4
$\begingroup$

I've created tag . I mean a mixture of variables of different types or scales.

We long have tags [continuous-data], [categorical-data], [binary-data], [count-data], [ordinal-data]. All of them are necessary and should exist, to me. (We also have tag [nominal] and I wonder shouldn't it be kept, dismissed or merged to [categorical-data].)

I think that we do need tag [mixed-data]. In many analyses, mostly multivariate exploratory methods (cluster, factor, irt,...), but also in other as well, mixing variables of different types in one set represent a special problem to solve.

I want therefore to ask you some questions:

  1. Do we need that tag?
  2. Shall it be worded [mixed-data] or [mixed-type-data]?
  3. Shall we narrow the definition - in the excerpt - so that we don't include the (very usual) case of using mixed type predictors (continuous+categorical) in regression-like modeling in the concept of "mixed data", or include that case in the concept too - along with using mixed data in clustering, PCA, etc.?
  4. Other thoughts from you, please welcome.
$\endgroup$
  • 1
    $\begingroup$ Why not let proposers decide? That's how tags are generated on SE in the first place. When somebody has a question that they feel truly needs a "mixed-data" tag then they can create it for that question. $\endgroup$ – whuber Mar 2 '18 at 17:01
  • 1
    $\begingroup$ @whuber, in this case ttnphns is the proposer, but he is opening his proposal for wider discussion fist (which is admirable IMO). $\endgroup$ – gung - Reinstate Monica Mar 2 '18 at 17:21
  • 1
    $\begingroup$ @gung By "proposer" I mean proposer of a question. If ttnphns is asking a question that he believes requires a "mixed-data" tag, then by all means he should apply it. That will create the tag. There's no need to ask the community to do that! $\endgroup$ – whuber Mar 2 '18 at 17:38
  • 2
    $\begingroup$ @whuber, I don't have a problem w/ someone proposing a useful tag & putting it on existing questions to group them. I've done that twice (glmm, & var), & both became productive tags. People asking questions might search for a tag, not find one, & not know--or have the required rep--to create their own. Someone more familiar w/ the site, like ttnphns, may be better able to see where a new tag is needed. $\endgroup$ – gung - Reinstate Monica Mar 2 '18 at 17:48
  • 7
    $\begingroup$ Regarding We also have tag [nominal] and I wonder shouldn't it be kept, dismissed or merged to [categorical-data]: Yes, I think [nominal] should be made a synonym of [categorical-data]! That was @gung's suggestion some time ago and we discussed it in a separate Meta post but you were strongly against this mapping so it was never implemented. Did you have second thoughts since then? I don't know about gung but I still think it would be a sensible tag synonym mapping. $\endgroup$ – amoeba says Reinstate Monica Mar 2 '18 at 22:44
  • 1
    $\begingroup$ @amoeba I would be for merging those tags. We would be doing site-users a favor by stressing these are identical terms. $\endgroup$ – AdamO Mar 14 '18 at 13:50
  • 2
    $\begingroup$ I sense the risk of vagueness is very high. At first blush, when I read "mixed-data" I think 1) data for mixed models (which can be of any sort), then 2) mixture data. Neither of these fit with what you're suggesting here. $\endgroup$ – AdamO Mar 14 '18 at 13:53
  • 1
    $\begingroup$ @AdamO, I probably will change the tag mixed-data to mixed-type-data (as I mentioned it) soon. Then I'll select and tag questions. I'll do it, please wait a bit. $\endgroup$ – ttnphns Mar 14 '18 at 17:08
  • 1
    $\begingroup$ Would [mixed-data] refer to variables with different measures, or to mixed continuous-discrete distributions, or to? $\endgroup$ – Alexis Apr 13 '18 at 17:57
  • $\begingroup$ @Alexis, no. mixed-type-data (that will be the tag) will refer only to mixture of variates (data columns) of different measures. $\endgroup$ – ttnphns Apr 13 '18 at 18:01
  • $\begingroup$ The [mixed-data] tag is already being used incorrectly (for mixed models). We really should rename the tag and provide a wiki excerpt to prevent proliferating the confusion. Otherwise I would simply delete the [mixed-data] tag for the time being. $\endgroup$ – amoeba says Reinstate Monica Apr 19 '18 at 9:57
  • $\begingroup$ @amoeba, we may delete [mixed-data]. The tag I'll organize will be [mixed-type-data] $\endgroup$ – ttnphns Apr 19 '18 at 12:02
  • $\begingroup$ OK, I've just deleted [mixed-data] then. By the way, what do you think about nominal/categorical-data -- see my comment above (this one). $\endgroup$ – amoeba says Reinstate Monica Apr 19 '18 at 12:17
  • 2
    $\begingroup$ @amoeba, thinking back, I perceive we shouldn't be excessively scholastic. Yes, I think you may merge nominal to [categorical-data]. $\endgroup$ – ttnphns Apr 19 '18 at 14:49
3
$\begingroup$
  1. Maybe, I think it depends on whether it would successfully group enough threads to usefully contribute to the organization of the material on the site (see below).
  2. We have several "mixed-" tags already, which naturally creates the potential for confusion. My preference would be for the name to be as distinct as possible, and the excerpt to delineate its usage clearly and up-front. My suggestion for the best name might be [mixed-data-types], although something else might be better.
  3. If the tag is circumscribed to only apply to questions about mixed data types in clustering and PCA, e.g., I think it would be too narrow to prove useful over time (see #1, above). If it would be applicable to questions about combining continuous and categorical data within regression models, e.g. (and you think those types of questions will emerge), then I think it may end up proving useful.
  4. (Hmmm, I'm a man of few thoughts, I'm afraid. I would that it were otherwise...)
$\endgroup$
  • 1
    $\begingroup$ I am also worried about your point #2. By the way, I noticed that we have [mixed] as a synonym of [mixed-model]. I think it would make sense to delete this synonym mapping (i.e. delete [mixed] altogether), given that we have several [mixed-something] tags, and might get even more. $\endgroup$ – amoeba says Reinstate Monica Mar 2 '18 at 22:30
  • 2
    $\begingroup$ @amoeba, I deleted [mixed]. It should disappear in a day or so. $\endgroup$ – gung - Reinstate Monica Mar 3 '18 at 1:05
  • 1
    $\begingroup$ FYI [mixed-data] started to be misused and I deleted the tag. Ttnphns says he is going to create [mixed-type-data] some time soon. If you don't like this name then perhaps comment under the Q. $\endgroup$ – amoeba says Reinstate Monica Apr 19 '18 at 12:25
  • $\begingroup$ @gung, shouldn't we rename tag [mixed-model] into [mixed-effect-model]? We have have tags [fixed-effect-model], [random-effect-model]. [mixed-model] sounds really ambiguous. $\endgroup$ – ttnphns Apr 20 '18 at 15:27
  • $\begingroup$ Maybe, @ttnphns, but that should be discussed somewhere on meta.CV other than here in these comments. My impression is that just saying 'mixed-model' is very common. $\endgroup$ – gung - Reinstate Monica Apr 20 '18 at 15:38
  • $\begingroup$ @gung, it is common. But people not inqrequently confuse it with 1) [mixed-type-data], 2) "gaussian/finite/etc mixture" or 3) distrubutions like Tweedie. See also AdamO's comments above. $\endgroup$ – ttnphns Apr 20 '18 at 15:44
  • $\begingroup$ @amoeba, will you have an opinion, too? (I mean last 3 comments on {mixed-model] here $\endgroup$ – ttnphns Apr 20 '18 at 15:45
  • $\begingroup$ @ttnphns I prefer the tag to be called [mixed-model]. This is IMHO by far the most common term. Actually, my long-standing plan is to eliminate [fixed-effect-model] and [random-effect-model] entirely because I think those are ambiguous, but this is such a big undertaking that I never started implementing it (see stats.meta.stackexchange.com/questions/4391). $\endgroup$ – amoeba says Reinstate Monica Apr 20 '18 at 15:48
  • $\begingroup$ @amoeba, it's a problem intrinsic to the inconsistent terminology that exists. ttnphns has a point about the potential for confusion w/ FMMs. $\endgroup$ – gung - Reinstate Monica Apr 20 '18 at 16:49

You must log in to answer this question.

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