7
$\begingroup$

https://stats.stackexchange.com/tags/weighted-sampling/info

There is no term "weighted sampling" in survey statistics world. Not at least that I know of. It is like "use Excel to prepare a JASA paper". I guess people meant "unequal probability sampling" when they tag their questions with this dubious phrase, and then it just propagates like "Free Wi-Fi" network in the airports (if you don't know this urban legend: somebody just created an ad-hoc network called "Free Wi-Fi", a bunch of stupid computers connected to non-discriminantly, and so after a little while this network used to be found in every US airport for a few years, even though there was no Internet access or anything; sort of performance-degrading virus).

Anyway, I hate this tag, and would like to kill it. Any comments on this wild desire? My understanding is that I would have to retag all of these questions though. A more generic sampling would be a better fit for most of them (and I believe I worked on that tag earlier).

$\endgroup$
  • 4
    $\begingroup$ Moderators can create a tag synonym and merge weighted-sampling with unequal-probability-sampling (master tag) so that every new suggested tag will appear as unequal-probability-sampling. Is that ok with you? There are several tags about 'sampling' but probably some questions need to be tagged as unequal-probability-sampling and we could keep the most generic one (although it is very broad and sometimes misused, IMHO). $\endgroup$ – chl Sep 26 '12 at 7:39
  • $\begingroup$ @chl I like your solution here. :) $\endgroup$ – Aarthi Sep 26 '12 at 16:09
  • $\begingroup$ I guess I just need to boost my own rating and get to the moderator level myself, so that I would not have to bother asking on Meta as to what to do on CV :) $\endgroup$ – StasK Sep 26 '12 at 21:12
2
$\begingroup$

I must disclose that I don't know anything about surveys, but in other areas "weighted sampling" is quite a common and well established term, to the point that there are more concurrent meanings in need for disambiguation. Think e.g. about sampling with a measure change (of which importance sampling can be seen as a special case). Or WLS in a simulation setting. Also "Weighted Monte Carlo" has atleast two different well defined meanings in quantitative finance, both of which are sometimes confusingly traded for "weighted sampling". Maybe by "unequal probability sampling" you meant just a measure change, but for many people (atleast all of us into simulation) such term rings no bell at all and is worse than obscure w.r.t. referencing (despite its obvious interpretation).

I can understand your concern though, and we should discuss it more extensively, what I suggest is just not to generalize to the whole forum a fix fit for a few areas only.

I cannot offer a solution right away, but certainly removing the tag altogether would cause additional problems, worse than the single relatively specific/limited issue solved. Those "misbehaving" posters should be advised to stay away from improper tagging instead, and on the other hand a disambiguation guideline would also help a lot.

For the proper uses I am even thinking about subtagging ("xxx weighted sampling" vs "yyy weighted sampling") where reasonable disambiguation synonyms cannot be found, so far from killing the term.

$\endgroup$
  • 2
    $\begingroup$ I see. Thanks for the input. I know at best a tiny bit about these areas (although of course I've spend many petaflops with my own simulation of various statistical estimators). I can see where it is misapplied in survey statistics applications. So probably a disambiguating wiki summary referring to better tags would be the best solution. $\endgroup$ – StasK Sep 28 '12 at 14:47
  • 1
    $\begingroup$ I also am not expert in all of these areas, but I can certainly say, as a matter of general CV philosophy, that I think good tag wiki excerpts are essential. If it turns out to be the case that this is a tag that can be correct in some settings, but may be misapplied in others, getting the tag wiki excerpt right is the ideal solution. $\endgroup$ – gung - Reinstate Monica Sep 28 '12 at 16:22
  • $\begingroup$ Relying on tag synonyms should help to solve those problems, IMO: unequal-probability-sampling will be a master tag, but if people think of 'weighted sampling' they will be suggested to look at this tag first. If the wiki description doesn't fit the bill, they can choose another tag or create a new one if they have enough reputation. $\endgroup$ – chl Sep 29 '12 at 10:32

You must log in to answer this question.

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