Weights when pooling data

[JoseAntonioOrtega]

First of all, congratulations for a great package make life simpler for people working with DHS data.

I have some issue with how to pool the data. In your vignette https://cran.r-project.org/web/packages/rdhs/vignettes/introduction.html you are providing a useful tutorial, but it has problems in the data pooling part.

DHS weights are normalized weights with a mean of one. This means that if you use hv005 "as is" for pooled data analysis, you are giving more weight to surveys with more data, not necessarily a good idea. In fact, the DHS recommendation on how to pool data is included in the DHS sampling manual https://dhsprogram.com/pubs/pdf/DHSM4/DHS6_Sampling_Manual_Sept2012_DHSM4.pdf , p.28. The suggested approach is to de-normalize and renormalize according to the relevant measure (eg: country's population, ...).

It would be a great addition to provide the possibility of renormalization of weights getting the data for renormalization from statscompiler.

For the moment it would be good if the vignette shows how can it be done, or at least provides a caution note that the use of single survey weights is not advised for pooled survey analysis.

Thanks!

Session Info