Comparing across 1980, 1990, 2000 Census and 2006-2008 3 year ACS


I am currently using 1980, 1990, 2000 Census, and 2006-2008 3 year ACS to build up the database. But it seems that Census and ACS database are quite different in terms of sample representativeness even after I adjust with “perwt” variable.

For example, if I simply sum up the population and employment within year after adjusting with “perwt” variable, I am getting the following number:

year 1980: employment 4,949,357; total population 1.13e+07

year 1990: employment 5,902,657; total population 1.25e+07

year 2000: employment 6,546,089; total population 1.41e+07

year 2008: employment 4,293,972; total population 8,965,060

Employment is based on “empstat” variable. I expected larger number of employment and total population for the year 2008, but number gets smaller. That is, it seems that ACS and Census are not compatible even after I adjust with “perwt” variable.
Do I need to further adjust the ACS database to make it compatible with Census database? Your help would be greatly appreciated!

The numbers you report are unweighted. Using your extract #13 and weighting your samples by PERWT, I get the following values for EMPSTAT=1:

1980: 98,987,140
1990: 117,161,898
2000: 130,869,287
2008 3-yr: 144,283,042

As you can see, the estimated number of employed persons in the U.S. increases over time, as you expected. The syntax for generating weighted counts varies depending on the statistical software you are using. In Stata, try the following: tab year empstat [fweight=perwt].

Hope this helps.