The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Topics » Wealth Index » Estimation of cutoff points for Rwanda 2014-15 wealth index
Estimation of cutoff points for Rwanda 2014-15 wealth index Tue, 25 April 2017 11:32
 avmujica Messages: 2Registered: May 2015 Location: Oxford Member
Hi,

I am trying to replicate the cutoff points for wealth quintiles in the Rwanda 2014-15 survey. I am working with the HR dataset and have almost been able to recreate the distribution of households across quintiles using hv271 (wealth index factor score, 5 decimals) - I am working on Stata:

gen hhweight = (hv005*hv012)/1000000
xtile hhwealth = hv271 [pw = hhweight], nq(5)

However, when I run the analysis to get the cutoff points, I do not get the same results as those reported in the country spreadsheet (see below). Is this because the combscor is not the same as hv271?

Would really appreciate if someone could help me sort this out. Thanks!

Ana

_pctile hv271 [pw=hhweight], nq(5)
return list

scalars:
r(r1) = -.6646999716758728
r(r2) = -.4281199872493744
r(r3) = -.1414700001478195
r(r4) = .4008800089359283

Statistics
combscor Combined national wealth score
Percentiles 20 -.6939782
40 -.4744206
60 -.1973164
80 .3421368

Re: Estimation of cutoff points for Rwanda 2014-15 wealth index [message #12387 is a reply to message #12298] Mon, 08 May 2017 16:26
 Trevor-DHS Messages: 799Registered: January 2013 Senior Member
Combscor is the same as hv271, but the percentiles from combscor that you show below are actually weighted percentiles for households (total of 12699 households), not for household members, so are not the actual quintile cutoffs.

Your code for calculating the quintiles produces virtually identical results to those in hv270, apart from four cases:
```. list hv271 hv270 hhwealth if hv270 != hhwealth

+-----------------------------+
|  hv271     hv270   hhwealth |
|-----------------------------|
7064. | -66470    poorer          1 |
7743. |  40088   richest          4 |
7973. | -63769    poorer          . |
11711. | -14147    richer          3 |
+-----------------------------+```

I think there are two reasons for the slight differences:
1) There is one case for which you have hhwealth == . This is because there is 1 household with 0 de jure members listed. For the few households that we get with 0 de jure members listed, we use the de facto number of members instead. thus we might have
```gen ppl=hv012
replace ppl=hv013 if hv012==0
gen hhweight = (hv005*ppl)/1000000```

2) I think there is some rounding going on. The other 3 cases above are all right at the quintile boundaries that you calculated. When we calculate the quintiles we actually have more digits of precision of the scores. We use the SPSS Rank command and I expect that there are slightly different rules about selection of quintiles between the SPSS Rank and the Stata _pctile commands, in addition to the precision issue.
So I can't fully explain the differences, but hopefully this gives you enough to work with.
Re: Estimation of cutoff points for Rwanda 2014-15 wealth index [message #12389 is a reply to message #12387] Mon, 08 May 2017 17:48
 avmujica Messages: 2Registered: May 2015 Location: Oxford Member
Many thanks Trevor - will re-estimate using adjusted weights.
 Previous Topic: PCA Results for Sierra Leone 2013 Survey? Next Topic: Construction
Goto Forum:

Current Time: Thu Jun 13 13:06:03 Coordinated Universal Time 2024