The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Sample weights and stratification - Nigeria 2008 and 2018
Re: Sample weights and stratification - Nigeria 2008 and 2018 [message #19275 is a reply to message #19218] Wed, 20 May 2020 08:28 Go to previous messageGo to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3087
Registered: February 2013
Senior Member

Following is a response from DHS Research & Data Analysis Director, Tom Pullum:

These are good questions. I will give recommendations but there is some room for different approaches.

I would definitely combine the surveys into a single file, but when you do this, you need an identifier for the survey. I would probably just construct survey=1 or 2 for 2008 and 2018, respectively. You will need unique identifiers for strata and clusters. I won't take the time to look up the stratum identifiers for these two surveys but let's assume it was v023 in each of them. You construct the combined stratum identifier with "egen stratum_ID=group(survey v023)". You construct the combined cluster identifier with "egen cluster_ID=group(survey v001)".

In each survey, v005 has been normalized so that the sum of the weights is the total number of cases (times 1000000), and therefore the mean weight in each survey is 1 (times 1000000). (Stata automatically re-normalizes pweights and that gets rid of the factor of 1000000.) When you combine the surveys, the overall mean of v005 will also be 1000000. You do not need to do anything with v005. That is, you do not need to do any re-scaling or re-normalizing.

You then use svyset including adjustments for weights, clusters, and strata.

I recommend that you only use the combined file for looking at differences between the two surveys. The aggregate of the two survey (e.g. the CPR for the pooled surveys) is not meaningful. If, say, you were combining surveys from several countries for a pooled analysis then you might want to re-scale the weights to take account of differences in sample sizes and/or population sizes but I would be very cautious with such an analysis, and fortunately that's not what you are talking about. Successive DHS surveys in the same country are well-suited for a difference-in-differences approach.

Good luck.
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Previous Topic: Pooling men, women and household DHS from Haiti
Next Topic: Weight about pooling 2003,2008,2013,2018 BR dataset in Nigeria
Goto Forum:

Current Time: Tue Jun 18 20:43:09 Coordinated Universal Time 2024