|Replicate Wealth Index India [message #509]
|Tue, 04 June 2013 06:03
Registered: June 2013
I am working with the indian NFHS-3 data. I intend to replicate the calculation of the wealth index in another dataset. In order to do so, I am first trying to understand how the index was calculated in the DHS data.
I found the list of variables used in the process in the report on the NFHS-3 (p.43) and I am calculating the principal components analysis in Stata.
I am using the following code and the varlist below. For categorical variables I generated dummies first.
# Principal component analysis:
pca varlist, components(1)
# Extract first component
Predict index, score
Doing so I get an index that is close to the original index (hv271). But in some cases the households are not classified in the same order. For example if I classify households in wealth index quintiles according to the original index and the one I reconstructed and then compare results, I end up with a diverging classification in about 4% of all cases.
Have I missed a variable or is there an option of pca that I should have set?
If anyone has managed to replicate the index exactly I would be very glad for any input!
Many thanks and regards
sh56a house has any windows
sh56b house has windows with glass
sh56c house has windows with screens
sh56d house has windows with curtains or shutters
hv201 source of drinking water
hv205 type of toilet facility
hv213 main floor material
hv215 main roof material
hv214 main wall [exterior] material
hv226 type of cooking fuel
sh58 owns this or any other house
memsleep number of sleeping rooms per household member (hv126/hv009)
hv247 owns a bank account or post office account
sh47c pressure cooker
sh47g electric fan
sh47i b&w television
sh47j colour television
sh47k sewing machine
hv243a mobile telephone
hv221 telephone (non-mobile)
hv243c animal-drawn cart
sh47u water pump