The DHS Program User Forum: Wealth Index » Zimbabwe 2015 Wealth Index quintiles

Home » Topics » Wealth Index » Zimbabwe 2015 Wealth Index quintiles

Show: Today's Messages :: Show Polls :: Message Navigator

Zimbabwe 2015 Wealth Index quintiles [message #13560]

Sat, 18 November 2017 16:17

amyfinnegan
Messages: 5
Registered: March 2013
Location: Durham, NC

Member

Hello -

I am working with the 2015 Zimbabwe DHS to replicate the wealth quintiles. I want to apply the cut-off values to a wealth index we collected in rural Zimbabwe (we asked the same questions used in ZDHS 2015) to show how our sample compares to the population of Masvingo county (where our survey was fielded).

I have downloaded the principal component output for Zimbabwe 2015 from here: https://dhsprogram.com/topics/wealth-index/Wealth-Index-Cons truction.cfm.

I am trying to replicate the quintiles in Stata. But they are not exactly as in the spreadsheet provided.

In the spreadsheet, the values are:

Mean -.1347808
Std. Error of Mean .00903679
Median -.3942115
Mode .03433a
Std. Deviation .92749292
Minimum -1.63862
Maximum 3.24577
Percentiles
20 -1.0280612
40 -.6095050
60 -.0076580
80 .8723670

Here is how I calculated the HHMEMWT:

gen dejure = hv012
replace dejure = hv013 if hv012 == 0
gen HHMEMWT = (dejure*hv005)/1000000

After weighting, I have the same number of observations and min/max as the spreadsheet but a different mean and slightly different standard deviation: :

. sum hv271 [aw=HHMEMWT]

Variable | Obs Weight Mean Std. Dev. Min Max
-------------+---------------------------------------------- -------------------
hv271 | 10,534 42909.8382 -183250.4 927336 -1638620 3245770

The output from using xtile in Stata gives the below, which does not match the spreadsheet:

. _pctile hv271 [pw=HHMEMWT], nq(5)

. return list

scalars:
r(r1) = -1050290
r(r2) = -654000
r(r3) = -175310
r(r4) = 860240

This matches with the summary by quintile in the ZDHS data:

. bysort hv270: sum hv271 [aw=HHMEMWT]

-> hv270 = poorest

Variable Obs Weight Mean Std. Dev. Min Max

hv271 1,758 8581.44865 -1267315 133072.1 -1638620 -1050780

-> hv270 = poorer

Variable Obs Weight Mean Std. Dev. Min Max

hv271 1,707 8581.49289 -851790.8 111743.7 -1050290 -654100

-> hv270 = middle

Variable Obs Weight Mean Std. Dev. Min Max

hv271 1,774 8582.97362 -441030.8 133150 -654000 -175310

-> hv270 = richer

Variable Obs Weight Mean Std. Dev. Min Max

hv271 2,690 8582.03223 409289.8 322580.7 -175210 860240

-> hv270 = richest

Variable Obs Weight Mean Std. Dev. Min Max

hv271 2,605 8581.89077 1234530 312694.8 860620 3245770

Why does creating quintiles in Stata not match the principal component output?

Many thanks!
Amy

Report message to a moderator

Re: Zimbabwe 2015 Wealth Index quintiles [message #14150 is a reply to message #13560]

Mon, 26 February 2018 11:47

Liz-DHS
Messages: 1516
Registered: February 2013

Senior Member

A response from Dr. Shea Rutstein:
Quote:

The statistics of the combined national wealth score are calculated using the household weights (hv005, not hhmemwt). The quintiles are calculated using hhmemwt. That is the difference.

Report message to a moderator

Re: Zimbabwe 2015 Wealth Index quintiles [message #14154 is a reply to message #14150]

Mon, 26 February 2018 14:00

amyfinnegan
Messages: 5
Registered: March 2013
Location: Durham, NC

Member

Hi Liz and Dr. Rutstein,

Thank you for this information!

I can see now that when we want to know the population level statistics of the wealth distribution, we use the household weight (hv005). When we want to know how to divide the households into quintiles, we must use the hhmemwt.

All the best,
Amy

[Updated on: Mon, 26 February 2018 14:00]

Report message to a moderator