[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Comparison of Dependent Proportions
At 09:21 02/08/02 -0400, Stephen Soldz wrote:
Doing a paired t-test for binary data won't get you far wrong, if your
sample size is large. The variance from the paired t-test is a special case
of the Huber variance for clustered data (where the clusters are the pairs
of responses and the observations are the individual responses). Of
course, it happens to be the maximum-likelihood variance for the bivariate
normal model as well. The main difference is in the degrees of freedom,
which is the number of pairs minus one for a paired t-test.
I'm trying to compare proportions in the same sample (e.g, rates of
agreement for 2 the same attitude statement for 2 different drugs). As I
have a student survey, I need to use the robust variance estimator to adjust
for the within-classroom design effect. What I want is an analog of the
paired t-test for binary data, with robust variance estimator. I thought I
could use svylc or svytest, but the only way I can do is after svymean,
resulting in an assumption of normality. Any ideas?
If your sample size is small, then it might possibly be better to restrict
the analysis to paired responses that are different, and calculate a
Binomial Clopper-Pearson CI using -ci- with the -binomial- option to
estimate the ratio of (1,0) responses to (0,1) responses. See -help ci-.
I hope this helps.
Lecturer in Medical Statistics
Department of Public Health Sciences
King's College London
5th Floor, Capital House
42 Weston Street
London SE1 3QD
Tel: 020 7848 6648 International +44 20 7848 6648
Fax: 020 7848 6620 International +44 20 7848 6620
or 020 7848 6605 International +44 20 7848 6605
Opinions expressed are those of the author, not the institution.
* For searches and help try: