[R] survey: estimating a covariance matrix

[David Winsemius]

David Winsemius <dwinsemius at comcast.net> wrote in 
news:Xns9A2E679818E94dNOTwinscomcast at 80.91.229.13:

> "Daniel Oberski" <daniel.oberski at gmail.com> wrote in
> news:817e25730801230506v3266a03fia82d7b22c5581153 at mail.gmail.com: 
> 
>> Hello
>> 
>> Does anybody happen to know if it is possible to use the survey
>> package to estimate a covariance matrix from a complex survey?
>> 
>> I have design weights and clusters (no strata), and want to get a
>> covariance matrix with preferably the effective sample size or else
>> an estimate of the variance-covariance matrix of the covariance
>> matrix ("asymptotic covariance matrix"). Is this possible?
> 
> Have you looked at Lumley's survey package?
> 
> <http://faculty.washington.edu/tlumley/survey/>
> 

Or more specifically:
<http://faculty.washington.edu/tlumley/survey/html/svyglm.html>

predict() returns a variance-covariance matrix by default. My thought was 
that newdata equal to estimated means should yield the vcov that you 
desired.

-- 
David Winsemius