Home » Topics » Mortality » Preceding Birth Inteval and child mortality
| Re: Preceding Birth Inteval and child mortality [message #854 is a reply to message #851] |
Mon, 21 October 2013 15:04   |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
Hi,
That's a pretty big question, but I can give a few hints that I've picked up working with analyzing mortality rates.
First, it sounds like you have the woman/individual recode to work with. For me, the birth recode is the right thing for mortality, because the observations are by child, not by woman (so no, like, B06_01, B06_02, etc... just B06 for each kid).
Second, be sure you drop all children who have not yet lived through the "exposure period." So if you are looking at U1 mortality, drop all kids born less than 12 months before the survey (you don't know who will be alive/dead by the time 12 months are up...well, you know some will be dead, but...).
Third, your choice of which birth interval variable to use shouldn't matter too much, but I'd say the imputed one (which should just have more/more precise lengths) makes sense. The question is whether there is some systematic problem where imputed birth intervals for, say, dead children are under/over estimated. I think that is probably a secondary concern here.
Fourth, pick some cap year, meaning maybe don't use every birth to every woman, just all after some date - some of these birth recodes will be from the 1970's or 80's.
Fifth, I would think that estimation strategy shouldn't matter too much here in a simple framework like this: you could use a linear probability model (OLS), probit, or some hazard function. They should all be similar. See the FAQ on the measureDHS site about weighting and standard errors.
Hope some of that helps, even if it isn't the exact answer you were probably looking for.
|
|
|
|
Current Time: Tue Aug 12 18:10:40 Coordinated Universal Time 2025
|
Members
Search
Help
Register
Login
Home
