American Community Survey Data Users Group
ACS Data Products & Resources
ACS Public Use Microdata Samples (PUMS)
ACS Summary Files
Application Programming Interface (API)
ACS Data Issues
ACS Data for Small Geographic Areas & Population Subgroups
Mapping ACS Data/GIS Applications
Measuring Trends Over Time with ACS Data
Measuring Trends Over Time…
Comparing 2005-9 ACS 5 year…
This group requires membership for participation - click to join
Joe D. Francis
Jan 15, 2016 3:19 PM
Comparing 2005-9 ACS 5 year with 2010-14 ACS 5 year
I am seeking advise on how to handle the issue of different population controls used in weighting the ACS 2005-09 period estimates from those used in weighting the ACS 2010-14 period estimates. The Census Bureau notes this as a special problem in making these comparisons which have different base populations. Considering that, what have others done to make this adjustment before making comparisons of ACS estimates and MOEs? It would seem that the standard calculations suggested in the Compass booklets would need to be adjusted. Any ideas?
Thanks in advance,
Jan 18, 2016 10:34 AM
Unfortunately, I don't have any ideas, but your post brings up some questions that might spark ideas. My first thought is this: Would the population controls for 2009 and 2010 data be different in the 2009-2013 PUMS file? Or did they adjust them to make them the same? With the multitude of data items in that file, "Year" is notably missing (although it can be derived from serial number). I wonder if that could be missing in order to discourage time series analysis because of different population controls?
I understood (and I know very little about this) that the population estimates are created by forcing the samples to match known quantities, and are likely created based on criteria that is not available to us (which would seem to make it beyond our capacity to reliably tweak the data). If there's a way to legitimately alter the calculations based on what is available to us, I too would be interested in knowing what it is.
- John Grumbine
Joe D. Francis
Jan 19, 2016 10:39 AM
Appreciate your suggestion. My guess is that the population controls are different as the Census Bureau bothered to put up a special webpage with warnings about comparing the two non-overlapping ACS 5 year releases. Specifically they warned that the 2005-9 ACS had a different base population than the 2010-14 resulting in different weighting being applied to the final population controls. I am wanting to compare ACS estimates and MOEs between the two 5 year time periods for counties in NY where some several of which have populations below the threshold to receive the 1 year ACS releases. As an aside, what a bummer to wait until a second, non-overlapping five year release to make comparisons and be hit with this problem.
Jan 20, 2016 1:07 PM
I think I found the page that you're referring to:
Based on that, the census bureau used different population controls for weighting different 5 year files. So it would be invalid (for this reason and others) to pull, e.g., the 2010 data out of two different files and try to compare them. Thanks for bringing this up,
Joe D. Francis
Jan 20, 2016 4:35 PM
That is the page indeed, John. I have asked a person on the ACS staff at the Census Bureau for help on this--suggestions, workarounds, etc. So far no response. To me this is a cruel turn of events for geographies that don't meet the threshold population for receiving ACS 1 year releases. These units have been waiting since 2009 to make statistically appropriate comparisons with non-overlapping pooled samples. First they take away the 3 year releases. Now this.