| Dropping observations/weights [message #11891] |
Sat, 25 February 2017 19:39  |
catherino222
Messages: 4
Registered: February 2017
Location: London ON
|
Member |
|
|
Hello,
I am new to STATA and DHS. I'm pooling Cameroon DHS 2004 and 2011 datasets. However, ethnicity is coded differently. The final reports have the same response options for ethnicity in 2004 and 2011 but the datasets have different options.
1. I would like to know how ethnicity for 2004 was recoded (grouped) to produce the final options that match the 2004 final reports and 2011 dataset
2. I'm looking at determinants of maternal mortality and dropped nulliparous women using V201. Is this a problem if I drop some observations? I have seen some comments about not dropping observations but they were not clear why this should not be done. Can you please clarify?
3. Also, must weights be de-normalized? The sampling frame for Cameroon were from 2005 for the CDHS 2011 and from 2002/2003 for CDHS 2004. I don't know if these make a distinct difference looking at the population sizes for both years(2002/2003 and 2005) which are close but I was considering using the weights as they are. I assume this will mean the 2011 dataset will have more weight than 2004. I would appreciate your input.
4. Lastly, can you provide information on how maternal mortality ratio was derived?
5. For v467d, 7964 of 15426 respondents had NOT applicable. Is there information as to why there was such a large number for NA?
Thank you very much
Catherine
|
|
|
|
| Re: Dropping observations/weights [message #11894 is a reply to message #11891] |
Mon, 27 February 2017 09:56  |
Bridgette-DHS
Messages: 3203
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS Stata Specialist, Tom Pullum:
There have been many forum posts on "pooling". You can put both surveys into a single file, with a variable that distinguishes one survey from another, and you can look at differences between the two surveys, but I would never recommend that you treat them as a single survey.
I am getting some help with the question about how the more detailed codes for ethnicity (v131 in the IR files) in the 2004 and 2011 surveys were mapped into the six main categories. Will post that separately.
Adult and maternal mortality rates and the maternal mortality ratio are calculated from the sisterhood module. They cannot be calculated from the respondents themselves because if a woman has died she cannot be in the survey. It would not be desirable to drop ANY woman who has any sisters. I can't think of a reason why you might even consider doing that.
There has been endless forum discussion on renormalizing the weights. If you are looking at differences or changes between the two surveys, then you use the weights exactly as they are in the data. Your question is only relevant if you want to combine the two surveys for a single estimate. I would not recommend doing that, but if you want to do it, the steps have been posted.
For a general description of the maternal mortality ratio and rate, see this WHO report: http://www.who.int/reproductivehealth/publications/monitorin g/maternal-mortality-2015/en/. For a description of DHS procedures, see this Methodological Report: https://www.dhsprogram.com/pubs/pdf/MR13/MR13.pdf.
For all of the v467 options (a through m) in the 2011 survey, there are 15,426 NA cases ("."). These would be women for whom the options were not relevant, perhaps because they had no need to seek medical help in a reference time period. The general strategy when you see this is to look at the questionnaire and find the filter that determined whether or not these questions would be asked. In some cases there may be subsampling, such that only a fraction of the women will be asked certain questions, such as domestic violence questions, but I don't think this was such a survey.
|
|
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
