I’m looking at family income distributions in MSAs using 1990 census, 2000 census, and the 3-year ACS surveys. In Stata, I created a single record for each family by identifying duplicates based on serial and famunit and dropping the duplicates. I coded 9999999 to missing. To make sure the income distributions are correct, I tried to compare the weighted mean and median family income in MSAs with the census estimates found in Social Explorer (http://www.socialexplorer.com) and American Factfinder (Social Explorer has both mean and median family income, but I found only median family income in American Factfinder).
The results are confusing. When I manulaly checked a few MSAs in 1990, 2007, and 2011, I found that the median and average family income is consistently off from the census estimates (off by 10,000 or 20,000). However, the mean and median household income looks much more reasonable (the difference from the census estimates is within a few thousand/hundred dollar).
Is this a unknown phenomenon with family income in the microdata? Is it because I’m using household weight to estimate family income? Is there anything that I did wrong?
Thank you so much!