How to handle missing data?

[Messu Melu]

Hi all,
I am investigating health disparities in some Sub-Saharan Africa countries. Ethnicity is one of my variables but missing for about 50% of the sample. However, it is not missing for most household hheads and their spouses, and a few children. Is it a good idea if I replace the missings with that of household heads or their spouse's in lieu of that than doing multiple imputations? The family structure in Africa, particularily in rural areas, is extended family structure that household members are highly related.
Any advice is appreciated!
Thank you!

[Carlo Lazzaro]

Messu:
your approach can be a good idea if those values were missing for sake of brevity in filling out the questionnaire (ie, husband's reply implies that all the household members share the same ethnicity).

[Messu Melu]

Hi Carlo,

I am not sure if you have ever done with Demographic Health Survey (DHS). I used DHS data sets in which the data are recoded in up to 6 recodes for each country. Some recodes such as individual women recode, birth recode and kids recode have some variables like variables about anthropometric info, child mortality, nutrition, contraception, ethnicity and etc that the recode for all individual household members of the selected households do not have. Thus, there were different questionnaires for each recode. That is why I just tried to extract such variables (by merging from the various datasets) which are actually the subsets of the recode for all members in the selected households(PR). My assumptions came from the data structure.

[Carlo Lazzaro]

Messu:
thanks for providing further clarifications.
You are correct: I'm not familiar with such data.
However, I think that the main issue still remains the same: if you're sure that the messingness is (I guess here) MCAR, you can follow your approach.

[Messu Melu]

Sure, it looks MCAR type of missing.
Thank you very much for taking time to respond to my questions!