[R] Problem with svyvar in survey package

[Thomas Lumley]

On Mon, Oct 24, 2011 at 4:21 AM, amitava <amtv.statpresi at gmail.com> wrote:
> I am facing a problem with a function in survey package. The function svyvar
> gives the estimated population variance from a given sampling scheme.  I am
> working with a data having more than four continuous variables. In order to
> have have population total for all those cont. variables I have written in
> the following format
>
> svyvar(~var1+var2+var3+var4+var5+var6,data) ; var1,var2,...,var6 are 6 cont.
> variables
>
> Now the problem is that,this function works properly when the number of
> cont. variables are less than or equal to three. But when the number of
> variables goes beyond three it gives a wrong answer,where we have two
> estimates for the 1st variable (here it is var1),and one of the variable is
> dropped,and also estimates differs. Sometimes it comes out to be a negative
> quantity.

You will have to give more information: what you describe does not
happen when I use svyvar. For example, using one of the built-in data
sets:

>data(api)
>dclus1<-svydesign(id=~dnum, weights=~pw, data=apiclus1, fpc=~fpc)
> v
         variance       SE
api00    11182.82 1386.414
api99    12735.21 1450.082
ell        285.01   54.081
mobility   127.11   55.790
meals      709.07  102.109
> v<-svyvar(~api00+api99, dclus1)
> v<-svyvar(~api00+api99+ell+mobility+meals, dclus1)
> svyvar(~api00+api99, dclus1)
      variance     SE
api00    11183 1386.4
api99    12735 1450.1
> svyvar(~ell+mobility, dclus1)
         variance     SE
ell        285.01 54.081
mobility   127.11 55.790
> svyvar(~meals, dclus1)
      variance     SE
meals   709.07 102.11

I assume you are using a current version of survey -- there was a bug
in printing svyvar results (though not in calculating them), but that
was fixed well over a year ago, as listed in the NEWS file.

-- 
Thomas Lumley
Professor of Biostatistics
University of Auckland