The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » weighting combined datasets
weighting combined datasets [message #478] Mon, 27 May 2013 00:52 Go to next message
marcelocr2 is currently offline  marcelocr2
Messages: 1
Registered: May 2013
Hi, I am doing a regression analysis about child nutrition. I am using 5 DHS datasets for the same country and I want to know if I have to use wights (because the DHS manual doesn't recommend it) and if I have to, how should I modify the weights in order to use the appended datasets.

thank you!
Re: weighting combined datasets [message #482 is a reply to message #478] Mon, 27 May 2013 14:45 Go to previous messageGo to next message
Messages: 292
Registered: March 2013
Senior Member

This discusses re-normalizing weights when you are using multiple surveys from multiple countries. If you want to re-normalize weights within county, this same procedure would apply: mp;goto=82&#msg_82

Now... I think re-weighting makes sense when you append multiple rounds of the same country, but you should be clear on what it means. Each survey is weighted to Ns - sample size of the survey. To re-normalize, you'd divide the DHS provided weight by the sum of the weights (which will be Ns or very close, if you lose some observations), and then they would sum to one. So, by re-normalizing, you are essentially weighting all the surveys to be of equal weight, regardless of sample size. The thought is that each survey is one nationally representative survey, and thus each woman in each survey is actually representing some real number of women. I think this is probably the right thing to do in most circumstances (I do it), but it is not intuitively clear to me it is always the right thing to do.

Re: weighting combined datasets [message #596 is a reply to message #482] Wed, 03 July 2013 18:27 Go to previous message
Messages: 12
Registered: March 2013
Location: Silver Spring Maryland
I don't think re-normalizing is weighting all surveys to be of equal weight. It simply scales the weight to the total number of observations and the relative variability remains the same.

I also don't think it is necessary, if you use survey software. The scale of the weights is irrelevant.

Bryan Sayer
Social & Scientific Systems, Inc.
Previous Topic: weighted median
Next Topic: Weighting the KR for multiple countries
Goto Forum:

Current Time: Sun Jun 23 22:50:39 Coordinated Universal Time 2024