5
$\begingroup$

I just looked at the wiki for regression and it only discusses linear regression (OLS). Should it include a list of, links to, or even text about other forms?

$\endgroup$
  • $\begingroup$ We have a tag for nonlinear-regression already, are you talking about a pointer to that, or something else? What "other forms" do you have in mind? $\endgroup$ – gung Sep 30 '12 at 3:30
  • $\begingroup$ I was talking about logistic regression, ordinal logistic, multinonial, Poisson and so on. $\endgroup$ – Peter Flom Sep 30 '12 at 11:40
  • $\begingroup$ So it sounds like you're mostly talking about the generalized linear model. We do have a tag for that too, of course (as you well know). So do you just want to link to that tag wiki? Also, are you talking about amending regression's tag wiki excerpt, full wiki, or both? $\endgroup$ – gung Sep 30 '12 at 13:29
  • 1
    $\begingroup$ I am not really sure - I am getting used to how the tags work (if you have a link to where this is discussed, I'd appreciate it). But when I think of "regression" I think of the whole panoply of regression methods; people also seem to be using the tag for regressions that are not OLS. $\endgroup$ – Peter Flom Sep 30 '12 at 13:35
  • 1
    $\begingroup$ I don't really know of a good link for tag wikis. FWIW, the best I can think of are the privilege descriptions: approve-tag-wiki-edits, & trusted-user. My general way of thinking about the issue is that tag wiki excerpts are really important for the smooth functioning / organization of the site as these are displayed w/ the tags (NB not all of the characters you are allowed when editing will show up, though), but that probably few people ever read the full wiki. $\endgroup$ – gung Sep 30 '12 at 15:42
  • $\begingroup$ hmmm. OK. Those are helpful. $\endgroup$ – Peter Flom Sep 30 '12 at 17:52
7
$\begingroup$

This is what the current tag wiki excerpt says:

Techniques for analyzing the relationship between one (or more) "dependent" variables and "independent" variables.

I would be fine with amending this to something like:

Refers to techniques for analyzing the relationship between one (or more) independent and dependent variables using ordinary least squares (OLS). "Regression" is sometimes used loosely to include other types of models (logistic regression, Poisson reg., etc). For these, please consider: generalized-linear-model, logistic, probit, nonlinear-regression, etc, instead.

For the full wiki, I'm less concerned, as I think they are rarely read. (As a case in point, I gather you have only just read it, even though you've been a member for >2 years, and with >10k rep*.) Of course, that doesn't mean we should feel free to have just anything there, but I see no reason why some additional paragraphs couldn't be added (after what currently exists) stating that people sometimes include things like logistic regression under the the rubric of regression and linking to Wikipedia or other tag wikis. For what it's worth, I also tend to think of logistic regression as falling under some general heading of 'regression', which is ironic, since OLS regression is a special case of the GLiM instead. That is probably a misconception worth clarifying in the full wiki.

*This is in no way a criticism.

$\endgroup$
  • $\begingroup$ All this would be fine with me. I didn't read this wiki, probably because I already know what regression is :-). $\endgroup$ – Peter Flom Sep 30 '12 at 22:30
  • $\begingroup$ That's what I figured. I didn't read it for a long time either. $\endgroup$ – gung Sep 30 '12 at 22:32
  • 2
    $\begingroup$ This definition seems to exclude the machine learning interpretation of "regression"...basically a predictive model where the predictions are numbers rather than discrete classes. $\endgroup$ – Michael McGowan Oct 1 '12 at 21:46
  • $\begingroup$ That's an interesting point, @MichaelMcGowan. I notice however that definition of 'regression' is an ex of doug's thesis that "disciplines use the same term to refer to completely different concepts". Thus, to include it as well under the reg tag would lead to a tag that is conceptually incoherent. Perhaps we should add another tag: ML-regression, IDK. But I don't think we should try to resolve this here in these comments. If you like, you could ask about what to do re this issue in a new meta question. $\endgroup$ – gung Oct 1 '12 at 21:59

You must log in to answer this question.

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