Interesting discussion. It raises another question for me, which is, if one
wants to use weights for post-stratification purposes only in a survey that
is otherwise conducted using SRS (e.g., to make, say, basic demographic
factors like age and sex match known population values in order to try to
get unbiased prevalence estimates), does one need to use the -svy-
functions? I thought that I had read somewhere that if weights were used in
this way, then one could think of them as aweights rather than pweights to
avoid having to use robust standard errors? But based on the discussion
below, maybe that is incorrect?
If I am reading the Stata manuals right (specifically, p. 346 of the Stata
8 User's guide, section 30.2) post-sampling adjustments should be handled
via pweights rather than aweights, and pweights will use robust
errors. Further, Stata's description of aweights does not sound consistent
to me with what you are describing. (However, given that I've never
actually used the svy commands, you may not want to accept me as the
ultimate authority on this.)
