Montserrat Rue <[email protected]> asks:
> I need to compare means in a cluster study. I use the svymeans and lincom
> commands and I get the following output. My question is: why the CIs for the
> means overlap so much but the p-value is significant?
[output omitted]
Your estimated means are correlated due to the fact that you have observations
from both comparison groups within one or more of your clusters. Recall, that
the variance of the difference of two means is
Var(mean1) + Var(mean2) - 2 * Cov(mean1, mean2)
and in your case the last term is not zero. Type
. mat list e(V)
after your -svymean- command to see that this is indeed the case.
Heuristically, if your means are positively correlated, then knowing one mean
implies knowing a lot about the other, and this in turn means that you can
more precisely estimate their difference. As a result, even though the means
themselves have high variance, their difference won't and even a "small"
difference will show up as significant.
--Bobby
[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/
