I am using the 1-year 2014 ACS data. I have taken a subsample where I only keep Federal Government Employees (CLASSWKRD==25). The weighted sample size using PERWT is over 5 million, and I get the same with the online data analyzer. According to OPM (https://www.opm.gov/policy-data-overs…), there were approximately 4,185,000 Federal Government Employees in 2014.
I assume this discrepancy is due to the sample weights not being intended to sum correctly to the population estimates for class of worker categories (unlike how a single state subsample would have the correct weighted N). Is that correct? Is there something that I should do to correct this discrepancy?