7
$\begingroup$

There are ×654 and ×101 tags. Meta-regression is a particular approach to meta-analysis so it's a subset tag. 90 out of 101 Qs are also tagged with [meta-analysis], i.e. there are almost no questions tagged with [meta-regression] only. Given that, do we really need this tag, or should it perhaps be merged into [meta-analysis] (which will only increase the question count there by 11)?

On a related note, I have recently learned that there are fixed effects models, random effects models, and mixed effects models in meta-analysis (see brief but very clear Chapter 2 in https://www.jstatsoft.org/article/view/v036i03/v36i03.pdf). Obviously, people use our [random-effect-model], [fixed-effect-model], and [mixed-model] tags in combination with [meta-analysis], but in fact these models in meta analysis are clearly distinct from mixed models (e.g. you cannot fit them with lmer). Confusing! So I suggest to create tags like [meta-analysis-fixed-effects], [meta-analysis-random-effects], and [meta-analysis-mixed-effects] as synonyms of [meta-analysis].

@mdewey wrote me elsewhere that

I do not see any pressing need for meta-analysis tagged questions to also have a random effects tag (or fixed effects for that matter)

which I think is in line with what I suggest (?).

Finally, as far as I could understand "meta-regression" is another name for "mixed effects model in meta-analysis". Which might be another reason to make it a synonym. If not, then alternatively [meta-analysis-mixed-effects] could point to [meta-regression].

$\endgroup$
  • 1
    $\begingroup$ Since meta-regression is a proper subset of meta-analysis making it a synonym is fine. By the same logic f-e, r-e and m-e meta-analysis might as well be synonyms of meta-analysis too. Or we could just delete f-e, r-e when we see them used with meta-analysis.. $\endgroup$ – mdewey Jan 24 '18 at 13:35
  • $\begingroup$ @mdewey I'd rather have meta-analysis synonyms for f-e, r-e, and m-e, because otherwise people will keep using other tags for this purpose. $\endgroup$ – amoeba Jan 24 '18 at 13:55
  • $\begingroup$ @mdewey There is also a small [network-meta-analysis] tag with ~10 Qs but here I have no opinion at all and am fine with keeping it as is. $\endgroup$ – amoeba Jan 24 '18 at 14:05
  • 1
    $\begingroup$ Although n-m-a is a subset of m-a it is sufficiently a niche area that a separate tag does seem helpful. $\endgroup$ – mdewey Jan 24 '18 at 14:25
  • $\begingroup$ I would be inclined to leave [m-r] as is; someone could well want to search on that & read those specific threads as distinct from regular, old m-a. Ideally, we would get people to not feel they need to use both tags (let me know if you have a solution for that ;-). The fixed & random effects versions of m-a are more of an issue; people should not be using our standard mixed tags for those. Creating the new tags to be synonyms is a good solution here. This is a case where I'd prefer to have the tags on 1 thread & don't merge, though, so they would be more salient in the future. $\endgroup$ – gung - Reinstate Monica Jan 24 '18 at 14:59
  • $\begingroup$ @gung I have no problem with subset tags, as long as the smaller tag has a clear scope and the bigger tag has enough Qs to warrant the subsetting. So even if 100% of [m-r] Qs were tagged with [m-a], it would be fine with me. The question is rather, is "meta-regression" a clearly defined and useful subset? I have no experience in meta analysis, so I have no opinion. We have two users who are very active in these tags and I pinged both of them to share their opinions. $\endgroup$ – amoeba Jan 24 '18 at 20:45
  • $\begingroup$ @gung Regarding synonyms, I strongly dislike keeping a synonym on 1 thread. It's cumbersome, it's against the SE guidelines, it's gaming the system, and it's not a stable solution (one edit to this thread and the full merge will happen). Also, I don't see a benefit. Ideally, this will be a part of my old plan on reorganizing fixed-effect and random-effect tags. My current thinking is to have meta-analysis-random-effect -> meta-analysis, mixed-model-random-effect -> mixed-model, econometrics-random-effect (on its own), and anova-random-effect -> anova. But let's deal with meta-analysis first... $\endgroup$ – amoeba Jan 24 '18 at 20:50
  • 1
    $\begingroup$ Chiming in here based on the request by @amoeba I think the meta-regression tag is useful. I skimmed through the questions that are tagged with m-r and see some cases where it is used when it doesn't seem relevant/applicable. As for the FE/RE/ME stuff - this one is tricky. In principle, the terms could have the same meaning in the context of meta-analysis (and imho they should) but they have taken on bizarre interpretations/definitions in the meta-analytic context that are really confusing. So not really sure what to do about these tags when they are used for meta-analysis questions. $\endgroup$ – Wolfgang Jan 24 '18 at 23:35
  • $\begingroup$ @Wolfgang Thanks for joining the discussion. Okay, we can keep [meta-regression]. Regarding FE/RE/ME, could you clarify what bizarre interpretations you mean? My understanding (based on your paper linked in my question here) was that all these models in meta-analytic context assume known (and heteroscedastic) error variances, unlike what happens in basically any other regression context. I guess this is not what you mean by "bizarre"? $\endgroup$ – amoeba Jan 25 '18 at 8:36
  • $\begingroup$ Wolfgand and @mdewey, I tried to edit the [meta-regression] wiki excerpt, can you please make sure that it makes sense and is useful? stats.stackexchange.com/posts/87546/revisions We could potentially cut out the 2nd sentence if you think it will be only confusing for the readers. $\endgroup$ – amoeba Jan 25 '18 at 9:24
  • $\begingroup$ When I looked we have 39 with m-a + r-e-m of which 10 also have f-e-m, there are 19 with m-m of which 4 have r-e-m and 1 f-e-m. f-e on its own is rare. Interestingly m-m occurs quite often with meta-reg which is adding nothing. @Wolfgang may have more to say here but none of this surprised me. $\endgroup$ – mdewey Jan 25 '18 at 13:38
  • $\begingroup$ The meta-reg excerpt looks fine but the wiki itself does not ring true to me. I will have a think unless someone gets there first. $\endgroup$ – mdewey Jan 25 '18 at 13:40
  • 1
    $\begingroup$ @mdewey My current inclination based on the discussion above is to remove all f-e-m, r-e-m, and m-m from Qs with m-a (around 50 threads involved probably?) and then create three synonyms of m-a: meta-analysis-fixed-effects, meta-analysis-random-effects, meta-analysis-mixed-effects. Meta-regression can be left alone based on Wolfgang's and gung's comments (btw, by now 100% of Qs with m-r are also tagged m-a). $\endgroup$ – amoeba Jan 26 '18 at 8:32
  • $\begingroup$ @mdewey (Of course not all instances of m-m need to be removed. If a question asks about "standard" mixed model vs meta-analytic mixed model then it's obviously fine to keep m-m. I would remove all f-e-m and r-e-m though because these two tags are ambiguous). It seems that there is no opposition here. Would you be willing to go through these threads? $\endgroup$ – amoeba Jan 30 '18 at 16:20
  • 1
    $\begingroup$ I will have a look at them and delete all the obvious ones with f|r-effect or m-m and come back with any puzzles. $\endgroup$ – mdewey Jan 30 '18 at 17:04
7
$\begingroup$

Based on the discussion in comments above, my current suggestions are as follows:

  1. stays as a distinct tag. Currently all questions with [meta-regression] are also tagged with , but it's still a useful subset tag to have.

  2. We remove all instances of and from [meta-analysis] threads. We also remove most instances of , leaving only the ones that are really about "standard" mixed models and meta-analysis.

  3. We create the following three tag synonyms:

Update: (2) done by @mdewey, (3) done by @gung.

$\endgroup$

You must log in to answer this question.

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