[R] Re : Complex survey analysis help

[Thomas Lumley]

On Sun, 1 Oct 2006, justin bem wrote:

> Dear Thomas !
>
> Thank you for you help ! I need and explaination. If I suppress second 
> and third stage what become the variance du to second and third stage ?
>
> I have try this
>     >E3stage<-svydesign(ids=~stage1id, weight=~wei ...)
> and
>     >E3stage<-svydesign(ids=~stage1id+stage2id+stage3id, weight=~weights...)
>
> And I get the same results

Yes, just as in the example on the help page for svydesign.

When the first stage is sampling with replacement the variance estimator 
is based on the p-expanded cluster totals being a simple random sample 
from a population distribution.

This is conservative, and better variance estimators are possible in 
principle.  I am hoping to include better estimators in a future version 
of the package, but I need to do some more reading first -- there does not 
seem to be agreement on the best variance estimators (except for the 
impractical one that requires pairwise sampling probabilities).  I would 
welcome suggested references (especially if they are in English).

 	-thomas

Thomas Lumley			Assoc. Professor, Biostatistics
tlumley at u.washington.edu	University of Washington, Seattle