17
$\begingroup$

We have the following rank-related tags:

Problems:

  1. [wilcoxon] is ambiguous. Worse, one meaning of [wilcoxon] is synonymous with [mann-whitney-u-test] and another meaning is synonymous with [signed-rank-test].
  2. [spearman] and [spearman-rho] are synonymous.
  3. [rank-correlation] might be better off sorted out into specific types.
  4. [ranking] and [ranks] have somewhat unclear scope.

My current proposal is:

  1. Create [wilcoxon-rank-sum] and [wilcoxon-signed-rank] tags. Done.
  2. Make [wilcoxon-rank-sum] and [mann-whitney-u-test] both synonyms of [wilcoxon-mann-whitney]. Done by @Glen_b.
  3. Make [wilcoxon-signed-rank] the synonym of [signed-rank-test]. Done by @Glen_b. Later reverted by @Scortchi, following @ttnphns' answer.
  4. Go through all [wilcoxon] questions and move those about rank sum into [wilcoxon-mann-whitney]. Done by @mdewey.
  5. Move all the remaining threads from [wilcoxon] into [signed-rank-test] via mod merge hammer. Done by @Glen_b.
  6. Merge [spearman] into [spearman-rho]. Synonym created via voting. Merge done by @Glen_b.
  7. Sort as many questions as possible from [rank-correlation] into [spearman-rho] and/or [kendall-tau] and/or [goodman-kruskal-gamma]. It looks like there will be ~30 questions left though. Done. Thanks @mdewey.

  8. Go through [ranking] and [ranks] and try to understand what to do with them. I leave this open for now because this involves 400+ threads.

$\endgroup$
  • 1
    $\begingroup$ +1 Good suggestions all over. Especially the wilcoxon is necessary because it clearly misleading users substantially. Would you make [rank] and [ranking] synonyms? $\endgroup$ – usεr11852 says Reinstate Monic Feb 18 '16 at 19:25
  • $\begingroup$ Given the confusion generated by [wilcoxon], I wonder if it would be better not to make it a synonym, but just to destroy it. I don't know. My thought is that someone might still come along later & type that & automatically get the one synonym, when they had meant the other. $\endgroup$ – gung - Reinstate Monica Feb 18 '16 at 19:49
  • 3
    $\begingroup$ @gung, I think we should treat [wilcoxon] similar to [student-t]. I would create two new tags [wilcoxon-ranksum] and [wilcoxon-signed-rank], then go through all threads in [wilcoxon] and sort them out into these two, and finally make [wilcoxon-ranksum] a synonym of [ranksum] (together with [mann-whitney-u-test]; the master thread name can be discussed though) and [wilcoxon-signed-rank] a synonym of [signed-rank]. I don't think we will need to blacklist [wilcoxon] after that. $\endgroup$ – amoeba says Reinstate Monica Feb 18 '16 at 21:26
  • 2
    $\begingroup$ @usεr11852, I guess that [rank] can be a synonym of [ranking], but somebody has to look at these tags in more detail to see if it makes sense. I am not sure we need a ranking tag on questions about ranksum test or spearman's correlation; this seems superfluous. So I have a feeling that in many cases it can be removed altogether (or replaced by a more specific tag). But there probably are some general questions on ranking or rank-based methods for which the tag would make sense. $\endgroup$ – amoeba says Reinstate Monica Feb 18 '16 at 21:30
  • $\begingroup$ Some good suggestions here. If I recall correctly, "Spearman rho" and "Kendall tau" are far from the only type of rank correlation though, so I don't think it is so easy to get rid of "rank correlation" so easily. $\endgroup$ – Silverfish Feb 19 '16 at 19:53
  • 4
    $\begingroup$ Would you say that these tags are... rank? $\endgroup$ – Sycorax says Reinstate Monica Feb 19 '16 at 21:28
  • 1
    $\begingroup$ @user777 Had to consult the dictionary to get the joke :-) Thanks. $\endgroup$ – amoeba says Reinstate Monica Feb 19 '16 at 22:36
  • $\begingroup$ @amoeba, voted. $\endgroup$ – gung - Reinstate Monica Mar 2 '17 at 13:21
  • $\begingroup$ I am working my way through rank-correlation. I have done all the easy ones. One thing I notice is that there are some about rank biserial correlation. We have no tag for biserial correlation and there are about 35+ questions about it. Shall I create it and tag them? Should I do this as a separate proposal? $\endgroup$ – mdewey Mar 12 '17 at 12:16
  • $\begingroup$ @mdewey To be honest, I've never heard about "biserial correlation" before. Do you mean this: en.wikipedia.org/wiki/Point-biserial_correlation_coefficient? The article says "The point-biserial correlation is mathematically equivalent to the Pearson (product moment) correlation". If this is true, then I guess the existing [pearson-r] tag is enough, isn't it? $\endgroup$ – amoeba says Reinstate Monica Mar 12 '17 at 20:31
  • $\begingroup$ ranking is the task for a respondent to rank stimuli (i.e. it is distinquished from rating or labeling/categorizing. ranks are transformed, ranked values. $\endgroup$ – ttnphns Mar 13 '17 at 13:38
  • 1
    $\begingroup$ I am doing them on an ad hoc basis. Some of them are not really about rank correlation at all. There will be a few left which have totally defeated me and I will let you lnow when I have to show the white flag. $\endgroup$ – mdewey Mar 13 '17 at 21:25
  • 1
    $\begingroup$ I have now reduced rank-correlation to a handful. These are ones where I really do not see what the best course of action might be and would welcom someone else having a go. The ones I eliminated were mostly spearman-rho, some kendall-tau, and a mixture of other statistics which were not really rank-correlation at all IMHO. In the process I created a tag somers-d with a wiki. There are a few more Somers' D questions which I shall find and tag but they did not have rank-correlation. $\endgroup$ – mdewey Mar 14 '17 at 12:48
  • 1
    $\begingroup$ @mdewey There are only 2 closed questions left in rank-correlation, so it should disappear after the questions are deleted (I think it's 9 days after closure so still in a couple of days). Now we will have to sort out [wilcoxon]... $\endgroup$ – amoeba says Reinstate Monica Mar 15 '17 at 20:02
  • 1
    $\begingroup$ I think most of them are due to me. I will have a look at some of the others starting from the oldest. $\endgroup$ – mdewey Apr 2 '17 at 20:18
5
$\begingroup$

Signed-rank-test to Wilcoxon-signed-rank test

[This answer was originally posted as a separate question (you still can see comments there) and now transferred to here as an answer. It is my request about the tags of two "Wilcoxon" tests, independent-sample and paired-sample.]

Moderators, please rename tag signed-rank-test (Wilcoxon paired sample test) to wilcoxon-signed-rank. We have curently such tag, but it is a junior synonym of signed-rank-test. So, please appoint wilcoxon-signed-rank to be the main tag and signed-rank-test its junior synonym: revert.

For the unpaired Wilcoxon, we have nice tag wilcoxon-mann-whitney. Thence it is logical to have wilcoxon-signed-rank as its paired counterpart.

Note that signed-rank-test isn't optimal because it doesn't mention "Wilcoxon". It thus can be confused with the sign test. Besides, there exist other, less known, rank tests based on signs of the rank differences, which adds to the confusion. Finally, a number of software (SPSS for example) and books use name "Wilcoxon" only with the paired signed-rank test and never with the unpaired test which they call "Mann-Whitney"; it might be an argument to state that "Wilcoxon" have stronger - at least not lesser - belonging to the paired test than to the unpaired and therefore should be included into the tag of the paired test.

An alternative name: create wilcoxon-paired-test instead of wilcoxon-signed-rank, if you would like it better. I like both variants of the tag name.

$\endgroup$
  • 2
    $\begingroup$ As I said elsewhere, the direction of a tag synonym (once the tags are merged, as these two are) makes almost zero practical difference. It's mostly a question of what tag name we like more. I don't have any strong opinion here, and I am certainly fine with doing the reversal that you suggest. Perhaps you are right and it will make more sense like that. The synonym was originally created this way simply because signed-rank-test was already an existing tag, there was no other reason. $\endgroup$ – amoeba says Reinstate Monica Mar 13 '17 at 13:36
  • 2
    $\begingroup$ @amoeba, thank you for reply and clarification, and for potential agreement. I hope that my arguments in the answer could weigh sufficiently to do the re-tagging. Both tags ought to start with wilcoxon..., for me. (Thank you for creating the very topical thread on ranks.) $\endgroup$ – ttnphns Mar 13 '17 at 13:42
  • 2
    $\begingroup$ I've swapped the tag names so that wilcoxon-signed-rank is now the synonym. $\endgroup$ – Scortchi - Reinstate Monica Mar 15 '17 at 11:01
2
$\begingroup$

TL;DR the tag now only contains tags which can be mapped to so if any moderator is reading this can they please carry out the merge operation mentioned by Amoeba in the original question.

Longer version

There were a few which did not map oonto [wilcoxon-mann-whitney] which I have re-tagged appropriately. One remaining issue is that there was one about the Wilcoxon-Gehan method. I have temporarily created the obvious tag for it. We have a handful of other questions which could be re-tagged to that tag. Shall I do that and provide a wiki for it or is there some other existing tag which could be used? I have an unmet training need for time-to-event analysis so would welcome second opinions here.


The remaining issue, as far as I can tell, is the [ranking] and [ranks] tags. My personal view is that there are too many of these to make a thorough examination of each one a viable proposition.

$\endgroup$
  • $\begingroup$ There might be a few left over threads, at least I find several with the following searches: stats.stackexchange.com/… (3 Qs), stats.stackexchange.com/… (2 Qs), stats.stackexchange.com/… (1 more Q). I did not look into them (yet), but some of them might need retagging too. $\endgroup$ – amoeba says Reinstate Monica Apr 9 '17 at 13:42
  • $\begingroup$ Re Wilcoxon-Gehan, I know nothing about it, but this search stats.stackexchange.com/search?q=Wilcoxon+Gehan+is%3Aquestion (only 5 Qs mentioning Wilcoxon and Gehan) suggests that this tag is not much needed. Perhaps tagging with [survival] is enough? $\endgroup$ – amoeba says Reinstate Monica Apr 9 '17 at 13:46
  • 1
    $\begingroup$ As far as your left over ones are concerned I think many of these are in fact signed ranks even if they speak about rank sum. I suppose I could double tag them which is what I did with some of the others. $\endgroup$ – mdewey Apr 9 '17 at 14:04
  • $\begingroup$ I am somewhat reluctant to have [wilcoxon] made a synonym of wmw, in that w could be used for a number of things. In the future, new or careless users will type w for what they want (something else) & then it will be mapped to wmw & thus mistagged. W/ 129 remaining, it's a heavy lift, but I wonder if manually retagging them & letting w die might be preferable. $\endgroup$ – gung - Reinstate Monica Apr 9 '17 at 15:40
  • $\begingroup$ @gung in fact the remaining ones are supposed to be wilcoxon signed ranks but I take your point. I think the safest thing is to do them manually. It is only 129 whereas there were nearly 300 so will not take many days to achieve. $\endgroup$ – mdewey Apr 9 '17 at 15:44
  • 1
    $\begingroup$ I have now re-tagged the Wilcoxon-Gehan one with [survival]. $\endgroup$ – mdewey Apr 9 '17 at 16:19
  • 1
    $\begingroup$ @gung The idea is not to make [w] the synonym of [wsr]. The idea is to transfer all threads from [w] to [wsr] via tag merge and then to delete [w] tag altogether. $\endgroup$ – amoeba says Reinstate Monica Apr 9 '17 at 17:55
  • 1
    $\begingroup$ @amoeba, my mistake. That's a fine solution, IMO. $\endgroup$ – gung - Reinstate Monica Apr 9 '17 at 18:00
  • $\begingroup$ I looked at the threads I linked above and I agree they are all about signed-rank. I'd say we can go ahead with the merge. $\endgroup$ – amoeba says Reinstate Monica Apr 9 '17 at 19:14
  • 2
    $\begingroup$ I merged as requested. There were a few threads that looked kind of borderline so if you see any that should have the WMW tag please add it. $\endgroup$ – Glen_b -Reinstate Monica Apr 10 '17 at 7:08
  • $\begingroup$ @Glen_b and mdewey: Thanks. I leave this thread open for the time being, because I am not sure what's the decision should be regarding [ranking]+[ranks]. Should they be synonymized? Should they be split into sub-tags? Should we leave them as is? What are these two tags even about, are they all about rank-based tests & rank-based correlations (what's the common term for that?), or are there other topics in these tags? $\endgroup$ – amoeba says Reinstate Monica Apr 10 '17 at 7:48
  • $\begingroup$ "ranking" could be relevant to more than just rank based tests and correlations but if in practice it isn't used for anything else that's fine. I don't know that the two should necessarily be covered by a single term. $\endgroup$ – Glen_b -Reinstate Monica Apr 10 '17 at 10:36

You must log in to answer this question.

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