The DHS Program User Forum
Discussions regarding The DHS Program data and results

Home » Topics » Wealth Index » Re: Calculating a Representative Wealth Index for Clusters Using DHS Sample Weights
Re: Calculating a Representative Wealth Index for Clusters Using DHS Sample Weights [message #30616 is a reply to message #30598] Wed, 08 January 2025 14:11 Go to next message
Janet-DHS is currently offline  Janet-DHS
Messages: 937
Registered: April 2022
Senior Member
Following is a response from DHS staff member, Tom Pullum:

Yes, the sample weight is the same for all households in the same cluster. If you want a single number to describe the cluster, it would be the unweighted mean. The best number to describe dispersion is probably just the standard deviation. The following analytical study could be helpful: https://www.dhsprogram.com/pubs/pdf/AS76/AS76.pdf.

Note that the wealth index is calculated with households as units. I would recommend calculating the cluster-level means and standard deviations within the HR file, and then copying those numbers into the other files you are using.
Re: Calculating a Representative Wealth Index for Clusters Using DHS Sample Weights [message #30657 is a reply to message #30640] Wed, 15 January 2025 08:51 Go to previous messageGo to next message
Janet-DHS is currently offline  Janet-DHS
Messages: 937
Registered: April 2022
Senior Member
Following is a response from DHS staff member, Tom Pullum:

These are good questions. The sample design does not take into account the locations of the clusters within the strata. They are not required to be spaced apart, for example, with a minimum distance between them. As a result, sometimes they are closely spaced. This would usually reflect variations in population density within the stratum. When the clusters are displaced, they are kept within the same level 2 areas. What you are finding is quite possible and "legal" within the sampling design. Not sure how you can handle it....

I don't think weights should play into the construction of the cluster-level WI. Say that you calculate the mean value of hv270 for the households in a cluster. Since all households in the cluster have the same weight, that mean will be the same whether or not you use weights. But then when you use the cluster-level mean in your analysis, you need to weight it by hv005 for the cluster.

I can imagine analyses in which the number of households in the cluster could be important. Could that account for the difference in fit that you are getting?
Re: Calculating a Representative Wealth Index for Clusters Using DHS Sample Weights [message #30705 is a reply to message #30669] Tue, 21 January 2025 16:06 Go to previous message
Janet-DHS is currently offline  Janet-DHS
Messages: 937
Registered: April 2022
Senior Member
Following is a response from DHS staff member, Tom Pullum:

I'm not sure that I understand the difference between your two calculations, but I will suggest a requirement or criterion: the weighted mean of the RWI should be the same as the weighted mean of hv270, in each cluster and in the sample of households as a whole. Here are the Stata commands I would use, illustrated for the Kenya 2022 survey, and within that, for cluster #1.

use "...KEHR8CFL.DTA" , clear
keep hv001 hv002 hv005 hv270

* Construct RWI
egen RWI=mean(hv270), by(hv001)
list if hv001==1, table clean nolabel

* compare the weighted means of hv270 and RWI within a specific cluster
summarize hv270 RWI [iweight=hv005/1000000] if hv001==1

* compare the weighted means of hv270 and RWI in the entire sample
summarize hv270 RWI [iweight=hv005/1000000]

If you are doing something other than this, I wonder what it is. If you are just comparing weighted and unweighted estimates, then it is not surprising that there will be some differences in any analysis. My preference would be for weighted estimates. Here I do not use weights to calculate RWI but DO use weights in any statistical analysis, exactly as I would do with hv270.
Previous Topic: How to arrive at final wealth index score using PCA
Next Topic: Type of house
Goto Forum:
  


Current Time: Sun Nov 15 23:22:10 EST 2026