| How to arrive at final wealth index score using PCA [message #28666] |
Mon, 19 February 2024 05:10  |
annabellebronson
Messages: 1
Registered: February 2024
|
Member |
|
|
I need to arrive at a final wealth index score for my analysis using PCA in stata. I am working with the Demographic and Health Survey dataset which provides only an asset-based wealth index for measuring socio-economic welfare. I am unsure about the variable normalisation process to arrive at the final wealth index score for each household. I have followed the variable preparation process using the Filmer and Pritchett /Demographic and Health Survey and coded each variable as a dummy. I have always assumed that if I predict the scores using the command predict score1 for example and apply the survey weights to generate break the scores up into quintiles, I arrive at quintiles representing different categories of wealth from the poorest to wealthiest. I, however, have been unable to duplicate the survey provided wealth index and quintile distribution using this approach. They provide the code in SPSS, and I am working in stata. I need to be sure what I am doing is right and not necessarily arrive at the very same index.
I did some more reading used the following document;http://www.psi.org/equity-wealth-quintileguide from the Population Service International (PSI). They show how to construct a wealth index based on a survey. In the document which is a guide they use the PCA command and then generate a factor score using the predict command, it appears they do not use this as the final wealth index score. They, however, go on to standardise each variable using the mean and standard deviation. They then proceed to create a final wealth index score for each household by multiplying the standardised variable by the factor scores obtained from the PCA and summing these up for each household. I guess my questions are;
1. Is predicting the relevant factor score using the predict command in stata alone adequate to arrive at a wealth Index score based on which to rank households.
2. Is ranking the households using quintiles and survey weight command sufficient to reflect the distribution of wealth based on this index.
3. is the process of generating the index using the standardised variable as described above same or similar to the stata predict command, ie. another way toobtain the score?
In the interest of brevity my commands are as follows;
pca n1, n2, n3 ... ni [aweight = weight], means
predict wealthscore1 wealthscore2
sort wealthscore1
xtile quintile= wealthscore1 [pweight=weight], nq(5)
Some clarity as soon as possible will be greatly appreciated. regards,
[Updated on: Mon, 19 February 2024 20:17] Report message to a moderator |
|
|
|
| Re: How to arrive at final wealth index score using PCA [message #28714 is a reply to message #28666] |
Wed, 28 February 2024 14:01  |
Janet-DHS
Messages: 698
Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS staff member, Tom Pullum:
Yes, you can match the SPSS procedure with Stata. Have you looked at the DHS website (https://www.dhsprogram.com/topics/wealth-index/index.cfm)? MANY variables go into that PCA. Various people have tried to simplify the construction, and I support that, but if you simplify it you will not get a match.
The conversion to quintiles is a little more complex. The following lines, applied the HR file, show how to do it ("mem" is the number of household members). You have to do it with the HR file because everyone in the same household has the same value of the continuous index and is in the same quintile.
set more off
use "...NGHR7AFL.DTA" , clear
keep hv001 hv002 hv005 hv012 hv013 hv270 hv271
gen mem = hv012
replace mem = hv013 if mem == 0
gen pwt=mem*hv005
gen wt=pwt/1000000
xtile hv270_test1=hv271 [pweight=pwt], nquantiles(5)
tab hv270 hv270_test1 [iweight=wt]
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
