One rule per answer, upvote if you agree, downvote if disagree.
Each statistical application/programming language gets its own tag
Like [R], [stata], [ruby], [python], [statistica], ...
If you don't know what tag to use, use [for-retag]
This will be an invitation for (powerful enough) users to retag the question appropriately. Old ones can be fixed during some retag campaigns coupled with chat discussion.
On the other hand, we can discuss lowering the reputation threshold to make such retags with SE team.
There will be no [statistics] or [statistical-analysis]
In theory it can be useful for pure statistical theory, but on the other hand it would be used by new users to tag almost all questions. On the other hand, [theory] can be used in this case.
(This will sound like a bunch of rules but it's really just one.)
Tags should accurately reflect both the nature of the question and the answers and comments that accompany it (i.e., tags need not describe only the question)
... in such a way that each individual tag clearly discriminates the thread from other threads
... but so that the collection of tags almost uniquely identifies the particular thread to which it's associated
... without being too numerous.
When this is not possible to do using current tags, either a new tag (or tags) need to be generated or--with due care not to upset older threads and their tags--existing tags could be modified.
Shorter tags are better
For instance [pca] instead of [principal-component-analysis]. Tags have length limit. Abbreviation conflicts can be fixed using synonyms.
Individual researchers/authors/etc. should not be tagged.
I could actually go either way on this one, but the recent question on Nassim Taleb got me wondering.
Popular packages/modules should have their own tags
I was the person that added the
e1071, so I obviously disagree with the idea proposed by @onestop.
The question being discussed currently has two tags:
e1071. I would argue:
- Just having the R tag by itself isn't particularly useful for any question.
- For this particular question, other tags don't really fit well, For example,
Bayesiandoesn't seem right. In the future the only person who will be interested in this question is someone who is using the
e1071package. Why not use it as a tag?
- We can have up to five tags. Most questions only have 2 or 3 tags.
- The R tag is the most popular and is currently used in 15% of posts. It seems like an obvious way to help searching for R is to use package names.
- @onstop suggests that only main packages have there own tags, i.e.
e1071is used by a number of bioconductor libraries. Therefore, (I would guess) it is probably used by more people than
ggplot2. Admittedly, most people don't know it ;)
- SO overflow use packages tags for searching. For example, an Rcpp tag exists.
- A tag is deleted after six months if it is an orphan.
- Most posts in this site suffer from a shortage of tags, why delete tags?
- The R & c++ example seems strange. Again taking SO as an example, no Rcpp post has a c++ tag. It seems like obfuscation to remove the Rcpp tag.
Site-wide retags (aka massive retags) need community consensus unless they involve one or more of the following;
- Spelling errors
- Retags that leverage established renaming rules
A retag of an individual question need not have a corresponding meta thread unless it attracts controversy.
Adding new tags by retag should be exceptional
... in general, we should use those rules to cope with arriving tags from user questions, not to proliferate new tags to fulfill some rule and then glue it to existing questions.
Exception: merging, unexpected special issues.
Each method gets its own tag
... but on reasonable generality level, eg. [cross-validation] not [elevenfold-cross-validation-with-random-resampling-for-butter-taste-data].
Reasonably general can be judged in such a way: we wait a month and see if any other question used the tag. Exception: if a [submethod] tags 100% of [method], we delete it.
Each notable term get its tag
Like [hetroscedasticity], [ordinal], [median], [categorical-data], ...
Some of those tags may be not fully justified, but they build some kind of context for questions.
Tag should be such its tag badge is reasonable
Users active in certain tags will be automatically awarded with tag badges (bronze and silver). We can't do anything with it, so the "senseness" of such badges should be an argument for tag existence.
No tags for software-specific commands, functions, modules, packages
Maybe allow exceptions for a very few of most popular ones or wide-ranging such as
ggplot2, but otherwise (re-)tag questions with the name of the statistical software / application / programming language and a tag for the method rather than a tag for the name of the command/function/package/module implementing the method in that particular statistical software / language.
Automatically kill tags that are too narrow where "too narrow = one question".
Unless the narrow tag is supported by a high rep user.
Each retag needs a meta discussion
As question-closing thread during good old times.
Each tag should have its 'caretaker'
... that will know what the tag mean, protect it from blood-wanting moderators, write its wiki, suggest synonyms and make necessary retags.