| Combining data from several countries and time periods [message #2505] |
Tue, 01 July 2014 04:08  |
Yohannes
Messages: 6
Registered: July 2014
|
Member |
|
|
|
I plan to undertake a cross country multivariate analyses where the unit of study would be children and wonder how the data from bigger countries such as India, Pakistan, Bangladesh, Nigeria, Ethiopia and others affect the resulting analyses. I also wonder if the existing sampling weights can still be valid if one breaks up these big countries into their larger admisrative units (such as breaking India and Nigeria into states, or Bangladesh into divisions or use 'killil' in Ethiopia) and enter these geographic units into the cross country analysis as if the individual states, divisions or killil (in the case of Ethiopia) constitute a country on their own.
|
|
|
|
|
|
| Re: Combining data from several countries and time periods [message #2559 is a reply to message #2541] |
Fri, 11 July 2014 12:47  |
Bridgette-DHS
Messages: 3214
Registered: February 2013
|
Senior Member |
|
|
Following is a response from our Senior Sampling Statistician, Ruilin Ren:
If you want to pool several countries' surveys together, you need to de-normalize the sampling weights and redefine the cluster and stratification variables, please search the forum for "de-normalization", as it's a frequently asked question.
As for breaking each country into larger units such as states, you can do it without any concern, once the sampling weight is de-normalized.
Thanks,
Bridgette
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
