5
$\begingroup$

I want to discuss the following three tags: (2000+), (580) and (400).

There is a well-known terminological ambiguity between econometrics and other statistics with regards to the "random effects" and "fixed effects". I feel that our tags add to this ambiguity, instead of clearing it up.

  • Mixed models are linear models that include random effects and fixed effects, i.e. random factors and fixed factors (or, rarely, only random factors). The terms "random/fixed effects model" are never (?) used to refer to mixed models. Update: Mixed models are also sometimes called "random effects models".

  • In econometrics, a "fixed effects model" is a way to treat panel data which is equivalent to a standard linear model with only fixed effects. In contrast, a "random effects model" is a way to treat panel data which is formally equivalent to a mixed model with a random intercept but which is estimated via FGLS and not via MLE (as mixed models are). (Evidence: @StasK's 2012 Meta answer and my recent question on the main site.)

So conceptually [fixed-effects-model] tag should only be applied to questions about econometrics; in practice people often use it on questions about mixed models to refer to fixed effects. This is confusing. And [random-effects-model] tag can be, and is being used to refer to both, mixed models and random effects models in the econometrics sense. So the mixed models questions get split between these two tags, which is very unfortunate.

What could be the ideal arrangement?

I'd say: rename these two tags such that they unambiguously refer to econometrics. For example: [panel-random-effects] and [panel-fixed-effects] (the word "econometrics" is too long to include in a tag name). And sort out the mixed models questions into [mixed-model].

$\endgroup$
  • 1
    $\begingroup$ I think [random-effects-model] is fine. Random-effects model can be used to refer to mixed models, see Laird and Ware (1982) "Random-effects models for longitudinal data". $\endgroup$ – Randel Oct 12 '16 at 1:48
  • 3
    $\begingroup$ I think the ambiguity in these tags on this site reflects real terminological ambiguity in the statistical literature, broadly defined. My impression is that outside of econometrics, many people (including often me) really do use the terms "mixed model" and "random-effects model" interchangeably. So I'm not sure I agree that "[random-effects-model] and [fixed-effects-model] tags should only be applied to questions about econometrics." I do however agree that the phrase "fixed-effects model" is used primarily in economics only. $\endgroup$ – Jake Westfall Oct 12 '16 at 2:16
  • $\begingroup$ Thanks @JakeWestfall and Jiebiao. So it seems that I was mistaken and in fact "random effects model" is used in literature and in real life to refer to "mixed models". But do you guys think that "random effects model" in mixed models (MLE) and "random effects model" in econometrics (GLS) are two sufficiently different concepts that it would be beneficial to separate the tags? Or do you rather think it's good that it is the same tag? I find it confusing because the first one is basically the synonym of [mixed-model] but the second one is not really. $\endgroup$ – amoeba says Reinstate Monica Oct 12 '16 at 9:28
  • $\begingroup$ I would say, the models are the same (except that one usually has only random intercept), but the main difference is just in the estimation. In other words, the two you refer to are indeed "random-effects model". The difference can be found from other tags like [panel-data] used together in the question and the context of the question. $\endgroup$ – Randel Oct 12 '16 at 14:07
  • $\begingroup$ I can see your point @JiebiaoWang. My biggest problem is that if not for econometrics, I would argue that [random-effects-model] should be a synonym of [mixed-model] (as both you and Jake implied above). I am however hesitant to suggest this synonym now, because it could be very confusing for all the econometrics people. What do you think? The current situation is unsatisfactory because navigating [random-effects-model] does not make a lot of sense: huge amount of [mixed-model] questions do not carry this tag. $\endgroup$ – amoeba says Reinstate Monica Oct 12 '16 at 14:26
  • $\begingroup$ I agree that we do not suggest [random-effects-model] as a synonym of [mixed-model]. People have their own preference for the terminology. That's why I set both [random-effects-model] and [mixed-model] as my favorite tags. :) But I am also curious how popular random-effects model is in econometrics, since people would worry about the correlation between X and the random intercept. $\endgroup$ – Randel Oct 12 '16 at 16:04
7
$\begingroup$

My question did not provoke much interest but I am still deeply annoyed by the current situation.

Here is what seems to me would be a better arrangement:


  1. For questions about "random effects model" in econometrics. This means a model with random intercepts which is estimated with FGLS.


  2. For questions about "fixed effects model" in econometrics. This actually means a simple linear model, but usually it is of a particular form and is estimated in a particular way.

  3. $\to$
    $\to$
    For questions about random effects and fixed effects in mixed models. We don't need separate tags for that, so these should be synonyms of the existing [mixed-model] tag.

With this arrangement, anybody typing "random-e..." in the tag field will get two suggestions and will be able to choose an appropriate one. The current tags and should be sorted out and eliminated. I don't see any disadvantage of my suggested arrangement, but I see multiple advantages (e.g. all questions about mixed models will be under [mixed-model] instead of being divided between [mixed-model] and [random-effects-model]).

@StasK wrote back in 2012 that

May be there should be random-effects-model-econometrics and fixed-effects-model-econometrics subtags

which is in line with what what I am suggesting here. The tag names have to be shorter though, due to the 25 chars limit.

This would require a substantial retagging effort but I would volunteer to do it. I looked the last 50 questions in both tags, and here is how they are split between topics:

                             econ      MM      both     neither      sum
 fixed-effects-model:         37       8       -        5            50
 random-effects-model:        15       35      -        -            50

Based on that, I would take the mixed-model Qs out of fixed-effects-model, and the econometrics Qs out of random-effects model, and then we could do tag renames for the rest.

There are 1001 threads tagged with either of these tags. The fraction to be manually retagged is ~15/50. So it's 1000*15/50 = 300 threads. With 10/day it will take 1 month.

$\endgroup$
  • 2
    $\begingroup$ This seems like a reasonable solution to me. $\endgroup$ – Jake Westfall May 24 '17 at 14:40
  • $\begingroup$ I agree that this is an annoying situation. I'm fine with your proposed solution, but here's a couple of additional possibilities: 1) instead of "random-eff-econometrics", make it "random-eff-fgls" to focus on the method instead of the field, 2) Make "fixed-eff-econometrics" or even the existing "fixed-effects-model" a synonym of "linear-model" (or whatever) since those are essentially the same. $\endgroup$ – mkt - Reinstate Monica Feb 5 at 15:24
  • 1
    $\begingroup$ @mkt, I disagree. The only people who would be asking questions about these are economists or others specifically using an econometric approach. They will have a particular thing in mind & we want to signal that to both askers selecting tags & answerers. I think having econometrics is important. I also think these excerpts are very good. I upvoted this long ago. Was there any reason we didn't pull the trigger then, amoeba? $\endgroup$ – gung - Reinstate Monica♦ Feb 5 at 15:38
  • $\begingroup$ @gung Not really -- it just looked like a lot of work, so I was reluctant to go ahead; and in addition I have decreased my participation since then... That said, this was my pet project so I guess I can look into it again. $\endgroup$ – amoeba says Reinstate Monica Feb 5 at 16:19
  • $\begingroup$ @mkt Would you be interested in joining the retagging efforts if we decide to go ahead? $\endgroup$ – amoeba says Reinstate Monica Feb 5 at 16:20
  • $\begingroup$ BTW, I agree strongly with what gung wrote in the comment. $\endgroup$ – amoeba says Reinstate Monica Feb 5 at 16:20
  • $\begingroup$ @gung Fair enough $\endgroup$ – mkt - Reinstate Monica Feb 5 at 17:59
  • $\begingroup$ @amoeba I could help out, but can't really commit to a regular 10/day. $\endgroup$ – mkt - Reinstate Monica Feb 5 at 18:00
  • 2
    $\begingroup$ Note: I made the synonyms in #3. $\endgroup$ – gung - Reinstate Monica♦ Feb 5 at 20:11

You must log in to answer this question.

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