As has been noted in other posts, we are unable to separate out the probabilities of selection for the cluster (v001 or v021)) and the household (v002). What we have is the product (the second probability is conditional on the selection of a cluster). There is no way to separate the two. That's unfortunate, and it means that svyset cannot be properly constructed for multi-level models. The basic reason for this is that if the separate probabilities were available, then the first one, with a little additional information, would allow one to identify the PSU. DHS does everything possible to protect the privacy and confidentiality of the respondents. Apart from that, the required information is deleted after the weights have been constructed. We couldn't retrieve it even if we wanted to.