the status quo. The issue is that historically lot of R questions have been asked on SO (that is my impression) and perhaps they are better suited for our site. Also, see the other thread I opened on a poll as to what to do. I agree that R need not be singled out but the issue of SO vs SA is most prominent with R. $\endgroup$
    – svadali
    Commented Oct 30, 2010 at 18:12
  • $\begingroup$ @Srikant The question about status quo was not about what has been posted where but rather about what the current policy is. Maybe that policy needs clarification rather than changing. $\endgroup$
    – whuber Mod
    Commented Oct 30, 2010 at 18:20
  • 2
    $\begingroup$ +1, We've already had discussions about statistical computing/programming questions. I'm a little uncomfortable with singling out one language, however popular. I'm also uncomfortable with an explicit migration policy that takes traffic away from SO -- SE's banner site for programming q&a. It seems this is a discussion that ought be initiated and decided by the vibrant R community on SO. It's somewhat impolitic for us to take such a step. My two cents. $\endgroup$
    – ars
    Commented Oct 30, 2010 at 18:53
4
$\begingroup$

This is what I would like to happen, but many people (especially active in SO [R] tag) would rather believe in idealistic "User should decide in her/his heart if it is or question and post it on SO or here accordingly. High-reps will be migrating misplaced questions."

I currently think that we should just live in the current chaos for a few more months and see what will happen. The flow of R questions here is visible now, so maybe this will happen organically... Or maybe the idealistic solution will justify and work well?

$\endgroup$
3
  • 1
    $\begingroup$ Well they do migrate questions to other websites. So, it is just a matter of obtaining consensus on meta.SO that R questions are off-topic as far as SO is concerned because they can be migrated to stats.SE now. $\endgroup$
    – svadali
    Commented Oct 27, 2010 at 23:19
  • 4
    $\begingroup$ Personally, I believe that questions regarding coupling of databases with R, how to call R from Python, implementation of exception handling in C scripts for R, ... do not belong on a site about statistics. There are a whole lot of questions about R that have absolutely nothing to do at all with statistics. $\endgroup$
    – Joris Meys
    Commented Oct 29, 2010 at 13:58
  • $\begingroup$ @Joris this not a discussion about how the site scopes but about what to do with R related questions posted here (and, parallelly, what to do with R related questions posted on SO). $\endgroup$
    – user88
    Commented Oct 29, 2010 at 16:17
3
$\begingroup$

I think programming questions related to (a particular) statistical analysis should be permitted on SE, irrespective of the tag (be it r, matlab...). We can be splitters (imagine, individual SE sites for linear regression, machine learning, assumption testing... :rolleyes:), but the lines has to be drawn somewhere. I suggest it be draw here. 

    you are here
|---------|ˇ---|

:)

$\endgroup$

You must log in to answer this question.