| Re: Weighting of pooled country and year data [message #30697 is a reply to message #30671] |
Tue, 21 January 2025 12:55  |
Bridgette-DHS
Messages: 3229
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
Unfortunately, Shireen no longer works at DHS.
Your question seems to be more about the specification of the svyset command than about weighting. There have been many related posts. The svyset command is determined by the sampling design. If you pool samples, you need to take into account that the clusters and strata are different in different surveys, even in the same command.
Construct a unique identifier for each survey, e.g. survey=1, 2, 3, etc. v000 does not actually serve this purpose because there can be two or even more successive surveys in the same country with the same value of v000. For each survey, define (say) cluster=v000 and stratum=v023 (or perhaps something else for older surveys). Then append the surveys and construct unique codes for cluster and stratum with
egen cluster_id=group(survey cluster)
egen stratum_id=group(survey stratum)
Then your overall svyset command will include cluster_id and stratum_id. There have been countless postings on the construction of weights in pooled files and I will not repeat any of them.
The birth year of respondents in the IR file is v010, and that is what I use to define birth cohort. I do not understand "birth_cohort is the cohort in which the observation was born." There must be a typo somewhere.
You do not need to adjust svyset for any covariates, birth cohort or anything else, because they are not part of the sampling process or design. You can analyze the data by age or cohort or year of data collection or region, etc., without changing svyset.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
