Should most R tags be replaced with package tags?

Question

The R tag covers an enormous number of questions, many of which aren't really about R in general, but how to use a specific package. There appear to have already been some grumblings about this in meta already:

Syntax highlighting with r tag

Should tags refer to the answer as well as the question?(see comment)

As I see it there are broadly three types of R question

Type A

Some questions that validly fit into a general R tag:

Difference between logit and probit models (questioner request an R answer)

Free resources for learning R

Type B

But other questions could be cut down to a more specific tag (as has been suggested at times, see meta examples above), for example:

Interpretation of R's lm() output (could be tagged lm)

Cost function for validating Poisson regression models(Is also tagged cv.glm)

Type C

While others use R heavily in the question example, but don't actually need R

Removal of statistically significant intercept term increases $R^2$ in linear model

How to interpret type I, type II, and type III ANOVA and MANOVA?

Which questions should be tagged as R?

Answer 1

A risk of adding package tags would be 1) overlapping nomenclature 2) frequent disagreement on tags and 3) difficulty consulting relevant questions from other tagged questions with answers. For instance, a user working with the survey package would probably often mistakenly use the survey tag instead of the survey-package tag. If a user tags their question with ggplot2 but the source of their problem was a data formatting issue, then should the ggplot2 tag be removed, even if the subtlety in data formatting was unique to that package? It could also greatly inflate the number of tags, leading to an overly granular approach to browsing questions (such as packages with overlapping functionality like lme4 and nlme or gee, geese, and sandwich for computing robust standard errors).

I think R packages come and go and there are way too many of them. I also think many expert statisticians can quickly look up documentation on a few package specific questions and quickly address the users questions without even downloading that package.

I think questions of type C which are methodological in nature but rely upon R to convey the idea and are tagged as such are in serious need of quality improvement. It seems that currently, expert users migrate these posts to stackoverflow where they often languish and die. I think experts should be more assertive in editing such posts to remove code/unformatted output that can be explained in concepts or algebra.

Answer 2

I don't think type C questions should be tagged R. Surely R tags should be for questions that require knowledge of R to answer - R is not a hard syntax, so for the examples above many non R users would understand and be able to answer the question. I don't think syntax highlighting is important enough, and anyway most people put the output in as well, which always highlights wrong (e.g. the "Interpretation of R's lm() output" link above).

Type B questions I'm more unsure of. On the one hand, they create an enormous number of R tags, but on the other hand they do need some R expertise. If there are any answerers who follow the R feed because they have lots of R experience, then it is important these are tagged as R. On balance I would say they should be tagged R.

Answer 3

If this is a strictly R programming/implementation question, it should go to SO. If this is a true statistics question, then an R package is just an implementation of the method, and one can often find alternative implementations in Stata, SAS, Matlab, or even (God forbid) Excel. So instead of cv.glm tag in the second type B question, it should have been the methods cross-validation tag. Any of the *lme* questions are actually mixed-model questions, etc.

On a number of occasions, I wrote in the tag excerpt something like "this tag is too generic, please use this for this, and that for that" (see, e.g., multilevel and gmm tags... I think the latter was actually killed, which isn't right, and should still be sitting around as a redirect). May be it is time to do something like this with r: officially declare R to be the default statistical package to ask questions about and give examples in, discourage the use of the r tag, and tell the users to use a tag like learning-r for the second question in A and the first question in B. Some other tags may need to be devised for "valid" R questions that CV rather than SO should handle, e.g., compare-stat-packages for tag combos like https://stats.stackexchange.com/questions/tagged/r+sas or https://stats.stackexchange.com/questions/tagged/r+stata.