st: Assigning pweight for -svy- data, based on data loss not survey design

[Nicholas Gebbia <nicholasjgebbia@gmail.com>]

Hey!

I'm working with pretest/posttest survey data evaluating an education
program, looking at changes in responses to Disagree/Neutral/Agree
questions, across paired pre and post surveys. I want to test for
significance in response changes (e.g. "For Q10, is the change in
Agree responses significant? ... Neutral responses? ... Disagree
responses?"), so naturally my inclination is to run -probit- and see
whether the dummy variable denoting pre/post survey turns out
significant. The problem is that I also want to account for the sample
population's differences from the entire population enrolled in the
program. I've already constructed a weight (I believe pweight?) to
account for gender-school representation (e.g. If 12% of all program
students are females at School B, and 7% are males from School D, then
my weighted sample reflects this). The normal -probit- and -logit-
commands don't take weights, so I've set up the data with -svyset-,
and running a -probit- appears to work correctly (the marginal effects
of "post" on agreeing to Q10, for example, equals the mean given by
-sum- for the change in Agree with weight applied). Especially as I'm
new to -svy- in Stata, I wanted to double-check that I've done the
-svyset- correctly, and that my standard errors are accurate.

The sample data is not sampled by design. Rather, all students in the
program are supposed to take the pre and post surveys; however, for
myriad data loss explanations, I only have matched pairs for about
half the population. The only weight I've yet constructed is the one
I've already referenced. My setup looked like this...

. svyset id [pw=weight]

where "id" simply denotes the ID of each student. Does this look okay?
Based on the marginal effects reported by

. svy: probit q10agree post

it looks like the weight has been applied as I intended (reasoning
mentioned above). Is there any way in which the weight I'm using is
not quite appropriate for the -svy- setup, especially insofar as it
might skew standard errors?

I've also given a preliminary check to the standard errors by
comparing to the standard errors reported by running the same -probit-
but sans -svy-, and there seems not to be any major difference (which
would be expected, as the weight does have an effect on responses, but
not a drastic one).

Any comments are much appreciated.
- Nick
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/