st: AW: population weights question

[Reinhardt Jan Dietrich <jan.reinhardt@paranet.ch>]

Do you want to compare your clinical sample with the general population or with your national sample? Do you want to compare the effect of gender on some outcome between the two groups, the prevalence of some trait or something else?
All this is important to know in order to decide if the weights for the national sample should be used or not in the comparative analysis (anyway, the weights for your clinical sample would always be 1).
Jan



________________________________________
Von: owner-statalist@hsphsun2.harvard.edu [owner-statalist@hsphsun2.harvard.edu]" im Auftrag von "Kyleigh Schraeder [kyschraeder@gmail.com]
Gesendet: Dienstag, 21. Februar 2012 18:28
An: statalist@hsphsun2.harvard.edu
Betreff: st: population weights question

Hi statalisters,

Steven helped me out with a similar question the other day but I would
like to rephrase.

I would like to compare a national health survey dataset to an
independently collected clinic dataset.  The clinic dataset was not
weighted.  I am having some difficulty understanding how to employ the
complex survey variables that were employed for the national dataset.
The design variables for the national dataset are:
xwgtreg (sampling weight variable)
strata (strata variable, 19 strata)
schlid (primary sampling unit/cluster - school id number)

Although national health survey data was collected from a
representative sample of grade 7-12 students (N exceeds 7,000
students), I only want to look at students between the ages 15-19 from
one stratum (stratum 19).  I am also interested in gender differences
between the national dataset and clinic dataset.  I am wondering how
to (and if I need to) -stset- the national health survey data for
analysis.  I have used the following commands:

use nationaldataset
keep if strata==19
keep if Age==16|Age==17|Age==18|Age==19
svyset schlid [pweight=xwgtreg], strata(strata)

**Then I ran tabulate and other descriptive commands. In order to
calculate appropriate variance estimates/confidence intervals, and
p-values, I have created subpopulations for gender analyses.  I read
that this step is necessary when analyzing survey data using compex
survey procedures..?*

gen Subpopmale = .
label variable Subpopmale "Subpopulation1: males 16 to 19 years old"
replace Subpopmale = 0 if (Gender == 2)
replace Subpopmale = 1 if (Gender == 1)
label define Subpopmale 1 "male" 0 "female"
label values Subpopmale Subpopmale

svy, subpop(Subpopmale): tabulate Mother_education
svy, subpop(Subpopmale): tabulate Father_education

**In order to compare the national health survey dataset to the clinic
dataset, I have used the following commands
use clinicdataset
append using nationaldataset
replace samptype = 2 if sample==.
label define samptype 1 "clinic" 2 "national"
label values samptype samptype
svyset schlid [pweight=xwgtreg], strata(strata)

Can I just go ahead and run comparative analyses (relative risk
ratios) on this combined dataset? I don't quite understand if the
population weights are being accounted for by the national dataset. .
Thanks in advance for your help,
Kyleigh
*
*   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/