The DHS Program User Forum: Mortality » Working through Odds Ratios

Home » Topics » Mortality » Working through Odds Ratios (problem in b6 and b7 variable in BR file)

Show: Today's Messages :: Show Polls :: Message Navigator

Working through Odds Ratios [message #29067]

Fri, 19 April 2024 07:08

shivani
Messages: 12
Registered: March 2021

Member

Hi,
I am want to calculate the odds ratio for NNMR, IMR, CMR across social categories for which I am using the variable b6 and b7.
However, there is some error I believe in data as the age at deaths (captured by b6 and b7) shows values greater than 60 months (in the variable b7).
Approx 8000 observations in b7 variable are greater than "60".
Can someone help me with that?
additionally, I hope my approach is correct in which I am only including those observations with b7==0 if the child didn't survive first three years as "having incurred neonatal mortality" in comparison to a live child of age 1 month?

Thanks

Report message to a moderator

Re: Working through Odds Ratios [message #29076 is a reply to message #29067]

Fri, 19 April 2024 17:09

Janet-DHS
Messages: 880
Registered: April 2022

Senior Member

Thank you for submitting your question. Could you please provide some more detailed information so we can better advise you?
Can you let us know:
• Which survey you are using (Include country name and year)?
• Which data files you are referring to?
• Which software you are using (Stata, SPSS, R, etc.)?
If you are trying to match a Table in a final report, please also indicate which table and which estimate you are trying to match.

Report message to a moderator

Re: Working through Odds Ratios [message #29078 is a reply to message #29076]

Sat, 20 April 2024 03:09

shivani
Messages: 12
Registered: March 2021

Member

Which survey you are using (Include country name and year)?Hi, Thanks for the reply. I am using DHS India survey dataset for the latest round 2019-21.
I am referring to BR file (the variable b6 and b7 are also included in the KR file though). Software that I am using is Stata, and I am not trying to match the table. I am trying to calculate odds ratio for nnmr, ima and car, across social groups in India. For that, I was creating a variable for "whether a child survived for 30 days". In order to do so, I was referring to b6 and b7 variable which tells one about the age of the child at death (b7 is imputed age in months).
The the variable has higher values than "60", which is odd. Values are as high as 335 months. Attaching the screenshot for your reference.

Attachment: Screenshot 2024-04-20 at 12.39.09?PM.png
(Size: 1.67MB, Downloaded 60 times)

Report message to a moderator

Re: Working through Odds Ratios [message #29111 is a reply to message #29078]

Wed, 24 April 2024 17:29

Janet-DHS
Messages: 880
Registered: April 2022

Senior Member

Following is a response from DHS staff member, Tom Pullum:

If you search the forum you will find many related exchanges. For your purposes you can ignore b6, and just use b7, which is months of age at death for children who died. However, for ages 2+ years, the months are constructed as years x 12. Thus, for "2 years" the number of months is "24". Then the only codes should be 36, 48, 60, 72, etc. I don't know where a value of "335" would have come from....

The NMR, IMR, CMR. U5MR are rates for synthetic cohorts, following life table procedures. Except for the factor of 1000, the IMR is 1q0, the CMR is 4q1, and the U5MR is 5q0. Rates refer to aggregates and can only be calculated for an aggregate. You are apparently looking for an individual-level analog. You can certainly construct binary (0/1) variables corresponding to age at death. For example you could initialize d_nmr=0, d_imr=0, d_cmr=0, and d_u5mr=0 and then construct d_nmr=1 if b7=1, d_imr=1 if b7<12, d_cmr=1 if 1<=b7<60, and d_u5mr=1 if b7<60. The only problem is how you then deal with censoring,

The DHS approach can calculate the U5MR for deaths during the past 5 years, because of the synthetic cohort approach, but children with full exposure to the risk of death in the past 5 years were actually born 5-9 years ago. Please look at the Guide to DHS Statistics, DHS reports on mortality, or other literature. The bottom line is that there is no easy way to translate rates for aggregates into individual-level indicators, which is what you want to do.

Report message to a moderator

Re: Working through Odds Ratios [message #29136 is a reply to message #29078]

Mon, 29 April 2024 05:50

shivani
Messages: 12
Registered: March 2021

Member

still facing problem in working with the data. I am dealing with the same BR file, and looking at the variable b8 (age of child in years), the range is between 0-40 years. I also tried to recheck by creating the age variable on my own. Seeking help from the recode file, I created the variable of "age of the child in months" by using the following commands:
gen age2= (v008a-b18)/30.4375
gen age3 = round(age2)

the problem remains intact. How is children aged more than 60 (when measured in months).
Attaching the screenshot for your reference

PS: I downloaded the dataset from DHS website only

Attachment: Screenshot 2024-04-29 at 3.18.33?PM.png
(Size: 1.54MB, Downloaded 59 times)

[Updated on: Mon, 29 April 2024 05:50]

Report message to a moderator

Re: Working through Odds Ratios [message #29178 is a reply to message #29136]

Fri, 03 May 2024 15:03

Janet-DHS
Messages: 880
Registered: April 2022

Senior Member

Following is a response from DHS staff member, Tom Pullum:
I entered the following two lines:

gen test=int((v008a-b18)/365.25)

correlate test b8

The correlation is exactly 1, so there is the kind of consistency you are looking for. I believe your problem was just that you used round() instead of int().

Report message to a moderator

Previous Topic:	Birth selection
Next Topic:	determinants of neonatal mortality

Goto Forum:

-=] Back to Top [=-

[ Syndicate this forum (XML) ] [

] [

]

Current Time: Wed Nov 13 18:09:52 Coordinated Universal Time 2024