37
$\begingroup$

Two hours ago (ish), a new user posted this question on two-tailed F-values - how to do them in R.

He (or she, but I'm going to say he) was told to take it to SO. So, he did what he can see to do -- he deleted* his question here, and subsequently reposts on SO.

* it's no longer deleted right now. See the update.

To little surprise, they see it as a stats question, and migrate it straight back here.

The OP has already been told it doesn't belong here, and deletes the second copy.

He's a new user who has been scared into deleting his own question, twice.

He's acted in good faith but risks getting himself auto-banned from being able to ask questions at all (and no, making a new account won't work either) -- a new user has no way of knowing this.

I'm somewhat disturbed by the eventual outcome here, especially in the way a new user that seemed to be trying to do the right thing was run around and ended up with a really bad outcome (two deleted questions). I think we should aim to find a way to do better than this for our new users, even if their questions do eventually close.

For openers, I think:

1) we need to avoid just saying "it belongs on SO". If it looks like a stats computing problem, there's a good chance we're wasting everyone's time. If you're not looking to see what happens to these posts when they go there, you probably should avoid sending people there. (Better to close a question, but we can do more than that.)

2) Even if we do decide to send them there, we need to make sure that new users know not to delete and repost. If it (deleting their own post) happens enough to a low rep user, it can kill their ability to post and we mods can't fix that ban. Generally it's better to migrate (and a few more heads get to have some input). But we should never migrate a poor question -- we have to improve it first, just as if it was a question here.

3) We also migrate too readily. Don't just migrate everything you think has something to do with computing; again, if you're not paying attention to what happens to these posts when they get migrated there, don't be quite so hasty. They often come straight back.

4) If a question must close, then so be it, but we should seek to give as much help as we can on the way.

5) I think we do need to take a slightly more lenient attitude to these questions (especially with new users). If you look back two or three years ago we have many similar questions that remained open, were answered and got upvotes. Some have a lot of upvotes.

Edit: In respect of (5) (and in turn, (1) and partly (3)) -- in particular there's a category of posts that if asked here tend to be migrated to SO and if asked on SO would tend to be migrated here, because enough people there see it as fundamentally a stats question to migrate (it would also include a fraction of the posts we close instead of migrating). To me this isn't a tenable situation, and I think we should look at trying to actually take a larger portion of those posts that a substantial portion of both sides think fit on the other side. This is not the same thing as taking a large fraction of posts that clearly belong on SO. I don't think this will vastly increase our volume -- it's not like the migrations are in the tens per day.


So to the main question: What are good ways to improve the outcome here?


Some followup:

I undeleted both questions; hopefully that helps avoid risking an auto-ban. I closed as duplicate the migrated one (which rejected the migration, oddly - I thought close as duplicate was one option that didn't reject the migration).

The original needs to be improved -- initially it should be clarified even if it migrates or we send the OP somewhere else; I think it can be rephrased to a statistics question ("How do I calculate two-tailed p-values with an F-test?" -- which certainly is on the 'requires statistical expertise' -- they shouldn't be dealing with that on SO) with a small R-component (i.e. giving the line of code required to implement the explanation).

I think this particular question is actually a stats question in disguise ... there are a bunch of issues about two-tailed F tests that are often handwaved or glossed over that should be discussed before going anywhere near how to implement any of it in any packages -- and I think once the statistical issues are clear, the computer problem is trivial (and would likely survive on SO if asked as its own question, though I doubt it's necessary).

$\endgroup$
  • 1
    $\begingroup$ if it's too much programming for xv, and too much statistics for so, does it belong on the data science se? $\endgroup$ – aeroNotAuto May 4 '15 at 12:56
  • 5
    $\begingroup$ I voted to close that Q, but I voted as "unclear...", not to migrate. In general, we should migrate Q's w/ a reproducible example (ie data & code), & not otherwise. $\endgroup$ – gung - Reinstate Monica May 4 '15 at 14:17
  • $\begingroup$ I voted "not to close" that question---the statistical content was clear, I am surprised it was migrated. $\endgroup$ – kjetil b halvorsen May 4 '15 at 16:37
  • $\begingroup$ @kjetil It wasn't migrated from here. The OP concluded from the advice there (the advice was fine; it's the outcome I'm concerned about) that such a question should be deleted here and reposted to SO. SO then migrated it here. [However, similar posts do migrate on the basis that it "belongs on SO", and if it hadn't been deleted it would have been likely to migrate.] $\endgroup$ – Glen_b May 4 '15 at 16:39
  • 1
    $\begingroup$ either way, this new format is really cheesing me off at least. I agree that history does not belong on here; etc. But why, in an age when data science,statistics, and programming are becoming more ubiquitous is there any distinction at all. Cross-breading has a huge benefit (mathematicians learn to program, programs learn new stats for AI, graphics; etc. win-win) $\endgroup$ – user3916597 May 7 '15 at 16:03
  • 1
    $\begingroup$ I think programming-related statistics questions should be welcomed here. I very often answer questions on SO which are basically statistics questions, but framed as programming questions in Matlab (or R, ...). $\endgroup$ – A. Donda May 8 '15 at 17:38
  • 1
    $\begingroup$ I don't have the faintest idea why anyone would think that is a SO question $\endgroup$ – shadowtalker May 9 '15 at 13:11
  • 2
    $\begingroup$ As a diamond mod on SO I can say that it being migrated back here was absolutely wrong. That question should have been closed there as off topic, not providing a clear, reproducable problem (in other words, "what have you tried, what is the error you're seeing"). Sorry that it got kicked back here when it shouldn't. $\endgroup$ – casperOne May 13 '15 at 12:29
15
$\begingroup$

I often, but not always, vote to close many such questions. Why?

There is no axiom that all questions have a place somewhere on SE. Questions on the routine use of some language don't belong here and they don't belong on SO if they are not about a (non-trivial) programming problem raised by the poster's own code. For example, questions about "How do I do X in Y?" don't belong in either place. I don't pose as a hard man in saying that, as I am active on SO too, I have answered thousands of such questions about Stata on Statalist and I respect greatly those who do similar things for SPSS, SAS, R, Python, MATLAB, Excel, etc. in appropriate forums. (It remains axiomatic also that people should Read The Fine Documentation.)

The argument is pragmatic as much as one of principle. If any and every question on use of statistical programs is allowed here, then we would be swamped. Traffic could easily rise by a factor between 2 and 10. If that happened, I think many active members would feel that the forum was too busy to manage. If that's not so, then we need a policy on precisely what is allowed, and we do have one.

I don't buy any implication that people new to the forum, or to statistics, or to any statistical computing platform have a right to ask different questions or a broader range of questions here. If the question on-topic. that is fine. If not, not. If I walk into a shop and the answer is "Sorry, we don't sell that here", then that's the way it is. I expect a similar attitude from new members. It's a little wry when new members start ruling for themselves what is on-topic as against moderators or members with moderate or high reputation, but there you go.

I try always to follow the statement on what is on-topic at https://stats.stackexchange.com/help/on-topic

  • Members who think the policy is too hard on certain kinds of statistical computing questions need to say what policy they would prefer. We've been around this more than once during my time on the forum (2 years) and no one ever has different wording to offer that buys widespread assent.

  • Members who think that particular decisions are being made in conflict with the policy need to spell that out, because if so decisions need to be revisited, which is always in order.

How does the example raised by @Glen_b bear on this? I have to say that as posted I think it is a weak question and doesn't really belong in either forum. I have no objection to people trying to improve it, but the main obligation is with the OP to improve it. (In particular, posts with the flavour "I don't understand" are making statements, not asking questions, a hard principle to learn.)

Elsewhere I am intensely interested in historical questions but here I don't find what was allowed a few years ago to be a compelling argument. A forum is allowed to, indeed expected to, evolve as it matures and works out its own style, policy and modus operandi. History is especially irrelevant for new users.

$\endgroup$
  • $\begingroup$ Voting to close is certainly a reasonable choice for questions like these; the OP still has a chance to improve the post, or to ask for migration (though I wouldn't migrate that one as it was). Additionally, our usual close reason for the code-related problems points to a list of resources, so there's some increased opportunity to find a solution. Even if we can't do anything else for them, closure is a good option. $\endgroup$ – Glen_b May 4 '15 at 8:40
  • $\begingroup$ I don't participate on this site, at all I think (through I have some interest in data analysis, as economist, I haven't found a reason to participate), but what you say about "There is no axiom that all questions have a place somewhere on SE" is another way of the saying "we don't migrate crap" which I wholeheartedly agree and follow. Crap is crap, wherever you see it. $\endgroup$ – Braiam May 9 '15 at 19:11
  • $\begingroup$ @Braiam I am pleased that you think you agree with me, modulo a difference in wording. But my main point is just that many questions are off-topic both here and on SE. Many would go fine in other forums. It is true that many do not show research effort, or are unclear or not useful, but that doesn't seem the point of this thread. $\endgroup$ – Nick Cox May 9 '15 at 19:38
9
$\begingroup$

Well done Glen for sorting this out! I think the OP in fact got good advice in the very first comment—to clarify their question, that questions purely on programming are for SO, & even what function to look into using—but they didn't know how to act on it. One, hopefully uncontroversial, lesson we can learn from this straight away is to take especial care to explain to new users what they should do & what will happen next: "Edit your question to make clear [...] & then, if it is purely to do with programming, it can be migrated to SO."

$\endgroup$
  • $\begingroup$ Indeed; @Glen_b tried very hard to get the OP to improve the question, evidently with no success (although perhaps time zones interfered). $\endgroup$ – Nick Cox May 4 '15 at 16:11
2
$\begingroup$

This is a typical question for someone learning R. R's docs are essentially incomprehensible for newbies, especially if someone's learning statistics at the same time. In fact many R old dogs tell newcomers to Google the answers when they complain for poor documentation.

Therefore, I don't think we should be supporting R programmers here. It would only perpetuate the unacceptable state of this package's documentation. Whoever cares for R should contribute to its documentation directly. It's open source software, so it's easy enough to do.

Take a look at examples of F-test documentation in R and Stata. R's examples are four lines of code with two lines of comments like "# Do x and y have the same variance?". Stata's doc has 10 fully documented examples with comprehensible text. The biggest difference is that Stata's help (in 19 pages!) is pretty much self-contained. This is very important for a beginner. An experienced dude will get a clue from R's doc, and will know what to do or will know where to get the rest information. The newbie will check the doc, get no idea of how to use the function and come here to ask a question. Stata and other commercial packages figured that a new user needs self-contained help sections, where the help has 99% of what you need to run the function.

Is it very difficult to write this in R? No. It's boring and time consuming, but whoever wrote the code must know all the stuff that Stata guys know. In fact, that's what I do quite often when using R's functions: I go pull SAS or Stata's doc to understand the method, then simply check the syntax in R's docs. This must stop. The basic usage of the R's functions should be explained in R's documentation, so we should stop supporting R users in this regard.

$\endgroup$
  • 3
    $\begingroup$ How would that work? Suppose I write an R package [that won't happen for other reasons, but consider the hypothetical] but don't document it well: do you then edit its documentation or do you write a separate guide? (I know this is a side-issue to this thread, but I am curious about what you envisage.) $\endgroup$ – Nick Cox May 7 '15 at 7:18
  • $\begingroup$ That would be maintained somewhere with sources. So I would send the pull request or a patch to documentation. You would consider my contribution and accept it. Since you don't give a dam about docs, it's likely that my contribution would make a big positive difference. $\endgroup$ – Aksakal May 7 '15 at 10:49
  • $\begingroup$ I'd be interested to know if this ever happens. In the Stata community user-written contributions are essentially open source too, but essentially what happens when programs are poorly documented is they are ignored or complained about, but other people competent in the same territory don't rewrite their lousy documentation; they're more likely to write their own programs instead. $\endgroup$ – Nick Cox May 7 '15 at 11:09
  • $\begingroup$ I doubt that it happens a lot. Writing the docs is time consuming. People want too be compensated for it, so they end up writing books. That's why there's so many books on open source, it works out well with poor documentation. The developers like developing they don't like writing docs. However, since all code repositories are public, it's possible in principle to contribute to documentation. $\endgroup$ – Aksakal May 7 '15 at 11:45
  • 3
    $\begingroup$ Interesting; thanks for the comments. By the way, I don't fully support your stance here about R. It just seems to me that "documenting the software" and "teaching the users about the methods" are treated rather differently by the R community than by firms selling proprietary software. $\endgroup$ – Nick Cox May 7 '15 at 11:55
  • $\begingroup$ Your examples aren't comparing the same things: R's is for, as it says, "an F test to compare the variances of two samples from normal populations"; Stata's "performs Wald tests of simple and composite linear hypotheses about the parameters of the most recently fit model". (I'm sure you could find better ones though.) $\endgroup$ – Scortchi - Reinstate Monica May 7 '15 at 15:59
  • 7
    $\begingroup$ CV is committed to willingness to answer statistical questions regardless of how well the software that people use documents the underlying methods. So, R users who ask statistical questions are just as entitled to support as anyone else. $\endgroup$ – Nick Cox May 7 '15 at 16:24
1
$\begingroup$

It belongs on R.SE. This was a straight R.SE question that just fell between the cracks. To avoid confusions like this one (and many others), I think the idea of setting up R.SE should be revived, even though the powers that be at the SE level close the proposals within minutes.

$\endgroup$
  • 1
    $\begingroup$ I once supported an Area 51 proposal for a Matlab site, which is closed now. But already then I had the feeling that maybe the community is too small to warrant its own site. On the other hand, I feel that the Matlab community on SO is something of a foreign object. What about a site that brings together people interested in programming for data analysis & statistics regardless of language, covering R, Matlab, Python with SciPy, etc. etc.? $\endgroup$ – A. Donda May 4 '15 at 14:50
  • 3
    $\begingroup$ Some would argue that DataScience.SE is such a forum. $\endgroup$ – StasK May 4 '15 at 15:20
  • 2
    $\begingroup$ In principle, yes, but it doesn't feel like that. I monitor the matlab tag on SE, and so I answer questions across SO, CV, math.SE, DSP etc., but there hasn't been a question on DS that I felt inclined to answer. One reason might be that it is much more "machine learning" than statistical data analysis. I'm aware that the two cannot be cleanly separated, but my feeling is "data science" is dominated by engineers that just want something to "work", and often operate under the assumption that there's nothing relevant that is older than 30 years. $\endgroup$ – A. Donda May 4 '15 at 15:38
  • 1
    $\begingroup$ Then how about welcoming statistical programming questions on CV? On SO, I often find that the OP needs help with understanding the statistics just as much as with implementing the analysis, so the line is blurred anyway. $\endgroup$ – A. Donda May 4 '15 at 15:39
  • 2
    $\begingroup$ 30 years??? These must the engineers who are in their sixties to recall that far back. And TAOCP was first published, mind you, more than 50 years ago. $\endgroup$ – StasK May 4 '15 at 15:43
  • 1
    $\begingroup$ I was thinking about Machine Learning, since 1986. $\endgroup$ – A. Donda May 4 '15 at 16:18

You must log in to answer this question.

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