Follow up to Question on Subsample Weighted N being larger than Group's Population

This is a follow up to the question answered at Weighted N for Federal Government Workers Subsample is Bigger than that Group's Population (using IPUMS-USA 2014) I filted the subsample down to Federal Government Employees (CLASSWKRD=25), who are citizens (CITIZEN=0-2), at work (EMPSTATD=10), and age 16 or older (AGE>=16). This leaves a sample size of 36,029, with a Weighted N of 3,555,779. This is still much larger than the number of U.S. Federal Government employees, 2,726,000 (not including armed forces as they would be filtered out by EMPSTATD=10). I would think that filtering to citizens should eliminate the vast majority of people who are not working for the U.S. Federal Government.

Can you think of other reasons why the number would be so far off? Do you know if government contractors count as government employees in the ACS? I am about to ask OPM if they count contractors in their totals. If you can’t think of another reason for the inconsistency, would you recommend simply footnoting this inconsistency, or weighting down the observations to match the population?

Thanks in advance for your help

It is difficult to say whether government contractors count as government employees as this question was answered by the respondent according to their own understanding and interpretation of the question. The exact text and instructions are available in the CLASSWKR Questionnaire Text. The fact that this data was collected from individuals while the OPM data was collected from the agencies is likely a large source of the discrepancy between the figure derived from the microdata and the OPM table. I believe noting this as a possible source of error is sufficient but you could attempt to reweight as well to see the degree to which this effects your results.

I am sorry I could not provide a more satisfying explanation.