The DHS Program User Forum: Weighting data » Weighting after pooling multiple countries

Home » Data » Weighting data » Weighting after pooling multiple countries

Show: Today's Messages :: Show Polls :: Message Navigator

Weighting after pooling multiple countries [message #24633]

Fri, 10 June 2022 10:21

dgodha
Messages: 44
Registered: November 2016
Location: India

Member

Hello DHS Experts,

Thanks to you and this forum, I have learnt a lot on pooling and weighting.

My current focus on pooling is to gain insight into stunting and neonatal mortality across Sub-Saharan Africa. I have pooled 19 countries but I don't want one country to dominate the results. I plan to do logistic regression instead of multi-level modelling.

Accordingly, I have given equal weight to each survey (before pooling) using the following commands:

scalar TOTWT=1000000
quietly summarize v005
scalar T=r(sum)
gen v005r=v005*TOTWT/T

In addition, I have modified the stratum and psu variables as follows:

gen cluster_id=survey*1000+v021 
gen stratum_id=survey*1000+v023

Now, I am not sure if I have surveyset correctly- do I still need to divide my weight variable by 1000000? My code is as follows:

gen wt=v005r/1000000
svyset cluster_id [pweight=wt], strata(stratum_id) vce(linearized) singleunit(centered)

I will appreciate your guidance on this.
Many thanks

Deepali

[Updated on: Fri, 10 June 2022 10:22]

Report message to a moderator

Re: Weighting after pooling multiple countries [message #24657 is a reply to message #24633]

Wed, 15 June 2022 09:03

Janet-DHS
Messages: 685
Registered: April 2022

Senior Member

Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

For pweights it does not make any difference whether you multiply or divide by a constant (such as 1000000). To confirm that, I suggest that you do a regression (for example) both ways. I don't think you will see a difference....

I recommend that you get the cluster and stratum id codes this way:

egen cluster_id=group(survey v021)
egen stratum_id=group(survey v023)

"egen" (extensions to "generate") is a very powerful set of commands.

Report message to a moderator

Re: Weighting after pooling multiple countries [message #24665 is a reply to message #24657]

Thu, 16 June 2022 08:55

dgodha
Messages: 44
Registered: November 2016
Location: India

Member

Many thanks for explaining- it clarifies the issue for always.

Deepali

Report message to a moderator

Previous Topic:	create a weighted population variable
Next Topic:	Weight to use for regression that has children and mother level variables

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Thu Apr 18 16:05:06 Coordinated Universal Time 2024