Weight for studying specific states or regions? [message #3494] |
Wed, 17 December 2014 18:30 |
vega25
Messages: 14 Registered: April 2014 Location: United States
|
Member |
|
|
Hi,
I just wanted to clarify: do the weights that we use in analysis for only one specific state or region (where the DHS is meant to representative) need to be different from the national weight variable v005 (women's individual sample weight)?
To put it differently:
Right now, for ease of programing in Stata, I calculate a new variable which is wt = v005/1000000. I suffix this to every single code in Stata using [iweight =wt] including when calculating summary statistics. Do I need to do something else if I want to limit the analysis to a particular state (say in India) or administrative region (say in Bangladesh)?
Thanks very much!
|
|
|
Re: Weight for studying specific states or regions? [message #3665 is a reply to message #3494] |
Fri, 23 January 2015 13:57 |
Bridgette-DHS
Messages: 3203 Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS Stata Specialist, Tom PullumThe weights that you use for a national-level analysis would apply for any sub-population, whether defined geographically (e.g. a region) or by some other cross-cutting variable (e.g. urban areas or the highest wealth quintile). Means, for example, that are weighted by v005, will be unbiased estimates of population means or sub-population means, however defined.
When you get down to a small subpopulation, such as the urban population of a single district, with very few clusters (maybe only one), the estimates will obviously deteriorate in quality. The standard errors will go way up because the sample size is small. However, the point estimates will still be unbiased.
|
|
|