The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Number of obs and population size not the same
Number of obs and population size not the same [message #26782] Mon, 01 May 2023 08:00 Go to next message
Johanne Elgaard is currently offline  Johanne Elgaard
Messages: 6
Registered: April 2023
Member
I'm working with the 2014 Ghana DHS. I am trying to calculate frequencies and percentages after weighting the data. My problem is that my number of observations are 945, but when using both svyset and [iweight] I get a population size of 925 and a number of obs of 945 in the output. When calculating the sum of the weighted frequencies together, it also adds up to 925. Can anyone explain the difference between number of obs and population size to me?

Thank you:)
Re: Number of obs and population size not the same [message #26784 is a reply to message #26782] Mon, 01 May 2023 08:44 Go to previous messageGo to next message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3210
Registered: February 2013
Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

I believe you are just seeing the difference between the weighted and unweighted numbers of cases. In the IR file, v005 is scaled or normalized so that the mean value of v005/1000000 is 1. However, for subpopulations, or in the KR file, where v005 is attached to the woman's children, the mean weights don't necessarily retain this property. That's why the weighted and unweighted totals differ. If you think this is not the explanation, please let us know.

Re: Number of obs and population size not the same [message #26785 is a reply to message #26784] Mon, 01 May 2023 09:10 Go to previous messageGo to next message
Johanne Elgaard is currently offline  Johanne Elgaard
Messages: 6
Registered: April 2023
Member
Thank you for your response :) That could be the explanation since I'm using the KR file. Should I then just weight the data when during my logistic regression but not interpret specifically on the frequencies since the weighted and unweighted numbers do not correspond here?
Re: Number of obs and population size not the same [message #26786 is a reply to message #26785] Mon, 01 May 2023 12:45 Go to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3210
Registered: February 2013
Senior Member

Following is a response from Senior DHS staff member, Tom Pullum:

You should not have to do any re-weighting. If you are using Stata, with svyset and [pweight=v005], Stata will always re-normalize so the weights have a mean of 1. As part of this, the factor of 1,000,000 is removed. (That's why you get the same results with [pweight=v005] or with [pweight=v005/1000000]. I don't know about packages other than Stata.

Previous Topic: All-women factor in trend analysis
Next Topic: Household weighted prevalence estimates
Goto Forum:
  


Current Time: Fri Dec 13 15:33:06 Coordinated Universal Time 2024