This R code computes exact Confidence Intervals for any Complex Survey under any subpopulation regarding the presence of missing data. This script replicates subpop command from Stata, based on the research presented in [1].
Many practitioners usually oversee this issue, forgetting that the computation of adequate Degrees of Freedom (df) for Confidence Interval's t-distribution must follow special rules. And, using a normal distribution or a gross approximation of df could mislead inference.
As mentioned in [1] this problem is still a matter of research. As far as I studied the literature, there aren't following investigations about this topic.
As far as I know, there isn't an R implementation about this. So, I hope someone finds this useful.
[1] West, Brady & Berglund, Patricia & Heeringa, Steven. (2008). A Closer Examination of Subpopulation Analysis of Complex-Sample Survey Data. Stata Journal. 8. 520-531.