Home » Countries » Nigeria » NGHR7BFL (NDHS 2018)
|
|
|
|
|
|
Re: NGHR7BFL [message #25127 is a reply to message #24993] |
Thu, 01 September 2022 16:34   |
Janet-DHS
Messages: 938 Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS Research & Data Analysis Director, Tom Pullum:
You are asking questions that go beyond DHS data and therefore beyond the scope of the forum.
DHS has very low levels of "missing" data. A blank or dot in a DHS data file should be interpreted as Not Applicable (NA). If you are thinking of "missing" as "don't know" or "refused" or something like that, we use special codes such as 8, 9, 9994, etc., depending on the variable. The frequencies of those codes are usually very low.
In general, to test whether "missing" is random with respect to some potential covariate, you construct a binary variable that is 1 if "missing" and 0 if "not missing" and do a logit regression of that variable on the covariate, to see whether there is a statistically significant relationship.
|
|
|
|
Re: NGHR7BFL [message #25360 is a reply to message #25350] |
Wed, 12 October 2022 09:37   |
Janet-DHS
Messages: 938 Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS staff member Tom Pullum:
Beginning with DHS-7, most surveys include hv270a in the PR file, v190a in the IR, KR, and BR files, and mv190a in the MR file. The "a" indicates that the wealth quintiles are residence-adjusted, i.e. calculated separately for urban and rural areas. A problem with the original, unadjusted wealth quintiles is that, in most surveys, there are very few households in the top quintile in rural areas and very few households in the bottom quintile in urban areas. If you use the unadjusted wealth quintiles in a model, much of the information is actually an urban/rural distinction. If you use the unadjusted wealth quintiles in a model, AND include urban/rural (hv025, etc.) then you have a better separation of wealth and residence, but the model may run into estimation issues because there are (typically) so few cases in the two combinations I mentioned.
Bottom line: if your model includes urban/rural, which it probably should, then you may want to use the adjusted wealth quintiles rather than the unadjusted. But there's no law saying you have to do that. It would be good to tell the reader which version you are using.
|
|
|
|
Re: NGHR7BFL [message #25602 is a reply to message #25585] |
Wed, 16 November 2022 09:10   |
Janet-DHS
Messages: 938 Registered: April 2022
|
Senior Member |
|
|
Following is a response from DHS staff member Tom Pullum:
The best variable for this purpose is hv000 (in Stata, HV000 in SPSS). That variable only takes one value, the string "NG7". If you tab that variable, without weights, you get 188,010 cases in the PR file, i.e. individuals in the household survey. If you do the same thing in the HR file, which has households as units, you get 40,427 households.
|
|
|
|
|
|
|
|
Goto Forum:
Current Time: Tue May 13 07:31:35 Coordinated Universal Time 2025
|