# What standard should be applied in deciding whether a question requires sufficient 'statistical expertise' to be posted to CV?

The FAQ currently states:

If the language is statistically oriented (such as R, SAS, Stata, SPSS, etc.), then decide based on the nature of your question: if it needs statistical expertise to understand or answer, ask it here; if it's about an algorithm, routine data processing, or details of the language, then please refer to the collection of links to resources we maintain.

What criteria should be used to determine whether a question requires sufficient statistical expertise to be posted to CV?

In your answers please provide a clear criterion and at least three examples that meet the criterion and three which do not.

• Note that this question is related to an answer posted to a previous question on meta, & the subsequent discussion in the comments to that answer. – gung - Reinstate Monica Jan 21 '13 at 16:53

I agree that there is no universal criterion, which is why I am glad this question refers to the plural, criteria.

### Negative Criteria

If any of the following hold, the question likely does not need statistical expertise to understand or answer:

I am not asserting that questions meeting any of these criteria automatically should be closed! These are flags that should cause us to read the question carefully and consider whether the questioner and the community would be better served by migrating it, closing it, or--possibly--leaving it open.

### Positive Criteria

There are too many to name. Generally, questions that ask for interpretations, applications, and examples of things that are explicitly statistical or applied to data analysis tend not to be controversial, whether or not they are couched in a particular programming language or tied to a particular platform. Our site currently has almost 15,000 such questions (out of 17,000 posted).

### On the Edge

Finally, because it might help clarify the preceding, let me describe some characteristics of "edge cases"--those questions that might legitimately cause controversy and be difficult to place. These might include

• Questions that otherwise would definitely be migrated but involve relationships among or interactions between specific statistical or data analysis software packages. It's possible the only people who could answer such questions are here and not on SO.

• Questions about improving the utilization of computing resources for calculations with statistical and data-analytic applications. Although these can be highly software-specific, they can have characteristics unique to statistical analysis and machine learning.

• Questions that look like they belong elsewhere but for which there is some evidence the proposers have some underlying statistical concern. These are questions like "I wrote this code to perform ANOVA but it's failing with negative numbers because the logarithm is not defined for them" (I paraphrase a recent question.) The programming solution would be to trap the bad input, but--recognizing that no correct ANOVA algorithm is going to involve logarithms--the statistical solution might be to ask about the underlying ANOVA question. The first solution would belong on SO (and the answers there would likely help the O.P. implement a working but horribly incorrect solution to their problem), but the statistical approach of eliciting a better question and then answering that would probably be much more helpful.

• Questions that obviously overlap multiple sites, such as applying SDEs (finance might be a better place to get answers); about the roles of data analysts in academia (there's a site specifically devoted to such academic topics); or (hypothetically) about statistical terminology in non-English languages (there are sites specializing in French, German, etc. usage).

• One could easily tag each of the questions under the "only tag" criteria with a relevant CV tag. I suggest this is not a valid criterion. – russellpierce Jan 27 '13 at 0:28
• Also, I'd note a silent creep in what has been written here. You have written that "I am not asserting that questions meeting any of these criteria automatically should be closed". Under the discussions so far under NO CASE should any of those questions be closed. The only appropriate response on the table for these types of questions is migration. – russellpierce Jan 27 '13 at 0:31
• Questions I have asked that have received positive moderation from the community, could all have easily fun afoul for your negative criteria. Please examine the following examples and explain how they meet or fail the second criteria you are proposing. – russellpierce Jan 27 '13 at 0:37
• – russellpierce Jan 27 '13 at 0:37
• Grammatical error on my part that I have now resolved. I was asking for question answers which contained a single criterion. That way each criterion could be assessed on its own merits to yield a set of criteria. – russellpierce Jan 27 '13 at 0:45
• Feel free to break up your response into separate answers so they can be moderated separately. – russellpierce Jan 27 '13 at 0:52
• Consider also this example: stats.stackexchange.com/questions/48520/… – russellpierce Jan 27 '13 at 4:11
• If you want to clarify your criteria by including the recent case stats.stackexchange.com/questions/51443/… feel free. In doing so, please square your response with your comment here stats.stackexchange.com/questions/48155/… that says that judgements must be made on the basis of the question itself, not the possible answers. – russellpierce Mar 8 '13 at 15:42
• That comment is consistent with my answer, @Russell, and adds nothing new concerning these criteria: It was based on my understanding of the question and not on reading any answer. – whuber Mar 8 '13 at 15:59

Proposed criterion:

The question is about a specific aspect of one programming language or statistical software, that does not require statistical expertise to be understood and answered. It could be formulated in an agnostic manner so that anyone used to that programming language or software might provide a definitive answer without resorting to their statistical knowledge. It is off-topic on this site.

Examples classified as off-topic:

Examples classified as on-topic:

There will always be borderline cases. Seconding @whuber's remark, the community has the means to express themselves to each new question, and many users do that already. I am very grateful to them. If a consensual decision could emerge from this thread, this would certainly facilitate the task of all reviewers on this site. In any case, we need to find a way to move forward.

I'm now making a little digression. It is difficult to delineate a clear frontier between applied statistics and the particular use of a statistical software, much like it is difficult to draw a consensual line between SO and CV when it comes to questions that use R or another software to answer a specific question in a somewhat vague statistical context. Because applied statisticians ought to rely on dedicated software, this site was from the beginning expecting questions related to the use of a statistical software to solve specific tasks. Does this mean that we have to collect questions about performing a left join operation using some lisp flavour with a postgresql backend? I don't think so. There are many R-only questions on this site, there are also many statistical questions on SO. An important distinction has to be made, though. This site has evolved since its inception, and so did its general policy on what is on- or off-topic. I believe that it has always been the reflect of what the community is expecting from this site, but I would be very happy to hear from the community in this particular case, and I invite everyone to discussion, here or in a dedicated chat room.

I should note that relying on old threads on our Meta does not do justice to the following:

1. For this site to grow and be accepted in the Stack Echange network, we answered a lot of questions during the beta and just after: I have noticed that, starting in 2011, a bunch of the newly posted questions were about mathematical statistics and applied data analysis, and it certainly helped to attract a new audience;
2. There has been a noticeable but joint evolution/split of CV and SO regarding R programming: it is clear that many users have turned to SO when it comes to answering R questions, but they are still visiting CV to share their expert knowledge in statistics.
3. Stack Exchange has evolved itself, with a new reviewing system and the fact that old threads can no longer be migrated to other sites: it is now easier to spot off-topic questions but we can't clean up all threads.
• Your first off-topic example is great. I worry about the other two RTFM ones. For example, stats.stackexchange.com/questions/5007/… is RTFM but has been highly moderated. Your example "Boxplots using ggplot2" is just a bad question. Too many layers and too specific. The "is it possible" one seems on-topic to me. You are suggesting it is off-topic because it is RTFM? – russellpierce Jan 27 '13 at 23:22
• @drknexus The first example indeed is not really a problem. The Stata question is really bad: The OP hasn't specified which Stata command was used and I only see a technical question about Stata programming. Note that I am not using "RTFM" in a pejorative manner, but those are often questions that show little research effort from the asker. (...) – chl Jan 28 '13 at 11:37
• (...) Questions about the use of legend parameter in a statistical package, or anything else which is about the tool itself (here, ggplot) but does not directly serve the purpose of data visualization to answer a specific statistical question (e.g., how to best represent variation of x and y conditional on z given that z is discrete and observed on a subset of cases for this particular statistical model?), are off-topic according to the above criterion. (...) – chl Jan 28 '13 at 11:41
• (...) Sidenote: Your question on ggplot was posted at the beginning of CV life, and it was a model of a well-formulated question (as are most of your Qs, btw), although I personally found it was OT. You may remember that at that time we were also trying to define good standards for posting Q&As. – chl Jan 28 '13 at 11:42
• I guess to even see how bad the Stata question is, one would have to know Stata (which I do not). Maybe I'm not tracking something, but it seems like the real problem with the two RTFM examples that they showed insufficient research on the part of the question asker... not that they are inherently OT. Although I see in the case of the ggplot example, I can see that you are saying that you think they are now OT for CV. – russellpierce Jan 28 '13 at 19:44
• Can you explain the distinction between the ggplot question and stats.stackexchange.com/questions/27400/…? The later is being used as an example of on-topic, but seems to distill down to an RTFM issue. The title of the question itself specifies 'in R' but the body of the question doesn't reiterate that point. Is it on topic simply because it doesn't hit the reader over the head with the specificity of the domain of the question? – russellpierce Jan 28 '13 at 19:53
• @drknexus Re: RTFM questions -- no, the problem is not much the lack of research effort than the absence of any statistical perspective behind the question, in my view. Re: grouped vs. case-wise data -- I think it has some statistical interest (ML estimates are identical in both cases, but the analysis of residuals or assessment of goodness-of-fit can differ). Moreover, some software do not require to explicitly code negative (0) cases. I'm afraid I may have failed to highlight this distinction in my answer: in fact, there's no real need to transform the grouped data to individual level data. – chl Jan 28 '13 at 22:12
• Which takes us back to an earlier point. The things you just listed are characteristics of possible answers, not inherent in the question itself. – russellpierce Jan 29 '13 at 20:02
• @drknexus I'm not sure I follow your point. The criterion I proposed applies to question content and in my earlier replies to your request for clarification I think I discussed the why and how those questions do not require much than the use of a particular tool regardless of a background statistical question. The last example includes "How could the same model be build using grouped data?". I feel that this is enough of a statistical question, but I may be wrong and I would be happy to hear about your thoughts on that point. – chl Jan 29 '13 at 22:27
• I've created a chat room for further discussion. If anybody want to join, let us know. – chl Jan 29 '13 at 22:27

No universal criteria are applicable. Community moderation in the form of upvotes, downvotes, and actual answers will appropriately control and govern the scope of CV.

Relevant discussions and standards (in chronological order): Example 1, Example 2, Example 3, Example 4, Example 5, Example 6, and Example 7

Example statements arguing in favor of this loose standard:

• "We agreed to welcome such questions, mainly because it is just hard to judge what is good and what is bad in this direction and we don't want to assume everybody here have also SO account" @mbq
• "Answer the question unless it clearly has no statistical content." and "... answer these questions and if no answer is forthcoming, then point them to SO. Not voting the question up for these types of questions will help too. I think it's easier to allow a wider range of questions rather than endless debating whether questions are on or off topic. Also if we close questions too quickly, it can put off people returning to the site - first impressions last and all that". @csgillespie
• "I would like to suggest not migrating existing questions on any basis. Let this site grow as it will" @whuber
• "To me, any SAS, R, S+, SPSS, Stata, etc. question is perfectly valid here" @Baltimark
• "The downside is that R questions will tend to be split between the stats.SE and SO sites. But that is happening already and our distinction of statistical questions here and programming questions on SO is rather fuzzy and almost impossible to moderate. Also, as somebody pointed out on another thread on meta, forcing people onto SO is not very friendly to newbies." @Rob Hyndman
• I agree with all of the points you gathered together here. I happen to have an SO account as well, and I it still seems insane to me that the second question I posted here was migrated to SO! I chose to post my question here instead of on SO for a reason! (original question stats.stackexchange.com/questions/62866/… ; meta-question meta.stats.stackexchange.com/questions/1640/… ) – A.M. Jun 30 '13 at 19:30
• See also: meta.stats.stackexchange.com/questions/1577/…. The only recourse to the continuing violation of community consensus on this point is to have members vote their conscience. If enough people earn enough rep then they can return CV to the path that has been selected by community moderation in meta. – russellpierce Jul 1 '13 at 3:13
• I had visited many questions on this issue, but I had missed the one at the link you posted. It is great. I am very confused, though, since "On topic: Program specific questions (e.g. R, SAS, SPSS, Matlab, C++, python, etc) with any statistical content" has 8 upvotes (including my +1 just now) and an up-voted supportive comment on that answer by the person who moved my question! – A.M. Jul 1 '13 at 16:42
• The people who are against 'programming' questions have a very nuanced position that they have not clearly articulated in a criterion by criterion breakdown where each criterion has been subjected to community review. Part of the problem I suspect is that they don't want to be bound to any one of their criteria because they want to stay flexible... but it has lead to uncertainty regarding what is considered acceptable here with the community consensus being quite far away from the vote choices of those with enough rep to close/move/migrate questions. – russellpierce Jul 1 '13 at 23:58
• Although I would say that although I find it objectionable @chi's criterion is the closest I've seen to anything clear. I really wish it were submitted to the above referenced on/off-topic meta thread. – russellpierce Jul 2 '13 at 0:02
• I am sure there is some great explanation, 'out there', about why having the same question in more than one place is such a terrible idea, but it sure would be nice if the default place to go when one has a question about stats is stats.stackexchange.com. I think this opinion by Jeff Atwood is relevant (and possibly carries even more weight when considered inter-site instead of just intra-site): blog.stackoverflow.com/2010/11/… – A.M. Jul 2 '13 at 1:51
• @A.M.: I think it is that generally people are already answering questions as a matter of their own good will and having multiple people answer it is a potential waste of time/energy/good-will. Jeff's point is a good one, but it certainly points against the value of having the same exact question posted in multiple places. – russellpierce Jul 3 '13 at 2:57
• I just want to point out that he is not just talking about variations of questions. "[...] people tend to ask and search using completely different words, and the better our coverage, the better odds people can find the answer they’re looking for. And isn’t that, really, the whole point of this exercise?" He is talking about questions differing just by search terms in some cases. ...and when you throw in the use of different SE site search bars, an argument even appears, yes, for the same exact question on 2 sites. (Different answerers being on different sites is yet another factor.) – A.M. Jul 4 '13 at 3:18
• I am not even advocating for that much duplication, but Jeff certainly seems to point a long way in that direction. – A.M. Jul 4 '13 at 3:21
• It seems to suggest that rather than questions being marked as duplicate a wiki model of being marked to merge would be helpful. Then presumably the top voted question would be the human readable version with all versions human readable if you decollapse them, but all would be machine readable and found by search engine. – russellpierce Jul 4 '13 at 12:07