Jim -
I frequently analyze survey data and I can tell you that the answers to
your questions may depend on the way the weights are constructed. Are
the weights adjusted at every way? If so, what are they adjusted for?
You should be able to use the same strata and PSU info (Q1) and use all
available records for the analyses (Q2). As for Q3 though, we've done
this three different ways - used wave-specific weights for each case,
used weights from the last wave for each case, and created special
overall w1-3 weights. The difference depended on the type of analyses
we were doing, the way the weights were constructed, and the individual
opinions of participating colleagues. I'm not sure if there is one
correct answer to that question. If someone else collected the data and
constructed the weights, I would check with them first. If you don't
find any answer there, I would check to see what kind of adjustments are
made to the weights at each wave.
Sarah
Rosenthal, James A. wrote:
Dear Statalist member:
Here is a general question about the survey data analysis as implemented
in STATA.
I have been analyzing Wave 1 data from a stratified, cluster sample (8
strata, 100 PSUs within the various strata, 5000 individuals within the
various PSUs).
There are now 3 waves of data available.
In my new analyses with 3 waves, I want to do the following: 1) use,
essentially, the same SVYSET command (this because the strata and PSU's
are unchanged), 2) use all available records for analyses (some cases
have less than 3 waves of data, so in this situation I want to use all
waves that are available for a case) and 3) use the wave-specific
sampling weights for each case (the sampling weights change modestly
from wave to wave).
Is this an appropriate analysis strategy? I have limited experience in
survey analysis and want to be sure I am not doing something (major)
wrong.
One concern I have is that the manual for the data set suggests using
one weight (not 3 separate ones) for all waves of a given case (this
suggestion presumes, to my best knowledge, that only cases with all 3
waves of data are included in analyses). Would I be better off choosing
one constant weight? If so, which weight would that be (1st, 2nd or 3rd
wave)? I think I am better off not excluding cases with fewer than 3
waves of data. Any comments on this?
I am going to be doing, basically, regressions. Population-averaged
results will be satisfactory, though I am as much modeling the data as
trying to make inferences regarding (finite) population parameters.
Thanks for any help that you might be able to offer on this.
Jim
Jim Rosenthal
Professor and Graduate Coordinator
University of Oklahoma
School of Social Work
1005 Jenkins Avenue
Norman OK 73069
405-325-1401
fax: 405-325-7072
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
--
Sarah A. Mustillo, Ph.D
Department of Psychiatry and Behavioral Sciences
Duke University School of Medicine
Box 3454
Durham NC 27710
919 687-4686 x231
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
