We sometimes get what I would describe as "wrong" questions.

For instance, a perennial favorite seems to be questions about binning continuous variables, like What is a best way of binning non-finite continuous variable?, which currently has zero votes, zero answers and a comment (by me) recommending not to bin the data at all - with seven upvotes. Or discretization to create intervals for continuous variables, where I copy-pasted the exact same comment.

I don't think these questions will get a lot of answers, and they kind of veer into "not even wrong" or "type III error" territory.

What should we do about these?

  1. Leave them open?
  2. Find some reasonable existing question we could close these as duplicates of (for the two questions above, this could be What is the benefit of breaking up a continuous predictor variable?), even if it is not a duplicate in the strong sense?
  3. Write an answer that essentially explains why the question itself is mistaken, even if such an answer does not answer the question?
  4. Implement yet another closure reason (I believe that we are already at the limit of the number of closure reasons the UI will support)?
  • 13
    (2) is problematic; (3) done politely is fine--the art here is to reframe the question and then answer your version of it--; but please consider (5): find, edit, or create a canonical answer that addresses the problematic procedure you are concerned about and propose it as the duplicate. – whuber Oct 25 '17 at 21:05
  • 6
    I had thought while reading to leave a suggestion -- that the best thing to do is often to find or in some way make a canonical question-and-answer that these can be pointed to. ... but whuber already made it. So ... I agree with him – Glen_b Oct 25 '17 at 21:56
  • 2
    I mostly agree, but I'm not as against the idea of closing as an imperfect duplicate. If understanding the information in the other thread would lead the OP to recognize the nature of the problem & take the appropriate action (ie, not binning), I think it may be acceptable. – gung Oct 26 '17 at 0:21
  • 1
    @gung: I don't know about that: explaining why the OP should take a different tack is one thing, but closing the q is another - as if we had a party line on the matter. – Scortchi Oct 26 '17 at 12:32
  • @Scortchi: what I'm afraid of is that the question will likely remain unanswered. Unless you are arguing for my alternative 3? – Stephan Kolassa Oct 26 '17 at 12:39
  • 1
    It's already answered (at least your 1st example), along the lines of your option #3. – amoeba Oct 26 '17 at 13:00
  • 2
    The canonical example here of a wrong question -- binning a continuous variable -- is such that (a) I too find myself repeatedly asking "Why do you want to do that? It would just be throwing away information." but (b) it's behind histograms (c) it's utterly standard as a device in economics and finance where looking at the top 10% ... bottom 10% of firms on some criterion is a common strategy. – Nick Cox Oct 26 '17 at 13:42
  • 3
    (ctd) I was grumbling about (c) to a senior biostatistician (disclosure: not Frank Harrell). His reaction was "Perhaps that's not as crazy as it sounds. Looking at the 10% most fragile ... 10% most robust patients might be a method we should be thinking about with medical data". So, what looks like a very misguided question may arise commonly because it's a frequent practice in some field; from one perspective it really doesn't seem misguided. Other examples could be over predictor selection in regression-like models or correcting P-values for multiple significance tests (or not). – Nick Cox Oct 26 '17 at 13:48
  • 3
    @StephanKolassa: I'd say answer along the lines of Option #3 either when you've something to say about why binning is a bad idea in the OP's particular situation or when there's not already a post dealing with the pros & cons of binning in general. Else leave a comment & a link, as you have been doing: if no-one cares to answer a (let's suppose) clear, focussed question on methods that are falling into desuetude (let's wish) except in special cases, what's the problem? – Scortchi Oct 26 '17 at 15:45
  • 2
    @Scortchi "desuetude" is a condition resembling that of old, battered, stained suede shoes lying in a cupboard. Not worn for years, but not yet thrown out. – Nick Cox Oct 26 '17 at 16:49
  • @NickCox: I agree that there is often room for argument, e.g. "stepwise regression has the following problems ..." is better than "stepwise regression is wrong." With multiple answer(er)s, we have room for a bit of nuance/variety of opinion. – Ben Bolker Aug 24 at 21:50
up vote 17 down vote accepted

As a beginning user, I would like to give my opinion on this and why I strongly favour (3) if it is possible to write such an answer clearly. An example where I learned a lot from reading a "wrong" question is this one by @gung.

  1. Write an answer that essentially explains why the question itself is mistaken, even if such an answer does not answer the question

In short: When posing a question about how to implement a method, the OP already assumes the method is valid. Of course there are also great questions of the form "Why is it bad to do X", but people who assume their method to be correct probably do not end up finding such questions if they already exist.

For example, stepwise model selection procedures is still taught in many statistics courses. Answers such as this one, on what is currently the top question sorted by 'frequent', shed light on topics that are very often misunderstood in statistics.

Similarly, binning continuous variables as you mention in the question is something that I think needs a good explanation why it is bad. This might result in more people wondering how to bin their continuous variable, searching or writing up a question and finding: "Ah, someone else already asked this, let's have a look."

In an ideal world everyone would first search for questions related to their method to find out if they are doing the right thing at all, but I think the fact that such questions keep reappearing suggests that many don't. Hence I really appreciate answers such as gung's which not only address a frequent 'problem' on the site, but are also very informative for those searching something along the lines of "How to do stepwise".

In fact, if there were a convenient way to find these "wrong" questions with great answers, I would enjoy reading them. They are also of great help when teaching students about common pitfalls and fallacies in statistics.

Of course, if such a Q&A already exists, it can be marked duplicate.

  • 3
    Thanks - this seems to be what the community likes best. You unerringly picked the alternative that will make the most work for us, of course. (We expect you to pitch in - after all, with 864 rep, you are hardly a "beginning user" any more ;-) – Stephan Kolassa Nov 2 '17 at 22:45
  • 2
    I would be happy to contribute and will continue to do so as long as my contributions are appreciated :) – Frans Rodenburg Nov 3 '17 at 14:31

I agree with Frans: #3 is the only constructive way imo.

I'm almost a newbie here and on both sides of the line: I sometimes ask, sometimes answer. I'm sometimes frustrated or puzzled at questions I feel as absurd. I sometimes feel angry at people considering my questions as "wrong" without a second look. If what I wanted to know was easy for me to find out by browsing the net, well usually, I would have found it somewhere else and wouldn't ask. Even if I am actually blinded by a common misunderstanding, don't know the keywords to use in Google, don't understand some articles I read, I am definitely in a situation where I can't clear it up by myself.

Unless the question is unclear, poorly formatted, unrelated to stats/ML, there is no wrong question. Saying that the point of view of the person who asks misses something is naive, a dead end or whatever IS definitely an answer to the question.

Like any answer it is worth explaining what we mean as clearly as possible, possibly providing references as usual. Also, we might be wrong, miss some knowledge or misunderstand what the person asks, or assume things that do not match the real situation where the question appears.

Writing clear and thoughtful answers allows (nice) confrontation in the community. Answers can be commented, improved, clarified, voted up/down... efficiently while it's messy with comments. Also some comments can be felt as aggressive sometimes. When you take time to write an answer, this aggressiveness disappears: since it's not so easy to explain, I guess it's not easy to understand either.

I think if a person makes the effort to write, update, improve... a clear question, then taking time to answer is not a waste of time (generally speaking).

  1. Many times a question is based on a false premise, or indeed several of them (a common example is a question about transforming to normality something that is not ever assumed to be normal - like the predictor variables in a linear regression model).

    In that case I think a perfectly reasonable response is to post an answer that corrects the erroneous idea on which the question is based -- but ideally one that offers some kind of path forward for the OP. This may require some steps of clarification of the issue the original poster is working with.

  2. These are often questions with what is sometimes called the XY problem (also see wikipedia).

    This happens when a user asks about an issue with their attempted (but often naive) solution to their original, underlying problem (typically not even mentioned in the question).

    For that reason it's often important to try to tease out what prompted the question, and either get the question edited in a way that relates to the real problem, or - where there's value in having an answer to the original question (e.g. because it's a common error that for some reason hasn't already been adequately dealt with on site) - to explain the mistaken premise but encourage the OP to ask a new question relating to the original problem.

You must log in to answer this question.

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