Discrepancies between ACS 5-year data from IPUMS vs. summary stats from AmericanFactFinder


I’ve been trying to supplement tables from ACS 5 year data (EEO tabulation) for 2006-2010 with ACS 5-year data for the same period from IPUMS. I’ve been flustered for about 24 hours since there seem to be considerable discrepancies in summary stats between the two sources, which makes me wonder if I’m doing something wrong.

For example, ACS 5 year data estimate (for the 2006-2010 period) available via AFF for the number and distribution of janitors (occ code 4220 in 2010) is the attached screenshot, showing a total of 2.6 million with 68.3% male and 31.7% female.

On the other hand, using perwt as weights, the ACS 5-year data for the same period indicates a slightly different total:

. tab sex if occ2010==4220 & datanum==5 [fw=perwt]

Sex | Freq. Percent Cum.
Male | 2,185,828 67.52 67.52
Female | 1,051,432 32.48 100.00
Total | 3,237,260 100.00

I’m wondering what accounts for the discrepancy between them (wrong weights?) Given the differences, it seemed a bit unwise to move on to the more finegrained data before making sure that I’m looking at the same apples.

Thank you!

It is difficult to know for sure what is causing this discrepancy. My best guess is that the American Fact Finder table you show here is defining the universe in a more restrictive way than you are when using IPUMS USA data. For example, in many of these tables the universe includes only those who are in the labor force (i.e. LABFORCE==2). Including this filter I find estimates that are relatively close to the figures reported in the American Fact Finder table. Finally note that, although we expect to calculate estimates within a given margin of error, we do not generally expect to exactly replicate officially published statistics with public use microdata. This is due to various confidentiality restrictions placed on public use microdata.

I hear you. That’s a frequent issue I have for which no one can unequivocally answer. Very frustrating.