[(denormalized weights*total number of women(15-49) interviewed in the survey)/total number of women (15-49)at the time of survey].

Thanks.]]>

Quote:

Firstly, the DHS survey final report usually does not provide the target population size estimation (women 15-49) which is the first piece of information needed for de-normalizing the women weight. So in general cases, you need to find out this piece of information from outside sources. But for the Nigeria 2008 DHS, we did provided an estimation of the target population size (women 15-49), as you cited 31,624,485, given in table B.1.]]>

Secondly, for the number of women interviewed in a DHS survey which is the second piece of information needed for de-normalizing the women weight, we provide this piece of information in the DHS survey final report, usually in chapter one, table 1.2 in the section "Response rate". For the Nigeria 2008 DHS, this number is 33385. The number you cited 36800 from table B.1 is not the actual number of women interviewed, it was the expected number of women interviews calculated in the sample design.

Hope this helps.

Thanks. ]]>

Thanks. ]]>

What is the total number of women aged 15-49 that you found from the UN census data?

]]>

The UN data link is below

esa.un.org/wpp/unpp/panel_indicators.htm

Thanks

]]>

I asked about the number of women you used because the de-normalizing process essentially re-scales the weights away from summing up to total sample size and to summing up to number of women the survey represents (this is what allows it to make the weights representational across countries). That 9m number should correspond to something like the number of women in that age group in all your countries, if you are doing the calculation right.

Also - the sum of the weights doesn't really matter within any one survey - you are using them for probability proportioning. My usual strategy is to just de-normalize such that the weights in each survey sum to 1 (new weight = oldweight/sum-of-weights). This makes each survey in total have the same weight in cross-country regressions. The DHS method essentially weights each individual survey by the number of (in this case) women aged 15-49 in each country.

But the main point is that interpreting the weighted averages as a number of observations is the wrong way to think about it. You want to think about it as doing two things: within survey it is distributing weight based on probability of selection; between surveys it is weighting by population.

]]>

After denormalization and using the denormalized weights we got a total sample of 9,954329 cases, however our total sample of 25,438 (unweight count). using denomalized size, all univariate comparisons were significant.

If we recalculate this with denomalized weight as below formula;

[(denormalized weights*total number of women(15-49) interviewed in the survey)/total number of women (15-49)at the time of survey].

This makes each survey in total have the same weight in cross-country regressions and univariate comparisons will be better.

It is the best, isn't it?

Is it possible to use this approach for univariate comparison?

Thank you for your collaboration

]]>