The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Domestic violence and HIV
Domestic violence and HIV [message #10280] Mon, 18 July 2016 14:14 Go to next message
Messages: 4
Registered: April 2016
I am looking at all women who were selected for the domestic violence module, who also have an HIV outcome and whose partners also completed the survey. Should I use d005 or the female's hiv05? Should the psu be v021?

I have done:
generate wgt = d005/1000000
gen psu=v021

svyset glm w_hiv03 violence [psu] [iweight=wgt], family(binomial) link(log)

Also, I cannot seem to use family with svyset in STATA. I would be grateful for any guidance. Thank you.
Re: Domestic violence and HIV [message #10288 is a reply to message #10280] Tue, 19 July 2016 11:20 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 2537
Registered: February 2013
Senior Member
Following is a response from DHS Specialists: Tom Pullum and Kerry MacQuarrie.

Comments from Kerry MacQuarrie: Both the domestic violence module and the HIV testing is likely to be a restriction on the sample. e.g., The DV module is administered to only one woman per household. HIV testing has a lower response rate than the survey overall (actually, the DV questions also have a lower response rate, but not as "lower" as HIV testing). Furthermore, HIV testing is often done not in every household, but in every second or third household. In both cases, the two weights account for their respective non-response and probability of selection. This leads me to the following recommendation:

In surveys where HIV testing is done in a subsample (1/2 or 1/3 of households), I'd go with the HIV weight.
In surveys where HIV testing is done in all households, I'd compare the n of the DV module and the n of the valid HIV test results to the total overall n. Whichever is the smaller n, I'd use that weight.

In AS36 (, we were using COUPLE data wherein both spouses had valid HIV test results and the woman had completed the DV module. We ended up using the MEN's HIV weight for all surveys (even those with testing in all households) because men's response rates are even lower than women's, and because the HIV samples were more restricted than the DV samples. We often recommend the men's weight for couple level data rather than the bother of constructing a true couple weight (and my understanding from Stan Becker's work is that creating a couple weight doesn't change much). We consulted with Ruilin about this before settling on this strategy. But we also ran at least some of our models (many not have been the final, final specifications) using the women's DV weight and there was almost no difference in point estimates or in associations found; there were some small differences in standard errors.

From Tom Pullum: Yes, the psu is given by v021. Usually v001 and v021 are identical. If they are not, use v021. We also recommend that you adjust for the strata, which are usually given by v023. (There have been other forum postings on identifying the strata.) Then you use svyset and svy like this:

svyset v021 [pweight=hiv05], strata(v023) singleunit(centered)
svy: glm w_hiv03 violence, family(binomial) link(log)

Re: Domestic violence and HIV [message #10378 is a reply to message #10288] Thu, 21 July 2016 15:38 Go to previous message
Messages: 4
Registered: April 2016
Thank you. This is extremely helpful!
Previous Topic: Domestic violence weight, denormalize, pooled cross-section, cross-tabulation
Next Topic: SPA survey Tanzania: Setting survey design and using weights for health facilities and health care p
Goto Forum:

Current Time: Thu Jun 30 09:15:28 Coordinated Universal Time 2022