Re: st:compute wealth index score with PCA

[Friedrich Huebler <fhuebler@gmail.com>]

Wei Yee,

Your code is correct but your are not using any weights with the
-xtile- command. I assume you run the -pca- command with data where
one observation represents one household. Without weights in the
subsequent -xtile- command, the resulting quintiles may not reflect
the number of persons living in each household. If we assume that
poorer households are larger, the quintiles may be unbalanced, with
more persons allocated to the poorest quintile (identified by your
variable score1_cat) and fewer persons allocated to the richest
quintile. For example, if the total number of household members is
100, the distribution may look as follows:

- Quintile 1: 30 persons
- Quintile 2: 25 persons
- Quintile 3: 20 persons
- Quintile 4: 15 persons
- Quintile 5: 10 persons

Ideally, the quintiles should have 20 persons each, although this may
not be possible if there is insufficient variation in the wealth
scores. To create proper quintiles, add a weight variable with
information on the number of household members to the -xtile- command.

Friedrich

On Wed, Sep 11, 2013 at 1:03 AM, Wei Yee <weiyee20@gmail.com> wrote:
> Hi all,
>
> I am trying to compute a wealth index score and then categorized them
> into quintiles. I have never done PCA or factor analysis before and
> have spent the past few days reading up but I think I am still a
> little confused. I really hope you guys will be able to shed some
> light.
>
> I started with converting the questions/variables (eg. if they own a
> tv, radio, mobile, flooring materials) to binary.Then I use the -pca-
> command,
> "pca naturalfloor rudimentaryfloor finishedfloor nonbiomass biomass
> radio tv mobile nonmobile watch bike motorbike cart car motorboat",
> followed by "predict score1".
> Then, I use "xtile score1_cat = score1, nq(5)" to  categorized into quintiles.
>
> My questions are:
> 1. Is this the correct method to obtain the quintiles representing
> different categories of wealth?
> 2. How is this different from using the command : -factor (var1, var2,
> ..), pcf ?
>
> Any suggestions or feedbacks will be greatly appreciated. Thank you so
> much in advance.
>
> Best Regards,
> Wei Yee.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/