When calculating nationwide employment among people aged 25-54, the values I get for the 1980-2000 census samples are almost exactly in line with CPS-derived estimates, found here for comparison Employment Level - 25-54 Yrs. (LNS12000060) | FRED | St. Louis Fed . In 2005, the ACS file suggests a sudden roughly 5% increase relative to the year 2000, and from there on exhibits the exact same trend as the CPS-derived series. I do not see anything in the PERWT documentation that would suggest that this would happen- did I miss something?
I see that the estimates of the number of employed people age 25-54 do not match very well between the ACS and CPS (estimates from FRED) from 2000-2005, compared to 2006-onward, and compared to the match between decennial census and CPS estimates.
The ACS is different from the decennial census, and in particular the earliest years of the ACS saw substantial variation in survey methodology. The first several years of the ACS, 2000-2004, were more preliminary runs of the ACS. The 2005 ACS was the first full 1% sample of the U.S. population. There were significant changes to the layout and wording of the ACS questionnaire between 2002 and 2003; between 2004 and 2005; and between 2007 and 2008. There were changes to the weighting methodology in 2006 due to some issues with weighting in the 2005 ACS. These changes could be relevant.
Additionally, the ACS is conducted on a rolling basis throughout the year, and estimates do not account for month of collection—the CPS estimates are seasonally adjusted. I would expect to see variation in how closely ACS estimates match CPS estimates across those first several years of ACS samples. I would also expect a difference between the CPS-ACS match and the CPS-census match. You can read more about the ACS, especially some important dates in the early years, on this page: What is the ACS? The Census Bureau also provides this helpful page on comparing ACS data to other data.
To ensure the populations you are studying are comparable, you should exclude from your analysis of ACS data people in the armed forces (EMPSTATD=14 or 15), and people living in group quarters (GQ=3 or 4; group quarters were sampled in the ACS beginning in 2006).