The Rao, Wu and Yue (1992) approach is widely used by Statistics Canada and, at least for the end user, is very similar to the BRR replicate sampling weights described on page 54 of the Stata Survey Data manual. Basically, the recipe goes as follows: draw a sample from your complex design sample following the sampling design (which means, among other things, sample clusters within strata) but with replacement, then recompute the weights so that the replicate sample truly be a sample of the population the original sample was drawn from, and repeat the whole procedure until you have 500 or 1000 sets of weights. When doing an analysis, get your point estimates the usual way, but compute your VCE using the 500 or 100 sets of weights.
Of course, only the people who can compute the original weights can produce these replicate bootstrap weights. Once the set of 500 or 100 replicate bootstrap weights is produced, all analysts will be bootstrapping over the same set of replicates.
Hope this helps.
Beno�t Laplante, professeur
Directeur des programmes de d�mographie
Centre Urbanisation, Culture et Soci�t�
Institut national de la recherche scientifique
Universit� du Qu�bec
Page web personnelle
Les programmes de d�mographie � l'INRS
Le babillard de la d�mographie � l'INRS
-----Message d'origine-----
De�: [email protected] [mailto:[email protected]] De la part de Ben Jann
Envoy�: 13 mars 2007 15:59
��: [email protected]
Objet�: Re: st: BOOTSTRAP and pweights
Stas wrote:
> I think the best idea with the bootstrap in complex surveys is to
> relegate everything to the weights and modifications of those, as
> described in Rao, Wu and Yue (1992) paper that I can send.
I'd love to see that, if you don't mind. Thanks.
> Doing the stuff properly with -pweights- would require something like
> PPS sampling, which is another disaster in the survey statistics world
> -- it is far more computationally intensive than it might seem.
This is what confuses me. When working on -kdens- I did some
simulations to compare the bootstrap CI's with the normal theory CI's
in the presence of -pweights-. I was convinced that I had to implement
PPS sampling for the bootstrap to be accurate. However, the results
were as follows:
- bootstrap with PPS sampling (and leaving away the weights within the
bootstrap samples) produced CI's that were like those obtained form
normal theory formulas with -aweights-(!)
- bootstrap with simple random sampling (and applying the weights
within the bootstrap samples) produced CI's that were like those
obtained form normal theory formulas with -pweights-
Well, I don't know. Hm ...
ben
*
* 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/
*
* 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/
