The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Dataset use in Stata » Multilevel analysis (Multilevel analysis on NDHS data)
Multilevel analysis [message #23563] Thu, 07 October 2021 23:51 Go to next message
Ugonna is currently offline  Ugonna
Messages: 17
Registered: June 2020
Member
Hello everyone,

I am doing a multilevel analysis in stata and I have some challenges I would like to be cleared on.

To find out the proportion of women in the community with at least secondary education, I did this:
gen sec_edu=0
replace sec_edu=1 if mat_edu>=4 & mat_edu<=5

So, to obtain the proportion at the cluster level, please what is the most appropriate option amongst these...

egen sec_edu_prop = pc(sec_edu), prop by(v021)
OR
egen sec_edu_prop = mean(sec_edu), by(v021)
OR
egen sec_edu_prop = rowmean(sec_edu), by(v021)



Then while running the analysis, do I need to apply a different kind of weight other than this?
svyset[pw=wt],psu(v021)strata(v022) singleunit(center)


Thank you for your anticipated response.
Re: Multilevel analysis [message #23570 is a reply to message #23563] Mon, 11 October 2021 06:35 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2400
Registered: February 2013
Senior Member

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


The usual variable for education is v106, with codes 0, 1, 2, 3. The recode would be

gen sec_edu=0 
replace sec_edu=1 if v106>=2 & v106<=3
The cluster-level proportion would come from 

egen sec_edu_prop = mean(sec_edu), by(v021)

The weights (v005) are the same for all the women in the same cluster. You can do the analysis with svyset as you give it (I would insert a space after "svyset"). However, because sec_edu_prop is the same for every woman in the same cluster, you could consider a multi-level model with women (or children) as level 1 and clusters as level 2. Methodological Report 27 (https://www.dhsprogram.com/pubs/pdf/MR27/MR27.pdf) will help if you go in that direction.

Re: Multilevel analysis [message #23573 is a reply to message #23570] Tue, 12 October 2021 03:29 Go to previous messageGo to next message
Ugonna is currently offline  Ugonna
Messages: 17
Registered: June 2020
Member
Hello Bridgette,

Thank you for your response.
I am going through the material sent and I saw IR file was used to get the level-weights at the individual level.

Being that I'm working with the child KR file, please I want to ask if the variables in the kr file can be used to for it.

Thank you
Re: Multilevel analysis [message #23574 is a reply to message #23573] Tue, 12 October 2021 05:55 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2400
Registered: February 2013
Senior Member
These are the different weights for different sample selections/units of analysis:

Households: hv005
Household members: hv005
Women or children: v005
Men: mv005
Domestic Violence: d005
HIV test results: hiv05

You need to divide the sampling weight you are using by 1,000,000.
Previous Topic: Cluster level population density data (Geospatial Covariate Datasets)
Goto Forum:
  


Current Time: Mon Oct 18 17:04:50 Coordinated Universal Time 2021