| Child Birth Weight - Not Recorded Data [message #24195] |
Fri, 11 March 2022 16:42  |
Aamna
Messages: 11
Registered: May 2019
|
Member |
|
|
Hello,
I need help in understanding how to handle a large percentage of data missing for the birth weight of children (m19). I am trying to find if access to private or public facilities impacts children's birth weight. I am looking at the children's data file. The Pakistan data shows that around 63% were not weighed at the time of birth. I have three queries with regards to it.
1) Do I consider these observations as missing?
2) If I consider them missing, do I drop the missing (not recorded birth weight) observations to do my regression analysis?
3) If I do not drop these observations, is it feasible to create a binary coded new variable with birth data as 1 and no birth data as 0 and then run my regression analysis? Or should I do something else in which I need your guidance on it?
I appreciate your input on my queries. I need help to understand the whole situation. Thank you!
Sincerely,
Aamna
|
|
|
|
| Re: Child Birth Weight - Not Recorded Data [message #24203 is a reply to message #24195] |
Mon, 14 March 2022 11:11  |
Bridgette-DHS
Messages: 3203
Registered: February 2013
|
Senior Member |
|
|
Following is response from DHS Research & Data Analysis Director, Tom Pullum:
I assume you are talking about the latest Pakistan survey, in the PK71 files. Yes, most children were not weighed at birth, including almost all of the home deliveries. For "not weighted at birth" m19 is coded 9996. There is also a code 9998 for "don't know". Those codes must be excluded from any analysis, by including "if m19<9996" in the estimation command or (as you say) you can construct a binary code that is, say, 1 for usable values and 0 otherwise, and then include something like "if condition==1" in the estimation command.
I have to say that I did a quick check of the difference between public and private hospitals in this survey, in terms of birthweight, and there is so much dispersion in weights that the means appear not to be different. However, I encourage you to proceed. It's an interesting research question.
|
|
|
|
| Re: Child Birth Weight - Not Recorded Data [message #24205 is a reply to message #24203] |
Mon, 14 March 2022 18:52 |
Aamna
Messages: 11
Registered: May 2019
|
Member |
|
|
Bridgette-DHS wrote on Mon, 14 March 2022 11:11
Following is the response from DHS Research & Data Analysis Director, Tom Pullum:
I assume you are talking about the latest Pakistan survey in the PK71 files. Yes, most children were not weighed at birth, including almost all home deliveries. For "not weighed at birth," m19 is coded 9996. There is also a code 9998 for "don't know." Those codes must be excluded from any analysis by including "if m19<9996" in the estimation command or (as you say) you can construct a binary code that is, say, 1 for usable values and 0 otherwise, and then include something like "if condition==1" in the estimation command.
I have to say that I did a quick check of the difference between public and private hospitals in this survey regarding birth weight, and there is so much dispersion in weights that the means appear not to be different. However, I encourage you to proceed. It's an interesting research question.
Hello,
Thank you for the reply. I am worried about skewed analysis as I have less than 20% data. I am thinking of creating a binary variable for the analysis. Hopefully, I will get some solid results from it. I understand your point about dispersion in weight; however, I will perform my analysis to explore this area. Thank you for the support.
I want to request that you kindly guide me towards the women's empowerment index, as this is the second part of my research. I want to read more about the index and factor analysis done in the DHS regarding it. I appreciate your help.
Thank you
Sincerely,
Aamna
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
