Why isn't the ACS 2014-2018 (or others) included in time series tables?

Hey guys, first of all the time series tables have been a huge help, so thank you for putting everything together! I have 2 related questions:

1. I’m just curious, why are there only 2 ACS 5-Year surveys included in time series tables? (They are the 2008-2012 and 2015-2019 tables.) Why not include all of the 5-Year ACS surveys?

2. I’m trying to compare population between census tracts for all years from 2010-2020 and wanted to use the 5-Year ACS to do that. If the time series tables only includes 2008-2012, 2015-2019, as well as the decennial 2010 and 2020 census, how do you recommend I go about standardizing all 2020 tracts to years 2010-2019?

  1. We plan to extend the time series tables to include all* 5-year ACS periods, hopefully soon. (*There are some complicating idiosyncrasies in data from 2005-2009, the first 5-year release, so we’ll probably continue to omit that period even after adding all others.) Our main concern is to avoid “overloading” the tables with much more content than most users need. Our system currently doesn’t allow users to select specific years for time series, so if a table included all ACS 5-year periods, a user would have no choice but to get all of them (11 periods now and more in the future). We expect most users would prefer a smaller set, especially because overlapping ACS 5-year periods are based on overlapping samples–e.g., the 2015-2019 and 2016-2020 periods include the exact same sample for 4 of their 5 years–so it’s very difficult to draw meaningful interpretations from comparisons of overlapping periods. We hope to enable year selection soon and then extend the tables after that, but we may still consider other strategies to include all periods even sooner than that.

  2. Using 5-year ACS data to create annual population estimates is problematic because the 5-year estimates don’t correspond directly to individual years, and as noted above, differences from one 5-year period to the next don’t effectively represent annual population changes due to the large overlap in samples. As an alternative, for a special release of annual 1990-2009 tract population data, we started with our standardized decennial tract-level time series, interpolated linearly to generate an initial set of annual estimates, and then adjusted the estimates within each county to match with annual county-level estimates from the Census Bureau’s Population Estimates Program. Our tract estimates page provides more detail on our method. We hope to extend this series to 2020 in the future, but we haven’t scheduled that work yet. In the meantime, you might try applying the same method yourself.

This is very insightful, thank you!