| Re: Contraceptive prevalence rate [message #30539 is a reply to message #30534] |
Tue, 17 December 2024 07:20  |
Bridgette-DHS
Messages: 3214
Registered: February 2013
|
Senior Member |
|
|
Following is a response from Senior DHS staff member, Tom Pullum:
Whenever I see a substantial number of cases with a dot (for Not Applicable or NA) on a key variable, I enter "lookfor select" to see if there is a selection variable for a subsample. One of the variables that comes up in this survey if you do that is "seligbm":
seligbm byte %8.0g SELIGBM hh selected for biomarker and long/short woman qre
which has the following distribution:
hh selected for biomarker and |
long/short woman qre | Freq. Percent Cum.
----------------------------------------+-----------------------------------
household selected for biomarker and fu | 10,053 33.42 33.42
household selected for full woman quest | 9,934 33.03 66.45
household selected for short woman ques | 10,091 33.55 100.00
----------------------------------------+-----------------------------------
Total | 30,078 100.00
This sort of thing is one of the hazards of using DHS data and the GitHub programs. Someone could say that analysts should read about the sampling design before starting to use the data but we (you and I!) usually just plunge in and then find these exceptional features the hard way. The good thing is that you tried to calibrate your estimate against the report and found a discrepancy, and it led to the evidence of subsampling. That's good practice. Subsampling is one of the main reasons why users cannot match reports.
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
