There are four correlation coefficients (cc) which currently lack tags. The point-biserial cc is that between a continuous variable and a binary variable. It is just the Pearson cc between them. There is also the less used rank-biserial cc which is between a ranked variable and a binary. By contrast the biserial cc assumes that the binary variable is the realisation of a latent normal and is an estimate of the correlation of the underlying bivariate normal. The polyserial cc generalises the biserial with an ordered categorical replacing the binary. We currently have 9 question with the word polyserial in them and 39 with the word biserial, most of which seem to be about the point-biserial. Although this is something of a niche area it would seem appropriate to have tags here. I think there are three main options:

  1. Create four tags, one for each cc. This strikes me as overkill
  2. Create a tag [biserial] and a tag [polyserial] and tag all the variants of biserial with [biserial]. This seems to me a good option as it does not rely on questionners knowing exactly which biserial they want
  3. Create [biserial] and tag all four cc with it. This does not seem quite right to me as the polyserial is not a special case of biserial but rather the other way around

If this meets with general approval I will re-tag them and write the appropriate wiki(s).

  • $\begingroup$ @AndreSilva I prefer #2 but if the consensus is for #1, #3 or do nothing then I will do that. $\endgroup$ – mdewey Apr 12 '17 at 14:49
  • $\begingroup$ If to create any tags at all (I don't currently have an opinion about that), I would opt for a joint tag for all of that. And I would prefer the tag name to have "correlation" in it. So smth like "biserial-polyserial-correlation" but shorter :) Also, one might consider putting tetrachoric/polychoric correlations into the same tag, see personality-project.org/r/html/tetrachor.html. I don't know if there is an umbrella term for all that. $\endgroup$ – amoeba says Reinstate Monica Apr 12 '17 at 16:10

My initial hunch is that we may be OK with the tags we already have. I would think we would do best to tag the threads you describe with something like one of the following:

  1. and ,         rather than [biserial] and [correlation]
  2. and ,               rather than [polyserial] and [correlation]
  3. and ,   etc.

I wouldn't want to have special tags for these (e.g., [point-biserial-correlation]), because I think it would be best to still have such treads categorized under the existing [correlation] tag, and using both of those would strike me as weird and redundant. That does leave the option of just, say, [biserial] and [correlation], but if we are going to add two tags to a thread, we already have tags that strike me as natural. Creating new tags that will sometimes be used and sometimes the existing tags will be used seems to me likely to lead to the information on the site being less well organized rather than better.

  • 1
    $\begingroup$ Since there have been no net upvotes for either the question or this answer I think the fourth option, do nothing, is the consensus so I am accepting this answer to close the thread. $\endgroup$ – mdewey Apr 19 '17 at 14:44
  • $\begingroup$ We could always revisit the idea in the future, @mdewey. There doesn't seem to have been much action on this at all. $\endgroup$ – gung - Reinstate Monica Apr 19 '17 at 15:24

You must log in to answer this question.

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