ACS Data Users Group

I seem to recall reading somewhere that the Census Bureau periodically revises past estimates based on their collection of new data; that is, older estimates from previous years may be changed by a few tenths of a percent based on more recent data analysis. Is this something the Census Bureau does or am I just imagining it? I only ask because there are some estimates I looked at in January 2019 that, when I look at the same estimates today, appear to have changed by a few tenths of a percent. Any help would be greatly appreciated!

If you're referring to Census's Pop Estimates Program (PEP), then Yes. With each new PEP vintage, the whole time series back to most recent decennial census is revised. You can find all the releases archved here www2.census.gov/.../

You're also correct that the revisions in vintage 2018 (published April 2019) were more noticeable than in some years.

I had a discussion with a manager in Census Population Division about the confusing nature of what's published on factfinder.census.gov. The issue is:
PEP considers the most recent vintage to be the best suited for use. (That is vintage 2018, data points for 2011, 12, 13, etc.)
But Factfinder queries for past years show the unrevised numbers, different from what is in the Excel files.

Population Division's answer to me was: (1) They are aware of the discrepancy. (2) Census's standard procedure for FactFinder is: After numbers are published in FactFinder, those numbers stay in FactFinder. So what a web user would see displayed is the vintage 2011 data points for 2011, vintage 2012 data points for 2012, etc.

Hope that helps.