I would be grateful if someone could help me understand the basic issue
involved in using sampling weights in the absence of stratification and
clustering.
If I compute the weighted mean and standard error of a variable in
SPSS(using the sample weight) or in Stata using -ci- and aweights, I get
the same estimates for the weighted mean and standard error. However, if
I use -svymean- with pweights, I get the same estimate for the weighted
mean, but the standard error is different. There is a note on page 350 of
[U] stating the standard error provided by SPSS or -ci- with aweights is
not correct. I'm not sure why.
Also, if I compute a 2x2 table in SPSS using the weights, I get a Pearson
chi-square value and p-value. However, in Stata, using -tabluate- with
aweights, the person chi-square cannot be computed. If I use -svytab-
with pweights, I get an uncorrected chi-square (which is sometimes a bit
different from that provided by SPSS) and a design based f-test. It seems
the recommendation is to use the design-based f-test. If the sample
weights are being used and there is no stratification or clustering, why
use the design-based f tests as opposed to the uncorrected Pearson
chi-square?
Thanks,
Mike Frone
*
* 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/