Re: st: combining DHS surveys: what weights to use

[Steve Samuels <sjsamuels@gmail.com>]

v005/1000000  will sum to sample size in the recoded DHS data sets;
the unscaled weights are apparently unavailable. If you want weights
for totals, in a single country or across countries, you _must_ first
divide v005 by 1,000,000 and then calculate the new weight NW as Bruno
shows.

The sum of the unscaled weights would also have been a valid estimate
of the population total.  Forcing the user to find an external total
estimate like P1549, is a major reason why I  have disliked the "sum
to sample size" weights. However,  even with unscaled weights, it is
very useful to have an external estimate of the population total for
comparison.  In a recent study, the two estimates differed by a
disturbing amount. I traced the reason to the fact that PSUs had been
selected with equal probabilities, not with probability proportional
to size.  Small PSUs were over-represented, and the sum of the
unscaled weights was too small.  Response might well have been related
to PSU size, so post-stratifying to  an external population total
would not have eliminated bias. The solution was to include  PSU size
category as one of the raking variables, with category  population
totals from the most recent census.

Steve

Steven Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax:    206-202-4783




On Tue, Jul 13, 2010 at 3:24 AM, SCHOUMAKER Bruno
<bruno.schoumaker@uclouvain.be> wrote:
> Do you want to do descriptivee analyses for the pooled data set (16
> countries together) ?
>
> In this case, I suppose you need to give greater weights to larger
> countries. Since all children of all women aged 15-49 are usually included
> in DHS, I would use the population of women aged 15-49 in a country to
> compute new weights. This population can be obtained from World Population
> Prospects online database.
>
> The idea is to compute the sampling rate in each country (sample size of
> women aged 15-49 divided by the population aged 15-49), and to compute the
> country weight as the inverse of the sampling rate. Then multiply the
> orginal individual weights by the country weights.
>
> *** in summary
> Original weight in DHS : v005 (which should preferably be divided by 1 000
> 000)
> Country specific weight :CSW= P1549/n1549 (population aged 15-49 in the
> country / sample size of )
> New weight : NW=v005*CSW
>
> Use pweights in the analyses (weights will be normalized by Stata).
>
>
> Hope this helps
>
> Bruno
>
>
>
> Le 12/07/2010 23:11, Stas Kolenikov a écrit :
>>
>> I am not sure why you would want to do anything at all with your
>> weights. Can you explain what kind of modifications did you have in
>> mind?
>>
>> On Mon, Jul 12, 2010 at 3:42 PM, Ana Gabriela Guerrero Serdan
>> <ag_guerreroserdan@yahoo.com>  wrote:
>>
>>>
>>> Dear all,
>>>
>>> I am working with DHS surveys for different countries at one point in
>>> time. I want to do regional analysis based on the surveys that I have
>>> (currently for 16 different countries).
>>>
>>> Does anyone has advice on how I should treat sample weights in this case,
>>> if I want to do simple descriptive statistics on the proportion of children
>>> that are undernourished or attending school?
>>>
>>> thanks,
>>> Gaby
>>>
>>>
>>>
>>> *
>>> *   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/
>>>
>>>
>>
>>
>>
>
>
> --
> Bruno SCHOUMAKER
>
> Centre de recherche en démographie et sociétés
> Université catholique de Louvain
> 1-17 PLace Montesquieu
> 1348 Louvain-la-Neuve
> BELGIUM
>
> Tel. +32 10 474136
> Fax. +32 10 472952
>
> bruno.schoumaker@uclouvain.be
> www.uclouvain.be/demo
>
> *
> *   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/
>



--

*
*   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/