ASEC vs. Census CPS numbers

I’m having some issues replicating the Census CPS numbers using the IPUMS ASEC samples. I’m using this report, see pg. 35-36 “All Races”.

For example, if I do a simple comparison of the number of observations, for 2005, the number of households is 114,384 thousand. Using the 2005 IPUMS ASEC, if I keep 1 observation per household and sum up the ASECWTH/1000, I get 113,475,724 which does not match exactly.

Am I using the weights incorrectly, or is there a reason for this discrepancy?

Edit: I have also come across some negative ASEC weights. These have come up in the 1969 files (unsure about others) see SERIAL == 2467 or 3397 or 22990. I was reading in the .dat file manually, so possible that it has been read in incorrectly but I think this is unlikely.


According to Income and Poverty in the United States: 2019 (the report that you linked to), the numbers are based on the number of households as of March of the following year (see the sub-heading at the top of page 35 of the pdf). To check that this solves your problem, I compared the number of households in the 2006 IPUMS CPS ASEC (114,510,050) with the number of households in 2005 according to the above-mentioned CPS report (114,384,000), which is a much closer match.