The DHS Program User Forum
Discussions regarding The DHS Program data and results
Home » Data » Weighting data » Pooled Cross sections
Re: Pooled Cross sections [message #10666 is a reply to message #10658] Mon, 29 August 2016 10:05 Go to previous messageGo to previous message
Bridgette-DHS is currently offline  Bridgette-DHS
Messages: 3017
Registered: February 2013
Senior Member
Following is a response from Senior DHS Stata Specialist, Tom Pullum:

The word "denormalize" means nothing to me, even though it is commonly used on the forum. When you use svyset, the type of weight is always pweight. Stata always normalizes pweights so that the mean pweight in the data file is 1, and the sum of the pweights--the weights that Stata is using in the calculation, regardless of what you say--is the same as the number of cases in the file. You can divide v005 by 1000000 or multiply it by whatever you want, and it will have no effect on your results.

However, you can re-weight or RE-normalize PARTS of a data file. When you are putting several countries or surveys into a single file, you can internally re-allocate in exactly the way you say. There have been many postings on this. You can re-weight so that the weighted number of cases for a survey is proportional to the country's population or part of the population (e.g. women 15-49) , as you do. OR you can re-weight so that each file counts equally--that is, if there are k files, the sum of the weights for each file is 1/k times the total number of cases in the pooled file. The danger of proportional allocation is that a country such as India or Nigeria will sompletely dominate the results.

If you are calculating survey-specific estimates or looking at differences between surveys, there is no need to re-weight. You can put the surveys into a single file for data processing efficiency, but leave the weights untouched. The kind of re-weighting you are talking about is only relevant if you want to combine surveys to produce a single estimate, e.g. the contraceptive prevalence rate in West Africa. But then I would ask, is it really meaningful to calculate something like that? The surveys are done at different times, so what is the reference time point of your pooled estimate? Do you have all the countries in West Africa? (The answer of course is "no".) Can't you just give the CPR, say, for the countries and dates that you have, for which no re-weighting is needed? A lot of the forum discussion about weighting in pooled files is based on the desire to do a type of analysis that maybe shouldn't be done at all....
 
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Previous Topic: Antenatal visit with skilled providers (Ethiopian DHS 2000, 2005, 2011 AND 2016))
Next Topic: Using R to weight data for domestic violence
Goto Forum:
  


Current Time: Thu Mar 28 22:35:18 Coordinated Universal Time 2024