Regarding dependent proportions in data (without survey design effects): it
reduces to the McNemar test.
This can easily be seen by encoding each of the two response variables (y1
and y2) as 1 for success and 0 for failure and computing d=y1-y2. It's
immediate that dbar=the difference in proportions. The variance, computed
under the null hypothesis (so you get var=(d2 instead of var=((d-dbar)2 )
reduces to the usual McNemar statistic.
If you use the design strata as covariates, you "should" get the statistic
you are looking for. (this is a guess, but I hope not too much of a
stretch)
Peter A. Lachenbruch
Director, Division of Biostatistics
Office of Biostatistics and Epidemiology
Phone: (301) 827-3320
FAX: (301) 827-5218
*
* 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/
