# Re: st: BRR Bootstrap weights w/xt commands

 From Stas Kolenikov To statalist@hsphsun2.harvard.edu Subject Re: st: BRR Bootstrap weights w/xt commands Date Mon, 2 Mar 2009 08:54:08 -0600

```Use -bs4rw- command that was written by Jeff Pitblado for this
purpose. As far as I know, Stat Canada uses SAS in house, and that is
almost impossible to change until a new generation of analysts grows
up who demand Stata... or at least until those analysts progress high
enough in the decision making circles to influence the software
choice.

I will be teaching a short course on survey bootstrap, bootstrap
weights and Stata implementation at the Joint Statistical Meetings in
Washington, DC.

Note that there are weird interpretational difficulties with
longitudinal survey data. What is the population to which the sample
results can be generalized? The straightforward interpretation of
weights as inverse probability of selection is that it is the
population that existed at the time when the sample was first taken.
If there have been any changes to the sample (refreshment, rotation,
etc.), then the population to which the sampling weights scale up the
sample may not have ever existed. David Binder and Georgia Roberts of
Stat Canada gave a talk about these issues a couple of years ago, but
to my knowledge the journal they sent it to could not find
knowledgeable enough referees to go through this, so it got kinda
stuck somewhere in the pipeline.

On 3/2/09, John Sandberg, Prof. <john.sandberg@mcgill.ca> wrote:
> Hi;
>
>  I'm new to bootstrap weighting but have a big problem to solve. I work
>  w/longitudinal data from Statistics Canada which does not release design
>  information w/their surveys for confidentiality reasons.  Instead, they
>  release a series of bootstrap replicate weights  along with the data and
>  std. errors are calculated, when using STATA at least using the svyset
>  with balanced repeated replications  such as:
>
>  svyset [pweight=weight], brrweight(bsw1-bsw1000) vce(brr)
>
>  The svy commands do not work with the xt commands, however. This really
>  restricts the analyses that can be done with the rich Stats Can. data
>  available.  My question is as follows. First has anyone faced and
>  resolved this problem before with STATA?  If not, would it be
>  technically possible to re-write the the vce(bootstrap)  code in the xt
>  ado files so that I could substitute the replicates provided by
>  statisitcs Canada?  Can I write my own weight code w/bsample? Any help
>  would be greatly appreciated. I believe this is a big reason STATA isn't
>  used as much as it could be with these data.
>
>  Best,
>  Jack
>
>
>
>  John Sandberg
>  Associate Professor
>  Department of Sociology
>  McGill University
>  Stephen Leacock Building Room 712
>  855 Sherbrooke Street West
>  Montreal, Quebec, H3A 2T7, Canada
>  (514) 398-2946
>  john.sandberg@mcgill.ca
>
>
>
>  *
>  *   For searches and help try:
>  *   http://www.stata.com/help.cgi?search
>  *   http://www.stata.com/support/statalist/faq
>  *   http://www.ats.ucla.edu/stat/stata/
>

--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```