| Wealth Index or CWI [message #8513] |
Fri, 06 November 2015 16:33  |
riti_s
Messages: 6
Registered: October 2014
Location: Los Angeles
|
Member |
|
|
Hi - I have a few questions regarding a multi-year and multi-country DHS study.
1. When summarizing the wealth index for full sample (also true for one country, one DHS year) we are finding that the category breakdown is not reflective of quintiles (i.e. 20% each) and the richest quintile tends to be underrepresented (e.g. finding for Bangladesh 2011, richest is around 14-15% unweighted and 15-16% when individ weight (v005/1000000) applied). Any insight into what might be going on? (likely a sample weight issue, so any direction on how to weight would be appreciated).
2. A bigger conceptual issue, however, is whether for our multi-country, multi-year study, we should actually be using CWI and not the wealth index. The reason why we are thinking we don't necessarily need to use CWI is that we are not interested in comparisons of wealth between countries or between years, rather we are using the wealth index as a covariate/control variable as it relates the association between our predictor and child health outcome. We are interested in seeing what your take is on this issue.
Many thanks!
|
|
|
|
| Re: Wealth Index or CWI [message #8522 is a reply to message #8513] |
Sun, 08 November 2015 16:53  |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
Just a note on 2: if you want wealth index as a predictor and are using multiple surveys, you have to rely exclusively on within-country-year variation in estimating your effect of interest. There is no reasonable comparison of wealth quintile effects across country or survey round, wealth index gives information only relative to others measured at the same place and same time. Isolating that kind of variation may be hard (de-meaning by country-year might help, but then still you are saying that the effect of changing a wealth quintile is the same in each country at each time period).
So it sounds like you know all that, and want to use the CWI to get around that problem. I would just say that this has a different kind of effect, allowing for the comparisons across-country but maybe doing less-well within country. If you want an "effect modifier" that describes differences in the relationship of interest across-countrys, this makes sense. If you want an effect modifier that describes within-country-time inequalities in health, you want the wealth index.
If you just want a "control" variable - I'd guess that you don't really need either one. If you want to know the impact of relative wealth at a given point in time and space on your outcome (not as an "effect modifier" but as a determinant of health, you'd want the wealth index. If you want to see how the outcome changes across space and time (how does child health respond to economic growth/changes) you'd want the CWI, and to think carefully about what variation from survey-to-survey you want to use.
You might be able to use both, to get at within- and across-country-year variation. But I guess mostly it just depends on what kind of "wealth effect" you are interested in.
|
|
|
|
|
|
| Re: Wealth Index or CWI [message #8541 is a reply to message #8513] |
Wed, 11 November 2015 21:41 |
Liz-DHS
Messages: 1516
Registered: February 2013
|
Senior Member |
|
|
Dear User,
Here is a response for #1 in your query from Dr. Sarah Staveteig:
Quote:Re #1. Wealth quintiles are computed for the de jure household population, not just the population of women 15-49. The proportion of women in each quintile will depend on their concentration in poorer households. The wealth index weight is created by multiplying the household weight by the number of de jure residents. If there are no de jure residents then we multiply the household weight by the number of de facto residents.
In Stata, you can verify the household population quintile weighting is evenly split by checking the Bangladesh household file:
use BDHR61FL.DTA, clear
gen usualhh=hv012
replace usualhh=hv013 if usualhh==0
gen popwt=hv005*usualhh/1000000
ta hv270 [iw=popwt]
The result produced from this is:
wealth |
index | Freq. Percent Cum.
------------+-----------------------------------
poorest | 15,786.124 20.00 20.00
poorer | 15,777.318 19.99 40.00
middle | 15,781.823 20.00 60.00
richer | 15,783.583 20.00 80.00
richest | 15,782.26 20.00 100.00
------------+-----------------------------------
Total |78,911.1086 100.00
(Note: everyone in the household has to be assigned the same wealth, so it is normal that some quintiles will be 20.01 or 19.99 percent).
|
|
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
