I am computing std errors using replicate weights for an ACS2016 extract using R. I noted that the PERWT sum to what I assume is the population total. However the sum of each replicate weight columns does not sum to anything even close. When I compute a statistic like population total, or even for a subpopulation (say male or female) the total using PERWT seems correct. However using each of REPWTPn gives an answer that is orders of magnitude smaller. For example the PERWT for my extract sum to 330799752. RPLWTP1 sums to 2232926. If I’m looking at say an estimate of population totals for women I will get vastly different extimates, When I take the squared differences I get a very large number that is certainly not correct. Am I missing something here?

Thanks.

Anne