| Re: Sample weight for regional level analysis [message #28505 is a reply to message #28501] |
Mon, 22 January 2024 07:38   |
Bridgette-DHS
Messages: 3199
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
I recommend that you just use the national weight for all your tables. This is hv005 in the HR and PR files, mv005 in the MR and CR files, and v005 in the IR, MR, and KR files. You could use the state weight for individual states, but I see that you also want regional estimates, for groups of states, and there are no regional weights. If you use the state weights for a group of states, then each state in a grouping will contribute equally, instead of in proportion to the size of each state in the grouping. That would not be a good thing.
If you use the national weight for individual states, then virtually everything will come out exactly the same as if you use the state weights. I recommend that you confirm this by producing some estimate with the state weight, and then the national weight, and compare the estimates. The only difference should be that the weighted number of cases will be different. The differences can be large. In that comparison, the state-weighted sample size will be closer to the unweighted sample size but the national-weighted number of cases will be approximately proportional to the population of the states.
Note that the state weights are given by a single variable, not by separate state-specific variables. For example, in the PR file for the NFHS-5, the state weights are given by shweight, regardless of which state you are working with.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
