st: Re: Bootstrap and pweights

["R.E. De Hoyos" <redeho@hotmail.com>]

Ben,

A good reference on the subject is:

Deaton, A. (1997) "The analysis of household surveys: a microeconometric 
approach to development policy", World Bank

It uses Stata (version 4.0) in its applications and the do-files are 
available online:

http://www.worldbank.org/lsms/tools/deaton/

As I understand, the problem of bootstrapping with weighted data (pweights) 
rests in the difficulty of respecting the total population where the data 
comes from. In the case of survey data, multiplying the observations by its 
respective -pweights- will give you the total population. When you 
use -bootstrap- or -bsample- the data generated violate this property.

A suggestion could be to expand the data using the -pweights- and then carry 
on the resampling, however if the survey design is a complex one --very 
likely--, you wont be able to reproduce the population with your 
bootstrapped sample. A second possible solution is to make the resampling 
within each value of -pweight- (usually the PSU or Strata or a combination 
of both under complex survey data). In this case you will be able to 
reproduce the total population.

see: -svybsamp- and -svybsamp2-


I hope this helps,

Rafa



----- Original Message ----- 
From: "Jann Ben" <ben.jann@soz.gess.ethz.ch>
To: <statalist@hsphsun2.harvard.edu>
Sent: Tuesday, October 25, 2005 11:36 AM
Subject: st: Bootstrap and pweights



> Hi, the issue of using bootstrap techniques with weighted
> data has been discussed several times on statalist. For
> example, see
>
> http://www.stata.com/statalist/archive/2003-09/msg00180.html
> http://www.stata.com/statalist/archive/2005-01/msg00218.html
> http://www.stata.com/statalist/archive/2002-12/msg00348.html
>
> The basic message is that bootstrap cannot be used with
> probability weights (pweights) in Stata. This is bad news
> and I think it would be very valuable to change that.
>
> Do I understand right, that the only problem with using
> pweights is that weighted sampling is not implemented in
> Stata's -bsample- command? In other words: Is it true that
> the bootstrap technique produces accurate results for
> weighted data if the weights are accounted for via a
> weighted sampling design?
>
> Furthermore: Is there a fundamental difficulty with
> weighted sampling? (I.e. why is weighted sampling not
> implemented in -bsample-?) How could weighted sampling be
> implemented in -bsample-? (References?)
>
> Any help will be greatly appreciated!
>
> Thanks,
> ben
>
> PS: Am I the only one who feels bothered by the no-pweights
> restriction of the bootstrap command? Is it true that
> -svy jackknive- could be an alternative?
>
> *
> *   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/