Roger Harbord wrote:
Does anyone know how to find a confidence interval for the covariance of 2
variables (NB not the correlation) based on a sample of iid observations
from (approximately) a bivariate normal distribution?
Stata code would be ideal (esp if it allows aweights), but a formula or a
reference would be a great help too.
I guess i could always bootstrap but my sample sizes are rather small
(often around 5 to 30).
--------------------------------------------------------------------------------
It doesn't allow aweights and it's asymptotic, which is a little dicey with
sample sizes of 5 to 30, but -xtmixed- will provide confidence intervals for
the variance of the random effect of iid pairs of observations, which is the
covariance of the two observations. You should also be able to get there
via a Fisher z-transform of the correlation coefficient, again asymptotic.
For a reference to the latter (and to an apparently more accurate "Ruben's
approximation"), see the link in the posting to a related question earlier
this month ( www.stata.com/statalist/archive/2005-09/msg00544.html ).
Joseph Coveney
clear
set more off
set seed `=date("2005-09-29", "ymd")'
set obs 30
generate float observation0 = invnorm(uniform())
generate float observation1 = 0.5 * observation0 + ///
sqrt(1 - 0.5^2) * invnorm(uniform())
correlate observation*, covariance
generate byte pair_id = _n
reshape long observation, i(pair_id) j(member_indicator)
xtmixed observation member_indicator || pair_id:, variance nolrtest
clear
set obs 5
generate float observation0 = invnorm(uniform())
generate float observation1 = 0.5 * observation0 + ///
sqrt(1 - 0.5^2) * invnorm(uniform())
correlate observation*, covariance
generate byte pair_id = _n
reshape long observation, i(pair_id) j(member)
xtmixed observation member || pair_id:, variance nolrtest
exit
*
* 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/