StataCorp has shut one door on this, but another is wide open. You can
use -glm, link(logit)- with aweights.
As for the general claim, here is one situation in which aweights seem
at first sight quite right for binary responses. In geography we can
have two kinds of land cover or land use (land/sea, urban/rural,
whatever) and we know or measure the fractions of each in various areas.
For example, 71% of the Earth is sea and 29% is land. There is not a
known or even natural integer for the number of observations each of
those fractions is based on.
Nick
[email protected]
Yulia Marchenko, StataCorp LP
"L S" <[email protected]> asked about using analytic weights with logit
regression:
> I am trying to run a weighted logit regression in STATA.
> It is easy to notice that weights are not allowed by the basic
> logit command (STATA gives the error 'aweight not allowed'). I have
> been told about the possibility of referring to a past version of
> STATA which supports aweights:
>
> version 9.0
> logit y x [aw=w]
Analytic weights, -aweights-, are used to represent the cell-means data
and,
thus, do not have meaningful interpretation with binary-response data.
As
Maarten mentioned, only -fweights-, -iweights-, and -pweigths- may be
used
with -logit-. Although never documented, -aweights- were allowed with
-logit-
prior to Stata's version 9.1, at which time they were removed because of
their
noninterpretability.
-- Yulia
[email protected]
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
