| Children ever death and children ever died [message #30404] |
Sun, 24 November 2024 07:51  |
Tesfay
Messages: 7
Registered: April 2021
|
Member |
|
|
Dear DHS experts, first I acknowledge the support you give to researchers, with a special thanks to Tom Pullum.
As part of my PHD work, I am looking on the effect of fertility rate on child mortality. I plan to follow to approaches:
1) I regress the probability of survival (b5) of under-five children on the number of children born in the last five years using the KR file.
2) I am trying to see the effect of the number of children ever-born (v201) on the number of children ever-died (v206+v207) using simultaneous equation modeling (Due to the simultaneity and endogeneity of the variables). I am using Ethiopia DHS (pooled from 2000 to 2019) and I understand that the IR file is the correct unit of analysis for the 2nd approach.
But I also need to control for other variables (b4, b0) and women's characteristics (maternal age at delivery (b3-v011), place of delivery (m15a) for all the children ever born, which are found in the BR file. So, please guide me on the ff issues.
1. Is my analysis plan and unit of analysis theoretically sound?
2. How can I get the variables in the BR file in the IR; I tried to merge using 1:m from BR to IR but it showed me an error message "variables caseid, v001, v002 do not recognize . . . "
3. Does it make sense to do the analysis using the BR file if it is impossible to merge?
Excuse me for asking conceptual questions, it is due to my lack of experience with DHS data.
[Updated on: Sun, 24 November 2024 07:57] Report message to a moderator |
|
|
|
| Re: Children ever death and children ever died [message #30427 is a reply to message #30404] |
Wed, 27 November 2024 11:46  |
Janet-DHS
Messages: 893
Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS staff member, Tom Pullum:
Thanks for the thanks!
There is no need to merge the IR and BR files. The BR file has one record for each birth in the woman's birth history. That record includes most of the mother's variables from the IR file, including the summary variables v201-v209. The IR file has one record for each woman (whether or not she ever had any children). It includes all of the b variables with subscripts. For the most recent birth, for example, b4_01 is the sex of the child and b5_01 is the survival status. I believe the question is which of these files do you want to use, NOT how do you merge them.
I'm not sure how you would apply SEM to the data. I suggest something a little different. Your research question could be, for example, "Does having a child death increase the probability of having another birth?" It could be that women (or parents) try to replace a child who has died. The probability could change over time and could depend on the sex of the child who died or the current sex composition. If this is what you are thinking of, please let us know and I can suggest a way to do it.
Another strategy with repeated surveys is to take a cohort perspective. For example, women born in 1980 (Gregorian calendar) will appear in all of the surveys, at different stages of family building. You can re-organize the data into a quasi-longitudinal structure. I do this sometimes when checking whether successive surveys in the same country are consistent with one another.
For SEM there is the crucial limitation that you only have a sequence of independent cross-sections and almost all the variables describe current status. There's not much retrospective information outside of the birth histories. Rather than jumping into a complex analysis, I strongly suggest that you start the analysis with a simple approach, using crosstabs and regressions, and then add complexity to the extent that it is possible and necessary.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
