Why dont the svyset above take account of post-stratification?
It is possible to split the final weight variable (e.g. ESTWTNR) into two separate components: that part which accounts for the use of unequal sampling fractions and that part which post-stratifies the achieved sample to match the profile of the population (see Section 7 of the WERS 2004 Technical Report for a discussion of how the weights in WERS are derived). It should then be possible to specify these two components separately on the svyset command, using the pweight and postweight() options. However, this is only feasible when estimating on the full sample, since population totals for specific sub-samples are not known. Additionally, in our trials we have found it difficult to get Stata to weight even the full sample properly under this arrangement. In practice, the reduction in standard errors that is theoretically achievable by separately specifying the post-stratification element of the sample design is relatively small: in our trials it altered standard errors at only the 3rd d
