I am trying to analyse first births among women aged 20-24, using surveys that use only ever-married women samples. However, I am a little confused as to how to go about using the all women factor. I've searched the forum to find an answer, and have come across a solution that is provided for SPSS, so I know that the denominator and the numerator have to be treated separately, but I'm not sure how to adjust for this using stata and in particular what happens when looking at only a sub-group (those aged 20-24). I am not using data for a specific country right now, but for the sake of working this out, let's say I am using Bangladesh 2011.

Any help would be appreciated. Many thanks. ]]>

The all-women factors are a major headache. Fortunately only a few surveys are limited to ever-married women.

The weight for the numerators is not affected, but the weight for the denominators will be multiplied by awfactt/100. The "t" after "awfact" tells you that that weight is for the total. If you want weight separately by urban-rural place of residence, or region, or one or two other covariates, you look for another letter (instead of "t"). What kind of analysis of first births do you want to do?

I will insert below a trick to get the mean of v201 using awfactt with Poisson regression and an offset. It may help.

* Strategy to implement awfactt for a mean

* Illustrate with Egypt

* I want to get:

* the unweighted mean ceb for emw

* the weighted mean ceb for emw

* the weighted mean ceb for all women (using awfactt)

set more off

cd e:\DHS\DHS_data\IR_files

use EGIR61FL.DTA, clear

keep v005 awfactt v201

rename v201 ceb

* the unweighted mean ceb for emw

summarize ceb

* the weighted mean ceb for emw

summarize ceb [iweight=v005/1000000]

* brute force

gen numerator=ceb*(v005/1000000)

gen denominator1=v005/1000000

gen denominator2=(v005/1000000)*(awfactt/100)

save temp.dta, replace

collapse (sum) numerator denominator*

gen mean1=numerator/denominator1

gen mean2=numerator/denominator2

* mean1 and mean2 are the weighted mean ceb for emw and all women, respectively

list, table clean

use temp.dta, clear

* Now get these three means using poisson regression

* this will match with the unweighted mean ceb for emw

poisson ceb, irr

* this will match with the weighted mean ceb for emw

poisson ceb [pweight=v005], irr

* this will match with the weighted mean ceb for all women

gen off=log(awfactt/100)

poisson ceb [pweight=v005], offset(off) irr

]]>

I'm am still a little confused as to how to use the all women's factor. For logistic regression, do I use it as part of the survey set?

svyset[pweight = sample_wt], psu (psu) strata(sample_domain)

svy: logistic overweight i.employmentFinal childage_mo ...

I am also using education a as covariate so then I would seem I need to use the all women's education factor for just that variable and the total for the other variables in the regression?

Thanks for your assistance !

Vanessa]]>

I think this may be getting even more complicated than is really necessary. The all-woman factors are needed when you are trying to estimate something for all women, but you have to work with ever-married women because that's all you have in the sample. An example is when you want to estimate a fertility rate for all women but you only have the births and exposure for ever-married women. You have to assume that (a) never-married women have no births and (b) awfact/100 is a multiplier to inflate exposure for ever-married women to exposure for all women. What kind of a logit regression are you doing for which this distinction is relevant? What is the binary outcome variable? Can you assume that all of the never-married women would have the same specific outcome (either 0 or 1) if you could measure it?

The all-woman factors are calculated for categories such as all urban women or all rural women. Those categories/variables must be present in the household survey. In a multivariate analysis, which your use of logit regression suggests, the factors may not be prepared in advance. If you want to do something complicated, it will be safest to bypass the all-woman factors all together and just say that you are working with ever-married women. You may have to accept that it will be difficult and maybe impossible to extrapolate to all women. If the officials in a country wanted to produce estimates that apply to all women, then they should have interviewed all women.

]]>

Thanks in advance

]]>