Hi Statalist,
I am working with a dataset that has in the past primarily been analyzed
using SUDAAN, but now some researchers want to use Stata to analyze it so as
to have access to some of Stata's more powerful estimators and overall
flexibility. The problem is in setting up the complex survey settings.
For these purposes, our data was collected in a two-stage sample. In the
first stage, there was a stratified random sample of US Census Metro
Statistical Areas. Some MSAs were chosen with certainty, while others were
chosen from a stratified sample that aimed to ensure certain types of MSAs
were included. These were sampled without replacement.
For the second stage, our data was collected as a stratified random sample
with replacement at the household level from within the MSAs. There are
weights to correct to population statistics included at this level.
We understand how to apply the finite population correction and we have
weights that scale to the population, but we're not sure how to deal with the
unequal probabilities in the first stage; and in particular how to deal with
the certainty PSUs. I've read that strata with single observations can be
set to be treated as selected with certainty, but I'm unclear as to how that
actually works. An MSA selected with certainty will still have hundreds of
observations; is it the psu identifier that Stata uses to identify a
"certainly selected" psu; that is, if the psu identifier is the same for all
observations in a strata, then Stata will treat it as a certainty PSU?
Also, we previously used the jointprob function in SUDAAN to implement a
joint inclusion matrix; is there an equivalent in Stata? Or will the svyset
commands accomplish the same ends.
Thank you very much for your help,
-Ralph
*
* 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/
